fix(cli): usage not red + skill subcommand + --help flag on groups

1. No-args usage uses printCliLine (not printCliError), exit 1
2. 'skill [topic]' as first-class command (help --skill kept as compat)
3. 'workflow --help', 'thread --help' etc. show group subcommands
4. Role prompts updated: 'uncaged-workflow skill develop'

240 tests (6 new), build clean.

Closes #83

小橘 🍊

docs/plans/2025-05-07-workflow-as-agent.md

@@ -0,0 +1,315 @@
+# Workflow-as-Agent Implementation Plan
+
+> **For Hermes:** Use subagent-driven-development skill to implement this plan task-by-task.
+
+**Goal:** Enable workflows to invoke other workflows as agents, backed by global CAS and refs tracking.
+
+**Architecture:** Migrate CAS from thread-local to global (`~/.uncaged/workflow/cas/`), add `refs` to RoleStep for GC traceability, then build `workflowAsAgent(name)` factory that resolves workflow name → bundle via registry and spawns a child thread.
+
+**Tech Stack:** TypeScript, Bun, Zod v4, monorepo with `packages/`
+
+**Issue:** https://git.shazhou.work/uncaged/workflow/issues/25
+
+---
+
+## Phase 1: Global CAS Migration
+
+Move CAS storage from `<threadDir>/<threadId>.cas/` to `~/.uncaged/workflow/cas/` (global, content-addressed, immutable). This is a **breaking change** — thread-local `.cas/` directories are abandoned.
+
+### Task 1.1: Add `globalCasDir` helper to `storage-root.ts`
+
+**Objective:** Provide a single function that returns the global CAS directory path.
+
+**Files:**
+- Modify: `packages/workflow/src/storage-root.ts`
+- Test: `packages/workflow/__tests__/storage-root.test.ts`
+
+**Implementation:**
+
+```typescript
+// storage-root.ts — add export
+export function getGlobalCasDir(storageRoot?: string): string {
+  const root = storageRoot ?? getDefaultWorkflowStorageRoot();
+  return join(root, "cas");
+}
+```
+
+Export from `packages/workflow/src/index.ts`.
+
+### Task 1.2: Update `cmd-cas.ts` to use global CAS
+
+**Objective:** CLI `cas get/put/list/rm` no longer needs threadId for storage location — CAS is global. But keep threadId in CLI for backward compat of planner/coder prompts (they pass threadId).
+
+**Files:**
+- Modify: `packages/cli-workflow/src/cmd-cas.ts`
+
+**Changes:**
+- `resolveCasDir` → use `getGlobalCasDir(storageRoot)` instead of deriving from thread data path
+- `cmdCasPut` / `cmdCasGet` / `cmdCasList` / `cmdCasRm`: threadId is still accepted (prompts pass it) but storage goes to global dir
+- Remove the `resolveThreadDataPath` dependency for CAS operations — thread doesn't need to exist to read CAS
+
+```typescript
+import { createThreadCas, getGlobalCasDir } from "@uncaged/workflow";
+
+export async function cmdCasGet(
+  storageRoot: string,
+  _threadId: string, // kept for CLI compat, not used for path
+  hash: string,
+): Promise<Result<string, string>> {
+  const cas = createThreadCas(getGlobalCasDir(storageRoot));
+  const content = await cas.get(hash);
+  if (content === null) {
+    return err(`cas entry not found: ${hash}`);
+  }
+  return ok(content);
+}
+// ... same pattern for put/list/rm
+```
+
+### Task 1.3: Update `cmd-thread.ts` — thread rm no longer deletes `.cas/`
+
+**Objective:** Since CAS is global, `thread rm` should NOT delete CAS entries. CAS cleanup is GC's job.
+
+**Files:**
+- Modify: `packages/cli-workflow/src/cmd-thread.ts`
+- Check: remove any `rmdir` / `unlink` of `<threadId>.cas/` directory
+
+### Task 1.4: Rename `createThreadCas` → `createCasStore`
+
+**Objective:** The name `createThreadCas` is misleading now. Rename to `createCasStore`.
+
+**Files:**
+- Modify: `packages/workflow/src/cas.ts` — rename function
+- Modify: `packages/workflow/src/index.ts` — update export (keep `createThreadCas` as deprecated alias for one release)
+- Modify: all consumers (`cmd-cas.ts`)
+
+### Task 1.5: Update tests
+
+**Objective:** All CAS-related tests use global dir instead of thread-local.
+
+**Files:**
+- Modify: `packages/cli-workflow/__tests__/commands.test.ts`
+- Verify: `bun test` passes
+
+### Task 1.6: Clean up old thread-local `.cas/` references
+
+**Objective:** Remove dead code that creates/reads thread-local `.cas/` directories.
+
+**Files:**
+- Search all `*.ts` for `.cas` path construction patterns
+- Remove orphaned helpers
+
+---
+
+## Phase 2: RoleStep `refs` Tracking
+
+Add `refs: string[]` to persisted role steps so GC can trace which CAS entries are alive.
+
+### Task 2.1: Add `refs` to `RoleOutput` and engine persistence
+
+**Objective:** Every role step can declare which CAS hashes it produced or consumed.
+
+**Files:**
+- Modify: `packages/workflow/src/types.ts`
+- Modify: `packages/workflow/src/engine.ts`
+
+**Changes to `types.ts`:**
+
+```typescript
+export type RoleOutput = {
+  role: string;
+  content: string;
+  meta: Record<string, unknown>;
+  refs: string[];  // CAS hashes produced/consumed by this step
+};
+```
+
+**Changes to `engine.ts`:**
+- `appendDataLine` for role steps: include `refs` field (default `[]` if not provided)
+
+### Task 2.2: Auto-populate refs from meta hashes
+
+**Objective:** The engine should automatically extract CAS hashes from `meta` to populate `refs`, so roles don't need to manually track them.
+
+**Strategy:** After meta extraction, walk the meta object and collect any string that looks like a CAS hash (Crockford Base32, 13 chars). This is a heuristic but works because CAS hashes are distinctive.
+
+Alternative (simpler): Let each `RoleDefinition` optionally declare a `extractRefs(meta: M) => string[]` function. For planner, this returns `meta.phases.map(p => p.hash)`. For coder, `[meta.completedPhase]`.
+
+**Recommended:** The explicit `extractRefs` approach — no magic, no false positives.
+
+**Files:**
+- Modify: `packages/workflow/src/types.ts` — add optional `extractRefs` to `RoleDefinition`
+- Modify: `packages/workflow/src/create-workflow.ts` — call `extractRefs` after meta extraction, set on `RoleOutput.refs`
+- Modify: `packages/workflow-role-planner/src/planner.ts` — implement `extractRefs`
+- Modify: `packages/workflow-role-coder/src/coder.ts` — implement `extractRefs`
+
+```typescript
+// types.ts — RoleDefinition addition
+export type RoleDefinition<Meta extends Record<string, unknown>> = {
+  description: string;
+  systemPrompt: string;
+  extractPrompt: string;
+  schema: z.ZodType<Meta>;
+  extractRefs?: (meta: Meta) => string[];  // CAS hashes to track
+};
+
+// planner.ts
+extractRefs: (meta) => meta.phases.map(p => p.hash),
+
+// coder.ts
+extractRefs: (meta) => [meta.completedPhase],
+```
+
+### Task 2.3: Update fork logic to preserve refs
+
+**Objective:** When forking a thread, `refs` from historical steps must be carried over.
+
+**Files:**
+- Modify: `packages/workflow/src/fork-thread.ts`
+- Verify: `ForkHistoricalStep` / `PrefilledDiskStep` include `refs`
+
+### Task 2.4: Tests for refs tracking
+
+**Files:**
+- Add: `packages/workflow/__tests__/refs-tracking.test.ts`
+- Verify: refs appear in `.data.jsonl` output
+
+---
+
+## Phase 3: CAS Garbage Collection
+
+### Task 3.1: Implement `gc.ts` in `@uncaged/workflow`
+
+**Objective:** Mark-and-sweep GC — scan all thread `.data.jsonl` files, collect `refs`, delete orphaned CAS entries.
+
+**Files:**
+- Create: `packages/workflow/src/gc.ts`
+- Export from: `packages/workflow/src/index.ts`
+
+```typescript
+export type GcResult = {
+  scannedThreads: number;
+  activeRefs: number;
+  deletedEntries: number;
+  deletedHashes: string[];
+};
+
+export async function garbageCollectCas(storageRoot: string): Promise<GcResult> {
+  // 1. Find all .data.jsonl files under storageRoot
+  // 2. Parse each, flatMap step.refs → Set<string>
+  // 3. List all CAS entries via createCasStore(globalCasDir).list()
+  // 4. Delete entries not in active set
+  // 5. Return stats
+}
+```
+
+### Task 3.2: Add `uncaged-workflow gc` CLI command
+
+**Files:**
+- Create: `packages/cli-workflow/src/cmd-gc.ts`
+- Modify: `packages/cli-workflow/src/cli-dispatch.ts` — add `gc` subcommand
+
+### Task 3.3: Run GC on `thread rm`
+
+**Files:**
+- Modify: `packages/cli-workflow/src/cmd-thread.ts` — after deleting thread data, optionally run GC
+
+### Task 3.4: Tests for GC
+
+**Files:**
+- Create: `packages/cli-workflow/__tests__/gc-cli.test.ts`
+
+---
+
+## Phase 4: `workflowAsAgent` Factory
+
+### Task 4.1: Create `workflowAsAgent` in `@uncaged/workflow`
+
+**Objective:** Factory function that takes a workflow name, resolves to bundle, returns an `AgentFn`.
+
+**Files:**
+- Create: `packages/workflow/src/workflow-as-agent.ts`
+- Export from: `packages/workflow/src/index.ts`
+
+```typescript
+import type { AgentFn } from "./types.js";
+
+export type WorkflowAsAgentOptions = {
+  storageRoot?: string;
+};
+
+export function workflowAsAgent(
+  workflowName: string,
+  options?: WorkflowAsAgentOptions,
+): AgentFn {
+  return async (ctx) => {
+    const storageRoot = options?.storageRoot ?? getDefaultWorkflowStorageRoot();
+
+    // 1. Read registry → resolve name to bundle hash + path
+    const registry = await readWorkflowRegistry(storageRoot);
+    const entry = getRegisteredWorkflow(registry, workflowName);
+    if (entry === null) {
+      return `ERROR: workflow "${workflowName}" not found in registry`;
+    }
+
+    // 2. Load bundle
+    const bundlePath = join(storageRoot, "bundles", `${entry.hash}.esm.js`);
+    const bundleExports = await extractBundleExports(bundlePath);
+
+    // 3. Create child thread input from ctx.start.content (parent prompt)
+    const input: ThreadInput = {
+      prompt: ctx.start.content,
+      steps: [],
+    };
+
+    // 4. Generate child threadId
+    const childThreadId = generateUlid();
+
+    // 5. Execute — collect all yields, return final content
+    const io: ExecuteThreadIo = { ... };
+    const result = await executeThread(bundleExports.run, workflowName, input, ...);
+
+    // 6. Return summary as agent content
+    return result.summary;
+  };
+}
+```
+
+### Task 4.2: System-level depth limit
+
+**Objective:** Prevent infinite recursion. Track depth via thread metadata, enforce a global max (default 3, configurable in `workflow.yaml`).
+
+**Files:**
+- Modify: `packages/workflow/src/types.ts` — add `depth` to `WorkflowFnOptions`
+- Modify: `packages/workflow/src/workflow-as-agent.ts` — increment depth, check limit
+- Modify: registry or config types for `maxDepth` setting
+
+### Task 4.3: Tests for workflowAsAgent
+
+**Files:**
+- Create: `packages/workflow/__tests__/workflow-as-agent.test.ts`
+- Test: name resolution, depth limit, child thread execution
+
+### Task 4.4: Integration test — nested workflow
+
+**Objective:** Create a minimal test workflow that calls another workflow via `workflowAsAgent`.
+
+**Files:**
+- Create: `packages/workflow/__tests__/workflow-as-agent-integration.test.ts`
+
+---
+
+## Execution Order
+
+```
+Phase 1 (Global CAS) → Phase 2 (refs) → Phase 3 (GC) → Phase 4 (workflowAsAgent)
+```
+
+Each phase is independently mergeable. Phase 3 depends on Phase 2 (needs refs to know what's alive). Phase 4 depends on Phase 1 (global CAS for cross-thread sharing).
+
+## Breaking Changes
+
+- CAS storage location moves from `<thread>.cas/` to `~/.uncaged/workflow/cas/`
+- `RoleOutput` gains required `refs: string[]` field
+- Existing threads with thread-local CAS will lose access to old CAS data (acceptable — those are short-lived workflow artifacts)
+- `createThreadCas` renamed to `createCasStore` (alias kept temporarily)

examples/hello-world.ts

@@ -28,6 +28,8 @@ const greeter: RoleDefinition<Roles["greeter"]> = {
   systemPrompt: "You greet the user briefly.",
   extractPrompt: "Extract the greeting string produced for the user.",
   schema: greeterMetaSchema,
+  extractRefs: null,
+  extractMode: "single",
 };
 
 const extract = createExtract({
@@ -47,4 +49,5 @@ export const run = createWorkflow<Roles>(
     agent: async (ctx) => `Hello, ${ctx.start.content}`,
   },
   extract,
+  null,
 );

packages/cli-workflow/__tests__/commands.test.ts

@@ -3,8 +3,9 @@ import { mkdir, mkdtemp, readFile, rm, unlink, writeFile } from "node:fs/promise
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 
-import { getRegisteredWorkflow, readWorkflowRegistry } from "@uncaged/workflow";
+import { getGlobalCasDir, getRegisteredWorkflow, readWorkflowRegistry } from "@uncaged/workflow";
 import { cmdAdd } from "../src/cmd-add.js";
+import { cmdCasGet, cmdCasList, cmdCasPut, cmdCasRm } from "../src/cmd-cas.js";
 import { cmdHistory } from "../src/cmd-history.js";
 import { cmdList, formatListLines } from "../src/cmd-list.js";
 import { cmdRemove } from "../src/cmd-remove.js";
@@ -15,6 +16,9 @@ import { addCliArgs } from "./bundle-fixture.js";
 const fixtureDescriptor = `export const descriptor = { description: "fixture", roles: {} };
 `;
 
+const wfPutImport = `import { putContentMerkleNode } from "@uncaged/workflow";
+`;
+
 describe("cli workflow commands", () => {
   let prevEnv: string | undefined;
   let storageRoot: string;
@@ -40,11 +44,13 @@ describe("cli workflow commands", () => {
     const bundlePath = join(bundleDir, "demo.esm.js");
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}import fs from "node:fs";
+      `${fixtureDescriptor}${wfPutImport}import fs from "node:fs";
 
-export const run = async function* (input) {
+export const run = async function* (input, options) {
   fs.existsSync(".");
-  yield { role: "noop", content: input.prompt, meta: { done: true } };
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, input.prompt);
+  yield { role: "noop", contentHash: h, meta: { done: true }, refs: [h] };
   return { returnCode: 0, summary: "done" };
 }
 `,
@@ -111,8 +117,8 @@ export const run = async function* (input) { return { returnCode: 0, summary: in
     const bundlePath = join(storageRoot, "solo.esm.js");
     await writeFile(
       bundlePath,
-      `export const run = async function* (input) {
-  yield { role: "x", content: input.prompt, meta: {} };
+      `export const run = async function* () {
+  yield { role: "x", contentHash: "STUBHASH00000000000000001", meta: {}, refs: [] };
   return { returnCode: 0, summary: "ok" };
 }
 `,
@@ -140,8 +146,11 @@ export const run = async function* (input) { return { returnCode: 0, summary: in
     },
   },
 };
-export const run = async function* (input) {
-  yield { role: "greeter", content: input.prompt, meta: { greeting: "hi" } };
+${wfPutImport}
+export const run = async function* (input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, input.prompt);
+  yield { role: "greeter", contentHash: h, meta: { greeting: "hi" }, refs: [h] };
   return { returnCode: 0, summary: "ok" };
 };
 `,
@@ -179,8 +188,10 @@ export const run = async function* (input) {
     const bundlePath = join(bundleDir, "demo.esm.js");
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "x", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "x" };
 }
 `,
@@ -208,8 +219,10 @@ export const run = async function* (input) {
     const dtsPath = join(bundleDir, "types.d.ts");
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "x", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "x" };
 }
 `,
@@ -239,8 +252,10 @@ export const run = async function* (input) {
     const bundlePath = join(bundleDir, "demo.esm.js");
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "x", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "x" };
 }
 `,
@@ -260,13 +275,17 @@ export const run = async function* (input) {
     const bundleDir = join(storageRoot, "src");
     await mkdir(bundleDir, { recursive: true });
     const bundlePath = join(bundleDir, "demo.esm.js");
-    const v1 = `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "v1", meta: {} };
+    const v1 = `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "v1");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "v1" };
 }
 `;
-    const v2 = `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "v2", meta: {} };
+    const v2 = `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "v2");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "v2" };
 }
 `;
@@ -298,13 +317,17 @@ export const run = async function* (input) {
     const bundleDir = join(storageRoot, "src");
     await mkdir(bundleDir, { recursive: true });
     const bundlePath = join(bundleDir, "demo.esm.js");
-    const v1 = `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "v1", meta: {} };
+    const v1 = `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "v1");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "v1" };
 }
 `;
-    const v2 = `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "v2", meta: {} };
+    const v2 = `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "v2");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "v2" };
 }
 `;
@@ -346,8 +369,10 @@ export const run = async function* (input) {
     const bundlePath = join(bundleDir, "demo.esm.js");
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "x", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "x" };
 }
 `,
@@ -357,8 +382,10 @@ export const run = async function* (input) {
     expect(add1.ok).toBe(true);
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "y", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "y");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "y" };
 }
 `,
@@ -371,14 +398,47 @@ export const run = async function* (input) {
     expect(bad.ok).toBe(false);
   });
 
+  test("cas put/get/list/rm use global cas dir (thread id not required for storage)", async () => {
+    const put = await cmdCasPut(storageRoot, "nonexistent-thread-id", "phase doc");
+    expect(put.ok).toBe(true);
+    if (!put.ok) {
+      return;
+    }
+    const hash = put.value;
+    const blobPath = join(getGlobalCasDir(storageRoot), `${hash}.txt`);
+    expect(await readFile(blobPath, "utf8")).toBe("phase doc");
+
+    const got = await cmdCasGet(storageRoot, "other-thread", hash);
+    expect(got.ok).toBe(true);
+    if (!got.ok) {
+      return;
+    }
+    expect(got.value).toBe("phase doc");
+
+    const listed = await cmdCasList(storageRoot, "another-thread");
+    expect(listed.ok).toBe(true);
+    if (!listed.ok) {
+      return;
+    }
+    expect(listed.value).toContain(hash);
+
+    const removed = await cmdCasRm(storageRoot, "rm-thread", hash);
+    expect(removed.ok).toBe(true);
+
+    const missing = await cmdCasGet(storageRoot, "after-rm", hash);
+    expect(missing.ok).toBe(false);
+  });
+
   test("rollback rejects missing bundle file for target hash", async () => {
     const bundleDir = join(storageRoot, "src");
     await mkdir(bundleDir, { recursive: true });
     const bundlePath = join(bundleDir, "demo.esm.js");
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "x", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "x" };
 }
 `,
@@ -392,8 +452,10 @@ export const run = async function* (input) {
     const hash1 = add1.value.hash;
     await writeFile(
       bundlePath,
-      `${fixtureDescriptor}export const run = async function* (input) {
-  yield { role: "a", content: "y", meta: {} };
+      `${fixtureDescriptor}${wfPutImport}export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "y");
+  yield { role: "a", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "y" };
 }
 `,

packages/cli-workflow/__tests__/fixtures/live/logs/C9NMV6V2TQT81/01LIVECMPLT01DDDDDDDDDDDDG.data.jsonl

@@ -0,0 +1,4 @@
+{"name":"demo-live","hash":"C9NMV6V2TQT81","threadId":"01LIVECMPLT01DDDDDDDDDDDDG","parameters":{"prompt":"hello","options":{"maxRounds":5,"depth":0}},"timestamp":1714963400000}
+{"role":"planner","contentHash":"FF7YQ5W3S2EV6","meta":{"phase":"plan","flags":[1,2]},"refs":[],"timestamp":1714963201000}
+{"role":"coder","contentHash":"EN34XX1W4WAFJ","meta":{},"refs":[],"timestamp":1714963202000}
+{"returnCode":0,"summary":"fixture completed"}

packages/cli-workflow/__tests__/fixtures/live/logs/C9NMV6V2TQT81/01LIVECMPLT01DDDDDDDDDDDDG.info.jsonl

@@ -0,0 +1,2 @@
+{"tag":"DEBUGTAG1","content":"bundle loaded","timestamp":1714963400050}
+{"tag":"DEBUGTAG2","content":"multi\nline","timestamp":1714963400500}

packages/cli-workflow/__tests__/fixtures/live/logs/C9NMV6V2TQT81/01LIVEINFLY01DDDDDDDDDDDDG.data.jsonl

@@ -0,0 +1,2 @@
+{"name":"demo-live","hash":"C9NMV6V2TQT81","threadId":"01LIVEINFLY01DDDDDDDDDDDDG","parameters":{"prompt":"hello","options":{"maxRounds":5,"depth":0}},"timestamp":1714963200000}
+{"role":"planner","contentHash":"P6M9FHE1GSBN0","meta":{"x":1},"refs":[],"timestamp":1714963201000}

packages/cli-workflow/__tests__/fixtures/live/logs/C9NMV6V2TQT81/01LIVEOLDER01DDDDDDDDDDDDG.data.jsonl

@@ -0,0 +1,2 @@
+{"name":"demo-live-old","hash":"C9NMV6V2TQT81","threadId":"01LIVEOLDER01DDDDDDDDDDDDG","parameters":{"prompt":"old","options":{"maxRounds":5,"depth":0}},"timestamp":1714963000000}
+{"returnCode":0,"summary":"older thread"}

packages/cli-workflow/__tests__/fork-cli.test.ts

@@ -2,6 +2,7 @@ import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
+import { createCasStore, getContentMerklePayload, getGlobalCasDir } from "@uncaged/workflow";
 import { cmdAdd } from "../src/cmd-add.js";
 import { cmdFork } from "../src/cmd-fork.js";
 import { cmdRun } from "../src/cmd-run.js";
@@ -9,7 +10,9 @@ import { pathExists } from "../src/fs-utils.js";
 import { addCliArgs } from "./bundle-fixture.js";
 
 /** Three-role workflow that respects `input.steps` for fork/resume. */
-const threeRoleBundleSource = `export const descriptor = {
+const threeRoleBundleSource = `import { putContentMerkleNode } from "@uncaged/workflow";
+
+export const descriptor = {
   description: "fork-cli",
   roles: {
     planner: { description: "planner", schema: {} },
@@ -17,20 +20,21 @@ const threeRoleBundleSource = `export const descriptor = {
     reviewer: { description: "reviewer", schema: {} },
   },
 };
-export const run = async function* (input) {
+export const run = async function* (input, options) {
+  const cas = options.cas;
   const has = (r) => input.steps.some((s) => s.role === r);
   if (!has("planner")) {
-    yield { role: "planner", content: "p1", meta: { k: "planner" } };
+    const h = await putContentMerkleNode(cas, "p1");
+    yield { role: "planner", contentHash: h, meta: { k: "planner" }, refs: [h] };
   }
   if (!has("coder")) {
-    yield { role: "coder", content: "c1", meta: { k: "coder" } };
+    const h = await putContentMerkleNode(cas, "c1");
+    yield { role: "coder", contentHash: h, meta: { k: "coder" }, refs: [h] };
   }
   if (!has("reviewer")) {
-    yield {
-      role: "reviewer",
-      content: "rev-" + String(input.steps.length),
-      meta: { k: "reviewer" },
-    };
+    const body = "rev-" + String(input.steps.length);
+    const h = await putContentMerkleNode(cas, body);
+    yield { role: "reviewer", contentHash: h, meta: { k: "reviewer" }, refs: [h] };
   }
   return { returnCode: 0, summary: "done" };
 };
@@ -107,7 +111,7 @@ describe("cli fork", () => {
     const sourceData = join(storageRoot, "logs", hash, `${sourceId}.data.jsonl`);
     const sourceRunning = join(storageRoot, "logs", hash, `${sourceId}.running`);
     await waitUntilRunningAbsent(sourceRunning);
-    await waitUntilMinDataLines(sourceData, 4);
+    await waitUntilMinDataLines(sourceData, 5);
 
     const forked = await cmdFork(storageRoot, sourceId, "planner");
     expect(forked.ok).toBe(true);
@@ -118,21 +122,22 @@ describe("cli fork", () => {
     const newData = join(storageRoot, "logs", hash, `${newId}.data.jsonl`);
     const newRunning = join(storageRoot, "logs", hash, `${newId}.running`);
     await waitUntilRunningAbsent(newRunning);
-    await waitUntilMinDataLines(newData, 4);
+    await waitUntilMinDataLines(newData, 5);
 
     const text = await readFile(newData, "utf8");
     const lines = text
       .trim()
       .split("\n")
       .filter((l) => l !== "");
-    expect(lines.length).toBe(4);
+    expect(lines.length).toBe(5);
     const start = JSON.parse(lines[0] ?? "{}") as Record<string, unknown>;
     expect(start.threadId).toBe(newId);
     expect(start.forkFrom).toEqual({ threadId: sourceId });
 
-    const last = JSON.parse(lines[lines.length - 1] ?? "{}") as Record<string, unknown>;
-    expect(last.role).toBe("reviewer");
-    expect(last.content).toBe("rev-1");
+    const lastRoleLine = JSON.parse(lines[lines.length - 2] ?? "{}") as Record<string, unknown>;
+    expect(lastRoleLine.role).toBe("reviewer");
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    expect(await getContentMerklePayload(cas, String(lastRoleLine.contentHash))).toBe("rev-1");
   });
 
   test("fork without --from-role retries last role", async () => {
@@ -157,7 +162,7 @@ describe("cli fork", () => {
     const sourceData = join(storageRoot, "logs", hash, `${sourceId}.data.jsonl`);
     const sourceRunning = join(storageRoot, "logs", hash, `${sourceId}.running`);
     await waitUntilRunningAbsent(sourceRunning);
-    await waitUntilMinDataLines(sourceData, 4);
+    await waitUntilMinDataLines(sourceData, 5);
 
     const forked = await cmdFork(storageRoot, sourceId, null);
     expect(forked.ok).toBe(true);
@@ -168,22 +173,23 @@ describe("cli fork", () => {
     const newData = join(storageRoot, "logs", hash, `${newId}.data.jsonl`);
     const newRunning = join(storageRoot, "logs", hash, `${newId}.running`);
     await waitUntilRunningAbsent(newRunning);
-    await waitUntilMinDataLines(newData, 4);
+    await waitUntilMinDataLines(newData, 5);
 
     const text = await readFile(newData, "utf8");
     const lines = text
       .trim()
       .split("\n")
       .filter((l) => l !== "");
-    expect(lines.length).toBe(4);
+    expect(lines.length).toBe(5);
 
     const replayCoder = JSON.parse(lines[2] ?? "{}") as Record<string, unknown>;
     expect(replayCoder.role).toBe("coder");
-    expect(replayCoder.content).toBe("c1");
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    expect(await getContentMerklePayload(cas, String(replayCoder.contentHash))).toBe("c1");
 
-    const last = JSON.parse(lines[lines.length - 1] ?? "{}") as Record<string, unknown>;
-    expect(last.role).toBe("reviewer");
-    expect(last.content).toBe("rev-2");
+    const lastRoleLine = JSON.parse(lines[lines.length - 2] ?? "{}") as Record<string, unknown>;
+    expect(lastRoleLine.role).toBe("reviewer");
+    expect(await getContentMerklePayload(cas, String(lastRoleLine.contentHash))).toBe("rev-2");
   });
 
   test("fork rejects unknown role with available names", async () => {
@@ -207,7 +213,7 @@ describe("cli fork", () => {
     const sourceData = join(storageRoot, "logs", added.value.hash, `${sourceId}.data.jsonl`);
     const sourceRunning = join(storageRoot, "logs", added.value.hash, `${sourceId}.running`);
     await waitUntilRunningAbsent(sourceRunning);
-    await waitUntilMinDataLines(sourceData, 4);
+    await waitUntilMinDataLines(sourceData, 5);
 
     const bad = await cmdFork(storageRoot, sourceId, "ghost-role");
     expect(bad.ok).toBe(false);

packages/cli-workflow/__tests__/gc-cli.test.ts

@@ -0,0 +1,165 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { spawnSync } from "node:child_process";
+import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { fileURLToPath } from "node:url";
+import {
+  createCasStore,
+  garbageCollectCas,
+  getGlobalCasDir,
+  putContentMerkleNode,
+} from "@uncaged/workflow";
+import { cmdThreadRemove } from "../src/cmd-thread.js";
+import { pathExists } from "../src/fs-utils.js";
+
+const cliEntryPath = fileURLToPath(new URL("../src/cli.ts", import.meta.url));
+
+async function writeDemoDataJsonl(params: {
+  path: string;
+  threadId: string;
+  bundleHash: string;
+  cas: ReturnType<typeof createCasStore>;
+  activeHash: string;
+}): Promise<void> {
+  const bodyHash = await putContentMerkleNode(params.cas, "p");
+  const text = [
+    JSON.stringify({
+      name: "demo",
+      hash: params.bundleHash,
+      threadId: params.threadId,
+      parameters: { prompt: "hi", options: { maxRounds: 5 } },
+      timestamp: 100,
+    }),
+    JSON.stringify({
+      role: "planner",
+      contentHash: bodyHash,
+      meta: {},
+      refs: [params.activeHash, bodyHash],
+      timestamp: 101,
+    }),
+    "",
+  ].join("\n");
+  await writeFile(params.path, text, "utf8");
+}
+
+describe("gc cli and garbageCollectCas", () => {
+  let prevEnv: string | undefined;
+  let storageRoot: string;
+
+  beforeEach(async () => {
+    prevEnv = process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    storageRoot = await mkdtemp(join(tmpdir(), "uncaged-wf-gc-"));
+    process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = storageRoot;
+  });
+
+  afterEach(async () => {
+    if (prevEnv === undefined) {
+      delete process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    } else {
+      process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = prevEnv;
+    }
+    await rm(storageRoot, { recursive: true, force: true });
+  });
+
+  test("garbageCollectCas keeps CAS entries referenced by thread refs", async () => {
+    const bundleHash = "C9NMV6V2TQT81";
+    const threadId = "01AAA1111111111111111111";
+    const logsDir = join(storageRoot, "logs", bundleHash);
+    await mkdir(logsDir, { recursive: true });
+
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    const activeHash = await cas.put("active-blob");
+    const orphanHash = await cas.put("orphan-blob");
+
+    await writeDemoDataJsonl({
+      path: join(logsDir, `${threadId}.data.jsonl`),
+      threadId,
+      bundleHash,
+      cas,
+      activeHash,
+    });
+
+    const gc = await garbageCollectCas(storageRoot);
+    expect(gc.ok).toBe(true);
+    if (!gc.ok) {
+      return;
+    }
+    expect(gc.value.scannedThreads).toBe(1);
+    expect(gc.value.activeRefs).toBe(2);
+    expect(gc.value.deletedEntries).toBe(1);
+    expect(gc.value.deletedHashes).toEqual([orphanHash]);
+
+    expect(await pathExists(join(getGlobalCasDir(storageRoot), `${activeHash}.txt`))).toBe(true);
+    expect(await pathExists(join(getGlobalCasDir(storageRoot), `${orphanHash}.txt`))).toBe(false);
+  });
+
+  test("garbageCollectCas deletes orphaned CAS when no threads reference them", async () => {
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    const orphanHash = await cas.put("lonely");
+
+    const gc = await garbageCollectCas(storageRoot);
+    expect(gc.ok).toBe(true);
+    if (!gc.ok) {
+      return;
+    }
+    expect(gc.value.scannedThreads).toBe(0);
+    expect(gc.value.activeRefs).toBe(0);
+    expect(gc.value.deletedEntries).toBe(1);
+    expect(gc.value.deletedHashes).toEqual([orphanHash]);
+    expect(await pathExists(join(getGlobalCasDir(storageRoot), `${orphanHash}.txt`))).toBe(false);
+  });
+
+  test("cli gc prints stats", async () => {
+    const bundleHash = "C9NMV6V2TQT81";
+    const threadId = "01BBB2222222222222222222";
+    const logsDir = join(storageRoot, "logs", bundleHash);
+    await mkdir(logsDir, { recursive: true });
+
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    const activeHash = await cas.put("keep-me");
+    await cas.put("drop-me");
+
+    await writeDemoDataJsonl({
+      path: join(logsDir, `${threadId}.data.jsonl`),
+      threadId,
+      bundleHash,
+      cas,
+      activeHash,
+    });
+
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const proc = spawnSync(process.execPath, [cliEntryPath, "cas", "gc"], {
+      env,
+      encoding: "utf8",
+    });
+    expect(proc.status).toBe(0);
+    expect(String(proc.stdout).trim()).toBe("scanned 1 threads, 2 active refs, deleted 1 entries");
+  });
+
+  test("thread rm triggers gc so unreferenced CAS is removed", async () => {
+    const bundleHash = "C9NMV6V2TQT81";
+    const threadId = "01CCC3333333333333333333";
+    const logsDir = join(storageRoot, "logs", bundleHash);
+    await mkdir(logsDir, { recursive: true });
+
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    const activeHash = await cas.put("pinned-by-ref");
+    await writeDemoDataJsonl({
+      path: join(logsDir, `${threadId}.data.jsonl`),
+      threadId,
+      bundleHash,
+      cas,
+      activeHash,
+    });
+
+    const orphanHash = await cas.put("orphan-after-rm");
+    const orphanPath = join(getGlobalCasDir(storageRoot), `${orphanHash}.txt`);
+
+    const removed = await cmdThreadRemove(storageRoot, threadId);
+    expect(removed.ok).toBe(true);
+
+    expect(await pathExists(orphanPath)).toBe(false);
+    expect(await pathExists(join(getGlobalCasDir(storageRoot), `${activeHash}.txt`))).toBe(false);
+  });
+});

packages/cli-workflow/__tests__/help.test.ts

@@ -0,0 +1,214 @@
+import { describe, expect, test } from "bun:test";
+import { runCli } from "../src/cli-dispatch.js";
+import {
+  formatSkillDoc,
+  formatSkillIndex,
+  formatSkillTopic,
+  getSkillTopics,
+} from "../src/cmd-help.js";
+
+const STORAGE_ROOT = "/tmp/help-test-storage";
+
+describe("help command", () => {
+  test("help returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["help"]);
+    expect(code).toBe(0);
+  });
+
+  test("no args prints usage (not red) and returns 1", async () => {
+    const code = await runCli(STORAGE_ROOT, []);
+    expect(code).toBe(1);
+  });
+});
+
+describe("skill command", () => {
+  test("skill (no topic) lists topics and returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["skill"]);
+    expect(code).toBe(0);
+  });
+
+  test("skill cli returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["skill", "cli"]);
+    expect(code).toBe(0);
+  });
+
+  test("skill develop returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["skill", "develop"]);
+    expect(code).toBe(0);
+  });
+
+  test("skill author returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["skill", "author"]);
+    expect(code).toBe(0);
+  });
+
+  test("skill unknown returns 1", async () => {
+    const code = await runCli(STORAGE_ROOT, ["skill", "unknown"]);
+    expect(code).toBe(1);
+  });
+});
+
+describe("--help flag on groups", () => {
+  test("workflow --help returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["workflow", "--help"]);
+    expect(code).toBe(0);
+  });
+
+  test("thread --help returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["thread", "--help"]);
+    expect(code).toBe(0);
+  });
+
+  test("cas --help returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["cas", "--help"]);
+    expect(code).toBe(0);
+  });
+
+  test("init --help returns 0", async () => {
+    const code = await runCli(STORAGE_ROOT, ["init", "--help"]);
+    expect(code).toBe(0);
+  });
+});
+
+describe("legacy help --skill compat", () => {
+  test("help --skill still works (lists topics)", async () => {
+    const code = await runCli(STORAGE_ROOT, ["help", "--skill"]);
+    expect(code).toBe(0);
+  });
+});
+
+describe("getSkillTopics", () => {
+  test("returns all topics", () => {
+    const topics = getSkillTopics();
+    const names = topics.map((t) => t.name);
+    expect(names).toContain("cli");
+    expect(names).toContain("develop");
+    expect(names).toContain("author");
+  });
+});
+
+describe("formatSkillIndex", () => {
+  test("lists all topics", () => {
+    const idx = formatSkillIndex();
+    expect(idx).toContain("cli");
+    expect(idx).toContain("develop");
+    expect(idx).toContain("author");
+    expect(idx).toContain("skill <topic>");
+  });
+});
+
+describe("formatSkillTopic('cli') — legacy formatSkillDoc", () => {
+  const doc = formatSkillDoc();
+
+  test("contains title", () => {
+    expect(doc).toContain("# uncaged-workflow CLI Reference");
+  });
+
+  test("contains all command group headers", () => {
+    expect(doc).toContain("### workflow");
+    expect(doc).toContain("### thread");
+    expect(doc).toContain("### cas");
+    expect(doc).toContain("### init");
+    expect(doc).toContain("### Top-level shortcuts");
+  });
+
+  test("contains core concepts", () => {
+    expect(doc).toContain("## Core Concepts");
+    expect(doc).toContain("Workflow");
+    expect(doc).toContain("Bundle");
+    expect(doc).toContain("Thread");
+    expect(doc).toContain("CAS");
+    expect(doc).toContain("Registry");
+  });
+
+  test("mentions all workflow subcommands", () => {
+    for (const sub of ["add", "list", "show", "rm", "history", "rollback"]) {
+      expect(doc).toContain(`workflow ${sub}`);
+    }
+  });
+
+  test("mentions all thread subcommands", () => {
+    for (const sub of [
+      "run",
+      "list",
+      "show",
+      "rm",
+      "fork",
+      "ps",
+      "kill",
+      "live",
+      "pause",
+      "resume",
+    ]) {
+      expect(doc).toContain(`thread ${sub}`);
+    }
+  });
+
+  test("mentions all cas subcommands", () => {
+    for (const sub of ["get", "put", "list", "rm", "gc"]) {
+      expect(doc).toContain(`cas ${sub}`);
+    }
+  });
+
+  test("contains exit codes section", () => {
+    expect(doc).toContain("## Exit Codes");
+  });
+
+  test("contains environment variables section", () => {
+    expect(doc).toContain("## Environment Variables");
+    expect(doc).toContain("UNCAGED_WORKFLOW_STORAGE_ROOT");
+  });
+
+  test("contains typical workflow section", () => {
+    expect(doc).toContain("## Typical Workflow");
+  });
+});
+
+describe("formatSkillTopic('develop')", () => {
+  const doc = formatSkillTopic("develop");
+
+  test("returns non-null", () => {
+    expect(doc).not.toBeNull();
+  });
+
+  test("contains thread ID info", () => {
+    expect(doc).toContain("Thread ID");
+    expect(doc).toContain("Crockford Base32");
+  });
+
+  test("contains CAS commands", () => {
+    expect(doc).toContain("cas put");
+    expect(doc).toContain("cas get");
+  });
+
+  test("contains meta output section", () => {
+    expect(doc).toContain("Meta Output");
+  });
+});
+
+describe("formatSkillTopic('author')", () => {
+  const doc = formatSkillTopic("author");
+
+  test("returns non-null", () => {
+    expect(doc).not.toBeNull();
+  });
+
+  test("contains bundle structure", () => {
+    expect(doc).toContain("Bundle Structure");
+    expect(doc).toContain(".esm.js");
+  });
+
+  test("contains descriptor info", () => {
+    expect(doc).toContain("WorkflowDescriptor");
+  });
+
+  test("contains role definition", () => {
+    expect(doc).toContain("Role Definition");
+  });
+});
+
+describe("formatSkillTopic unknown", () => {
+  test("returns null for unknown topic", () => {
+    expect(formatSkillTopic("nonexistent")).toBeNull();
+  });
+});

packages/cli-workflow/__tests__/init-template.test.ts

@@ -0,0 +1,142 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { mkdir, readFile, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import { runCli } from "../src/cli-dispatch.js";
+import { cmdInitTemplate, cmdInitWorkspace } from "../src/cmd-init.js";
+import { pathExists } from "../src/fs-utils.js";
+
+describe("init template", () => {
+  let parent: string;
+
+  beforeEach(async () => {
+    parent = join(
+      tmpdir(),
+      `wf-init-template-${Date.now()}-${Math.random().toString(36).slice(2)}`,
+    );
+    await mkdir(parent, { recursive: true });
+  });
+
+  afterEach(async () => {
+    await rm(parent, { recursive: true, force: true });
+  });
+
+  test("creates templates/<name> with expected files", async () => {
+    const ws = await cmdInitWorkspace(parent, "my-workflows");
+    expect(ws.ok).toBe(true);
+    if (!ws.ok) {
+      return;
+    }
+    const root = ws.value.rootPath;
+
+    const created = await cmdInitTemplate(root, "review-pr");
+    expect(created.ok).toBe(true);
+    if (!created.ok) {
+      return;
+    }
+
+    const tdir = join(root, "templates", "review-pr");
+    expect(created.value.templatePath).toBe(tdir);
+    expect(await pathExists(join(tdir, "package.json"))).toBe(true);
+    expect(await pathExists(join(tdir, "tsconfig.json"))).toBe(true);
+    expect(await pathExists(join(tdir, "src", "roles.ts"))).toBe(true);
+    expect(await pathExists(join(tdir, "src", "moderator.ts"))).toBe(true);
+    expect(await pathExists(join(tdir, "src", "index.ts"))).toBe(true);
+
+    const pkg = JSON.parse(await readFile(join(tdir, "package.json"), "utf8")) as {
+      name: string;
+      type: string;
+      dependencies: Record<string, string>;
+    };
+    expect(pkg.type).toBe("module");
+    expect(pkg.dependencies["@uncaged/workflow"]).toBeDefined();
+    expect(pkg.dependencies.zod).toBeDefined();
+    expect(pkg.name).toContain("review-pr");
+
+    const idx = await readFile(join(tdir, "src", "index.ts"), "utf8");
+    expect(idx).toContain("WorkflowDefinition");
+
+    const roles = await readFile(join(tdir, "src", "roles.ts"), "utf8");
+    expect(roles).not.toContain("interface ");
+    expect(roles).not.toContain("?:");
+    expect(roles).not.toContain("export default");
+
+    const moder = await readFile(join(tdir, "src", "moderator.ts"), "utf8");
+    expect(moder).not.toContain("export default");
+  });
+
+  test("finds workspace walking up from nested cwd", async () => {
+    const ws = await cmdInitWorkspace(parent, "ws");
+    expect(ws.ok).toBe(true);
+    if (!ws.ok) {
+      return;
+    }
+    const root = ws.value.rootPath;
+    const nested = join(root, "a", "b");
+    await mkdir(nested, { recursive: true });
+
+    const created = await cmdInitTemplate(nested, "nested-tpl");
+    expect(created.ok).toBe(true);
+    if (!created.ok) {
+      return;
+    }
+    expect(await pathExists(join(root, "templates", "nested-tpl", "src", "index.ts"))).toBe(true);
+  });
+
+  test("errors when not inside a workflow workspace", async () => {
+    const orphan = join(parent, "nowhere");
+    await mkdir(orphan, { recursive: true });
+    const r = await cmdInitTemplate(orphan, "x");
+    expect(r.ok).toBe(false);
+    if (!r.ok) {
+      expect(r.error).toContain("templates/*");
+    }
+  });
+
+  test("errors when template directory already exists", async () => {
+    const ws = await cmdInitWorkspace(parent, "ws");
+    expect(ws.ok).toBe(true);
+    if (!ws.ok) {
+      return;
+    }
+    const root = ws.value.rootPath;
+
+    const first = await cmdInitTemplate(root, "dup");
+    expect(first.ok).toBe(true);
+
+    const second = await cmdInitTemplate(root, "dup");
+    expect(second.ok).toBe(false);
+    if (!second.ok) {
+      expect(second.error).toContain("already exists");
+    }
+  });
+
+  test("errors on invalid template name", async () => {
+    const ws = await cmdInitWorkspace(parent, "ws");
+    expect(ws.ok).toBe(true);
+    if (!ws.ok) {
+      return;
+    }
+    const bad = await cmdInitTemplate(ws.value.rootPath, "a/b");
+    expect(bad.ok).toBe(false);
+  });
+
+  test.serial("runCli init template uses cwd and succeeds in workspace", async () => {
+    const ws = await cmdInitWorkspace(parent, "cli-ws");
+    expect(ws.ok).toBe(true);
+    if (!ws.ok) {
+      return;
+    }
+    const root = ws.value.rootPath;
+    const prev = process.cwd();
+    try {
+      process.chdir(root);
+      const code = await runCli(join(parent, "_storage"), ["init", "template", "from-cli"]);
+      expect(code).toBe(0);
+      expect(await pathExists(join(root, "templates", "from-cli", "package.json"))).toBe(true);
+    } finally {
+      process.chdir(prev);
+    }
+  });
+});

packages/cli-workflow/__tests__/init-workspace.test.ts

@@ -0,0 +1,152 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { mkdir, readFile, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import { formatCliUsage, runCli } from "../src/cli-dispatch.js";
+import { cmdInitWorkspace } from "../src/cmd-init.js";
+import { pathExists } from "../src/fs-utils.js";
+
+describe("init workspace", () => {
+  let parent: string;
+
+  beforeEach(async () => {
+    parent = join(tmpdir(), `wf-init-${Date.now()}-${Math.random().toString(36).slice(2)}`);
+    await mkdir(parent, { recursive: true });
+  });
+
+  afterEach(async () => {
+    await rm(parent, { recursive: true, force: true });
+  });
+
+  test("creates expected files and directories", async () => {
+    const created = await cmdInitWorkspace(parent, "my-workflows");
+    expect(created.ok).toBe(true);
+    if (!created.ok) {
+      return;
+    }
+
+    const root = created.value.rootPath;
+    expect(await pathExists(join(root, "package.json"))).toBe(true);
+    expect(await pathExists(join(root, "biome.json"))).toBe(true);
+    expect(await pathExists(join(root, "tsconfig.json"))).toBe(true);
+    expect(await pathExists(join(root, "AGENTS.md"))).toBe(true);
+    expect(await pathExists(join(root, "README.md"))).toBe(true);
+    expect(await pathExists(join(root, "templates"))).toBe(true);
+    expect(await pathExists(join(root, "templates", ".gitkeep"))).toBe(true);
+    expect(await pathExists(join(root, "workflows", "package.json"))).toBe(true);
+
+    const rootPkg = JSON.parse(await readFile(join(root, "package.json"), "utf8")) as {
+      workspaces: string[];
+    };
+    expect(rootPkg.workspaces).toEqual(["templates/*", "workflows"]);
+
+    const wfPkg = JSON.parse(await readFile(join(root, "workflows", "package.json"), "utf8")) as {
+      type: string;
+      dependencies: Record<string, string>;
+    };
+    expect(wfPkg.type).toBe("module");
+    expect(wfPkg.dependencies["@uncaged/workflow"]).toBeDefined();
+    expect(wfPkg.dependencies.zod).toBeDefined();
+
+    const tsconfig = JSON.parse(await readFile(join(root, "tsconfig.json"), "utf8")) as {
+      compilerOptions: { strict: boolean; module: string; target: string };
+    };
+    expect(tsconfig.compilerOptions.strict).toBe(true);
+    expect(tsconfig.compilerOptions.module).toBe("ESNext");
+    expect(tsconfig.compilerOptions.target).toBe("ESNext");
+  });
+
+  test("AGENTS.md contains coding agent guide sections and terms", async () => {
+    const created = await cmdInitWorkspace(parent, "my-workflows");
+    expect(created.ok).toBe(true);
+    if (!created.ok) {
+      return;
+    }
+
+    const agentsPath = join(created.value.rootPath, "AGENTS.md");
+    const body = await readFile(agentsPath, "utf8");
+
+    for (const section of [
+      "项目结构",
+      "核心概念",
+      "开发流程",
+      "编码规范",
+      "Template",
+      "Build",
+      "常见陷阱",
+    ]) {
+      expect(body).toContain(section);
+    }
+
+    for (const term of [
+      "RoleDefinition",
+      "WorkflowDefinition",
+      "Moderator",
+      "AgentFn",
+      "ExtractFn",
+      "RoleMeta",
+    ]) {
+      expect(body).toContain(term);
+    }
+
+    expect(body).toMatch(/type[\s\S]*interface/i);
+    expect(body).toMatch(/function[\s\S]*class/i);
+    expect(body).toContain("Crockford Base32");
+    expect(body).toMatch(/no[\s\S]*default export/i);
+    expect(body).toMatch(/no[\s\S]*console/i);
+    expect(body).toMatch(/no[\s\S]*dynamic import/i);
+
+    expect(body).toContain("bun run check");
+    expect(body).toContain("bun test");
+    expect(body).toContain("uncaged-workflow");
+    expect(body).toContain("bun build");
+    expect(body).toContain("CLAUDE.md");
+    expect(body).toContain("docs/architecture.md");
+  });
+
+  test("errors when directory already exists", async () => {
+    const first = await cmdInitWorkspace(parent, "dup");
+    expect(first.ok).toBe(true);
+
+    const second = await cmdInitWorkspace(parent, "dup");
+    expect(second.ok).toBe(false);
+    if (!second.ok) {
+      expect(second.error).toContain("already exists");
+    }
+  });
+
+  test("errors on invalid workspace name", async () => {
+    const slash = await cmdInitWorkspace(parent, "a/b");
+    expect(slash.ok).toBe(false);
+
+    const dots = await cmdInitWorkspace(parent, "..");
+    expect(dots.ok).toBe(false);
+
+    const empty = await cmdInitWorkspace(parent, "");
+    expect(empty.ok).toBe(false);
+  });
+
+  test("usage lists init subcommands", () => {
+    const u = formatCliUsage();
+    expect(u).toContain("uncaged-workflow init workspace <name>");
+    expect(u).toContain("uncaged-workflow init template <name>");
+  });
+
+  test("runCli rejects unknown init subcommand", async () => {
+    const code = await runCli(join(parent, "_storage"), ["init", "bogus", "name"]);
+    expect(code).toBe(1);
+  });
+
+  test.serial("runCli init workspace uses cwd", async () => {
+    const prev = process.cwd();
+    try {
+      process.chdir(parent);
+      const code = await runCli(join(parent, "_storage"), ["init", "workspace", "from-cli"]);
+      expect(code).toBe(0);
+      expect(await pathExists(join(parent, "from-cli", "workflows", "package.json"))).toBe(true);
+    } finally {
+      process.chdir(prev);
+    }
+  });
+});

packages/cli-workflow/__tests__/live.test.ts

@@ -0,0 +1,369 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { spawn, spawnSync } from "node:child_process";
+import { cp, mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { fileURLToPath } from "node:url";
+
+import { createCasStore, getGlobalCasDir, putContentMerkleNode } from "@uncaged/workflow";
+
+import {
+  formatLiveDebugLine,
+  formatLiveTimeLabel,
+  LIVE_CONTENT_MAX_LINES,
+  type LiveRoleRow,
+  renderLiveRoleStepLines,
+} from "../src/cmd-live.js";
+import { parseLiveArgv } from "../src/live-argv.js";
+
+const cliEntryPath = fileURLToPath(new URL("../src/cli.ts", import.meta.url));
+const fixtureRoot = fileURLToPath(new URL("./fixtures/live", import.meta.url));
+
+/** Bodies for Merkle content nodes; hashes must match `.data.jsonl` fixtures. */
+const LIVE_FIXTURE_PLANNER_BODY =
+  "alpha\nbeta\ngamma\nLINE4\nLINE5\nLINE6\nLINE7\nLINE8\nLINE9\nLINE10\nLINE11";
+
+describe("live helpers", () => {
+  test("formatLiveTimeLabel pads HH:MM:SS", () => {
+    const label = formatLiveTimeLabel(new Date("2024-06-01T09:08:07.000Z").getTime());
+    expect(label).toMatch(/^\d{2}:\d{2}:\d{2}$/);
+  });
+
+  test("formatLiveDebugLine flattens newlines in message", () => {
+    const line = formatLiveDebugLine(0, "TAG1", "a\nb");
+    expect(line).toContain("[TAG1]");
+    expect(line).toContain("a b");
+    expect(line).not.toContain("\n");
+  });
+
+  test("renderLiveRoleStepLines truncates content to LIVE_CONTENT_MAX_LINES", () => {
+    const lines = Array.from({ length: LIVE_CONTENT_MAX_LINES + 3 }, (_, i) => `L${i + 1}`);
+    const row: LiveRoleRow = {
+      role: "r",
+      content: lines.join("\n"),
+      meta: { k: "v" },
+      timestamp: 0,
+    };
+    const out = renderLiveRoleStepLines(row, "r");
+    const body = out.filter((l) => l.startsWith("  L"));
+    expect(body.length).toBe(LIVE_CONTENT_MAX_LINES);
+    expect(out.some((l) => l.includes("more line"))).toBe(true);
+    expect(out.some((l) => l.startsWith("  meta: "))).toBe(true);
+  });
+});
+
+describe("parseLiveArgv", () => {
+  test("parses thread id and flags in any order", () => {
+    const a = parseLiveArgv(["01ABC", "--debug", "--role", "planner"]);
+    expect(a.ok).toBe(true);
+    if (a.ok) {
+      expect(a.value.threadId).toBe("01ABC");
+      expect(a.value.latest).toBe(false);
+      expect(a.value.debug).toBe(true);
+      expect(a.value.role).toBe("planner");
+    }
+    const b = parseLiveArgv(["--latest", "--role", "x"]);
+    expect(b.ok).toBe(true);
+    if (b.ok) {
+      expect(b.value.latest).toBe(true);
+      expect(b.value.threadId).toBe(null);
+      expect(b.value.role).toBe("x");
+    }
+  });
+
+  test("rejects --latest with thread id", () => {
+    const r = parseLiveArgv(["--latest", "01ABC"]);
+    expect(r.ok).toBe(false);
+  });
+});
+
+describe("live CLI", () => {
+  let prevEnv: string | undefined;
+  let storageRoot: string;
+
+  beforeEach(async () => {
+    prevEnv = process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    storageRoot = await mkdtemp(join(tmpdir(), "uncaged-wf-live-"));
+    process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = storageRoot;
+    await mkdir(join(storageRoot, "logs", "C9NMV6V2TQT81"), { recursive: true });
+    await cp(
+      join(fixtureRoot, "logs", "C9NMV6V2TQT81", "01LIVECMPLT01DDDDDDDDDDDDG.data.jsonl"),
+      join(storageRoot, "logs", "C9NMV6V2TQT81", "01LIVECMPLT01DDDDDDDDDDDDG.data.jsonl"),
+    );
+    await cp(
+      join(fixtureRoot, "logs", "C9NMV6V2TQT81", "01LIVECMPLT01DDDDDDDDDDDDG.info.jsonl"),
+      join(storageRoot, "logs", "C9NMV6V2TQT81", "01LIVECMPLT01DDDDDDDDDDDDG.info.jsonl"),
+    );
+    await cp(
+      join(fixtureRoot, "logs", "C9NMV6V2TQT81", "01LIVEINFLY01DDDDDDDDDDDDG.data.jsonl"),
+      join(storageRoot, "logs", "C9NMV6V2TQT81", "01LIVEINFLY01DDDDDDDDDDDDG.data.jsonl"),
+    );
+    await cp(
+      join(fixtureRoot, "logs", "C9NMV6V2TQT81", "01LIVEOLDER01DDDDDDDDDDDDG.data.jsonl"),
+      join(storageRoot, "logs", "C9NMV6V2TQT81", "01LIVEOLDER01DDDDDDDDDDDDG.data.jsonl"),
+    );
+
+    const cas = createCasStore(getGlobalCasDir(storageRoot));
+    await putContentMerkleNode(cas, LIVE_FIXTURE_PLANNER_BODY);
+    await putContentMerkleNode(cas, "patch");
+    await putContentMerkleNode(cas, "still running");
+  });
+
+  afterEach(async () => {
+    if (prevEnv === undefined) {
+      delete process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    } else {
+      process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = prevEnv;
+    }
+    await rm(storageRoot, { recursive: true, force: true });
+  });
+
+  test("prints role steps and summary for a completed thread", async () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const proc = spawn(process.execPath, [cliEntryPath, "live", "01LIVECMPLT01DDDDDDDDDDDDG"], {
+      env,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+    const stdout = await new Promise<string>((resolve, reject) => {
+      let buf = "";
+      proc.stdout?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.stderr?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.on("error", reject);
+      proc.on("exit", (code: number | null) => {
+        if (code === 0) {
+          resolve(buf);
+        } else {
+          reject(new Error(`exit ${code}: ${buf}`));
+        }
+      });
+    });
+
+    expect(stdout).toContain("planner");
+    expect(stdout).toContain("coder");
+    expect(stdout).toContain("meta:");
+    expect(stdout).toContain('"phase":"plan"');
+    expect(stdout).toContain("LINE10");
+    expect(stdout).not.toContain("LINE11");
+    expect(stdout).toContain("more line");
+    expect(stdout).toContain("completed: returnCode=0");
+    expect(stdout).toContain("fixture completed");
+  });
+
+  test("--latest tails the newest thread by start timestamp", async () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const proc = spawn(process.execPath, [cliEntryPath, "live", "--latest"], {
+      env,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+    const stdout = await new Promise<string>((resolve, reject) => {
+      let buf = "";
+      proc.stdout?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.stderr?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.on("error", reject);
+      proc.on("exit", (code: number | null) => {
+        if (code === 0) {
+          resolve(buf);
+        } else {
+          reject(new Error(`exit ${code}: ${buf}`));
+        }
+      });
+    });
+
+    expect(stdout).toContain("fixture completed");
+    expect(stdout).not.toContain("older thread");
+  });
+
+  test("--debug prints .info.jsonl records after data output", async () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const proc = spawn(
+      process.execPath,
+      [cliEntryPath, "live", "01LIVECMPLT01DDDDDDDDDDDDG", "--debug"],
+      {
+        env,
+        stdio: ["ignore", "pipe", "pipe"],
+      },
+    );
+    const stdout = await new Promise<string>((resolve, reject) => {
+      let buf = "";
+      proc.stdout?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.stderr?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.on("error", reject);
+      proc.on("exit", (code: number | null) => {
+        if (code === 0) {
+          resolve(buf);
+        } else {
+          reject(new Error(`exit ${code}: ${buf}`));
+        }
+      });
+    });
+
+    expect(stdout).toContain("[DEBUGTAG1]");
+    expect(stdout).toContain("bundle loaded");
+    expect(stdout).toContain("[DEBUGTAG2]");
+    expect(stdout).toContain("multi line");
+  });
+
+  test("--role filters out non-matching roles", async () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const proc = spawn(
+      process.execPath,
+      [cliEntryPath, "live", "01LIVECMPLT01DDDDDDDDDDDDG", "--role", "planner"],
+      {
+        env,
+        stdio: ["ignore", "pipe", "pipe"],
+      },
+    );
+    const stdout = await new Promise<string>((resolve, reject) => {
+      let buf = "";
+      proc.stdout?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.stderr?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.on("error", reject);
+      proc.on("exit", (code: number | null) => {
+        if (code === 0) {
+          resolve(buf);
+        } else {
+          reject(new Error(`exit ${code}: ${buf}`));
+        }
+      });
+    });
+
+    expect(stdout).toContain("planner");
+    expect(stdout).not.toContain("patch");
+    expect(stdout).toContain("completed: returnCode=0");
+  });
+
+  test("--latest --debug --role combine", async () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const proc = spawn(
+      process.execPath,
+      [cliEntryPath, "live", "--latest", "--debug", "--role", "planner"],
+      {
+        env,
+        stdio: ["ignore", "pipe", "pipe"],
+      },
+    );
+    const stdout = await new Promise<string>((resolve, reject) => {
+      let buf = "";
+      proc.stdout?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.stderr?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.on("error", reject);
+      proc.on("exit", (code: number | null) => {
+        if (code === 0) {
+          resolve(buf);
+        } else {
+          reject(new Error(`exit ${code}: ${buf}`));
+        }
+      });
+    });
+
+    expect(stdout).toContain("[DEBUGTAG1]");
+    expect(stdout).toContain("planner");
+    expect(stdout).not.toContain("patch");
+    expect(stdout).toContain("fixture completed");
+  });
+
+  test("unknown thread id exits 1", () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const r = spawnSync(process.execPath, [cliEntryPath, "live", "01UNKNOWNXXXXXXXXXXXXXXXXX"], {
+      env,
+      encoding: "utf8",
+    });
+    expect(r.status).toBe(1);
+    expect(String(r.stderr ?? "")).toContain("thread not found");
+  });
+
+  test("follows file until WorkflowResult is appended", async () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
+    const dataPath = join(
+      storageRoot,
+      "logs",
+      "C9NMV6V2TQT81",
+      "01LIVEINFLY01DDDDDDDDDDDDG.data.jsonl",
+    );
+
+    const proc = spawn(process.execPath, [cliEntryPath, "live", "01LIVEINFLY01DDDDDDDDDDDDG"], {
+      env,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+
+    await new Promise((r) => setTimeout(r, 120));
+    const prior = await readFile(dataPath, "utf8");
+    await writeFile(
+      dataPath,
+      `${prior.replace(/\s*$/, "")}\n${JSON.stringify({ returnCode: 0, summary: "caught up" })}\n`,
+      "utf8",
+    );
+
+    const stdout = await new Promise<string>((resolve, reject) => {
+      let buf = "";
+      proc.stdout?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.stderr?.on("data", (c: Buffer) => {
+        buf += c.toString("utf8");
+      });
+      proc.on("error", reject);
+      proc.on("exit", (code: number | null) => {
+        if (code === 0) {
+          resolve(buf);
+        } else {
+          reject(new Error(`exit ${code}: ${buf}`));
+        }
+      });
+    });
+
+    expect(stdout).toContain("planner");
+    expect(stdout).toContain("completed: returnCode=0");
+    expect(stdout).toContain("caught up");
+  });
+});
+
+describe("live --latest with empty storage", () => {
+  let prevEnv: string | undefined;
+  let emptyRoot: string;
+
+  beforeEach(async () => {
+    prevEnv = process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    emptyRoot = await mkdtemp(join(tmpdir(), "uncaged-wf-live-empty-"));
+    process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = emptyRoot;
+  });
+
+  afterEach(async () => {
+    if (prevEnv === undefined) {
+      delete process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    } else {
+      process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = prevEnv;
+    }
+    await rm(emptyRoot, { recursive: true, force: true });
+  });
+
+  test("exits 1 when no threads exist", () => {
+    const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: emptyRoot };
+    const r = spawnSync(process.execPath, [cliEntryPath, "live", "--latest"], {
+      env,
+      encoding: "utf8",
+    });
+    expect(r.status).toBe(1);
+    expect(String(r.stderr ?? "")).toContain("no threads");
+  });
+});

packages/cli-workflow/__tests__/storage-env.test.ts

@@ -0,0 +1,54 @@
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { getDefaultWorkflowStorageRoot } from "@uncaged/workflow";
+import { resolveWorkflowStorageRoot } from "../src/storage-env.js";
+
+describe("resolveWorkflowStorageRoot", () => {
+  let savedInternal: string | undefined;
+  let savedUser: string | undefined;
+
+  beforeEach(() => {
+    savedInternal = process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    savedUser = process.env.WORKFLOW_STORAGE_ROOT;
+    delete process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    delete process.env.WORKFLOW_STORAGE_ROOT;
+  });
+
+  afterEach(() => {
+    if (savedInternal === undefined) {
+      delete process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+    } else {
+      process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = savedInternal;
+    }
+    if (savedUser === undefined) {
+      delete process.env.WORKFLOW_STORAGE_ROOT;
+    } else {
+      process.env.WORKFLOW_STORAGE_ROOT = savedUser;
+    }
+  });
+
+  test("returns default when no env vars are set", () => {
+    expect(resolveWorkflowStorageRoot()).toBe(getDefaultWorkflowStorageRoot());
+  });
+
+  test("WORKFLOW_STORAGE_ROOT overrides default", () => {
+    process.env.WORKFLOW_STORAGE_ROOT = "/tmp/custom-storage";
+    expect(resolveWorkflowStorageRoot()).toBe("/tmp/custom-storage");
+  });
+
+  test("UNCAGED_WORKFLOW_STORAGE_ROOT takes priority over WORKFLOW_STORAGE_ROOT", () => {
+    process.env.WORKFLOW_STORAGE_ROOT = "/tmp/user-path";
+    process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = "/tmp/internal-path";
+    expect(resolveWorkflowStorageRoot()).toBe("/tmp/internal-path");
+  });
+
+  test("ignores empty WORKFLOW_STORAGE_ROOT", () => {
+    process.env.WORKFLOW_STORAGE_ROOT = "";
+    expect(resolveWorkflowStorageRoot()).toBe(getDefaultWorkflowStorageRoot());
+  });
+
+  test("ignores empty UNCAGED_WORKFLOW_STORAGE_ROOT and falls through to WORKFLOW_STORAGE_ROOT", () => {
+    process.env.UNCAGED_WORKFLOW_STORAGE_ROOT = "";
+    process.env.WORKFLOW_STORAGE_ROOT = "/tmp/user-fallback";
+    expect(resolveWorkflowStorageRoot()).toBe("/tmp/user-fallback");
+  });
+});

packages/cli-workflow/__tests__/thread-cli.test.ts

@@ -4,7 +4,9 @@ import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
+import { getGlobalCasDir } from "@uncaged/workflow";
 import { cmdAdd } from "../src/cmd-add.js";
+import { cmdCasPut } from "../src/cmd-cas.js";
 import { cmdKill } from "../src/cmd-kill.js";
 import { cmdPause } from "../src/cmd-pause.js";
 import { cmdPs } from "../src/cmd-ps.js";
@@ -12,9 +14,12 @@ import { cmdResume } from "../src/cmd-resume.js";
 import { cmdRun } from "../src/cmd-run.js";
 import { cmdThreadRemove, cmdThreadShow } from "../src/cmd-thread.js";
 import { cmdThreads } from "../src/cmd-threads.js";
-import { pathExists } from "../src/fs-utils.js";
+import { pathExists, readTextFileIfExists } from "../src/fs-utils.js";
 import { addCliArgs } from "./bundle-fixture.js";
 
+const wfPutImport = `import { putContentMerkleNode } from "@uncaged/workflow";
+`;
+
 const threadFixtureDescriptor = `export const descriptor = {
   description: "thread-cli",
   roles: {
@@ -29,18 +34,26 @@ const threadFixtureDescriptor = `export const descriptor = {
 `;
 
 const fastBundleSource = `${threadFixtureDescriptor}
-export const run = async function* (input) {
-  yield { role: "planner", content: "plan", meta: { plan: input.prompt } };
-  yield { role: "coder", content: "code", meta: { diff: "y" } };
+${wfPutImport}
+export const run = async function* (input, options) {
+  const cas = options.cas;
+  let h = await putContentMerkleNode(cas, "plan");
+  yield { role: "planner", contentHash: h, meta: { plan: input.prompt }, refs: [h] };
+  h = await putContentMerkleNode(cas, "code");
+  yield { role: "coder", contentHash: h, meta: { diff: "y" }, refs: [h] };
   return { returnCode: 0, summary: "done" };
 };
 `;
 
 const slowPlannerBundleSource = `${threadFixtureDescriptor}
-export const run = async function* (input) {
+${wfPutImport}
+export const run = async function* (input, options) {
   await new Promise((r) => setTimeout(r, 400));
-  yield { role: "planner", content: "plan", meta: { plan: input.prompt } };
-  yield { role: "coder", content: "code", meta: { diff: "y" } };
+  const cas = options.cas;
+  let h = await putContentMerkleNode(cas, "plan");
+  yield { role: "planner", contentHash: h, meta: { plan: input.prompt }, refs: [h] };
+  h = await putContentMerkleNode(cas, "code");
+  yield { role: "coder", contentHash: h, meta: { diff: "y" }, refs: [h] };
   return { returnCode: 0, summary: "done" };
 };
 `;
@@ -48,27 +61,38 @@ export const run = async function* (input) {
 const cliEntryPath = fileURLToPath(new URL("../src/cli.ts", import.meta.url));
 
 const abortablePlannerBundleSource = `${threadFixtureDescriptor}
-export const run = async function* (input) {
+${wfPutImport}
+export const run = async function* (input, options) {
   await new Promise((r) => setTimeout(r, 600));
-  yield { role: "planner", content: "plan", meta: { plan: input.prompt } };
-  yield { role: "coder", content: "code", meta: { diff: "y" } };
+  const cas = options.cas;
+  let h = await putContentMerkleNode(cas, "plan");
+  yield { role: "planner", contentHash: h, meta: { plan: input.prompt }, refs: [h] };
+  h = await putContentMerkleNode(cas, "code");
+  yield { role: "coder", contentHash: h, meta: { diff: "y" }, refs: [h] };
   return { returnCode: 0, summary: "done" };
 };
 `;
 
 const pauseResumeBundleSource = `${threadFixtureDescriptor}
-export const run = async function* (input) {
-  yield { role: "first", content: "f", meta: {} };
+${wfPutImport}
+export const run = async function* (_input, options) {
+  const cas = options.cas;
+  let h = await putContentMerkleNode(cas, "f");
+  yield { role: "first", contentHash: h, meta: {}, refs: [h] };
   await new Promise((r) => setTimeout(r, 1500));
-  yield { role: "second", content: "s", meta: {} };
+  h = await putContentMerkleNode(cas, "s");
+  yield { role: "second", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "done" };
 };
 `;
 
 const delayedFirstYieldBundleSource = `${threadFixtureDescriptor}
-export const run = async function* (input) {
+${wfPutImport}
+export const run = async function* (_input, options) {
   await new Promise((r) => setTimeout(r, 900));
-  yield { role: "only", content: "x", meta: {} };
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  yield { role: "only", contentHash: h, meta: {}, refs: [h] };
   return { returnCode: 0, summary: "done" };
 };
 `;
@@ -175,15 +199,67 @@ describe("cli thread commands", () => {
     expect(await pathExists(dataPath)).toBe(false);
   });
 
+  test("thread rm runs GC and removes CAS blobs not referenced by any remaining thread", async () => {
+    const bundleDir = join(storageRoot, "src");
+    await mkdir(bundleDir, { recursive: true });
+    const bundlePath = join(bundleDir, "demo.esm.js");
+    await writeFile(bundlePath, fastBundleSource, "utf8");
+
+    const added = await cmdAdd(storageRoot, addCliArgs("solve-issue", bundlePath));
+    expect(added.ok).toBe(true);
+    if (!added.ok) {
+      return;
+    }
+
+    const ran = await cmdRun(storageRoot, "solve-issue", "hello", 5);
+    expect(ran.ok).toBe(true);
+    if (!ran.ok) {
+      return;
+    }
+
+    const threadId = ran.value.threadId;
+
+    let threads = await cmdThreads(storageRoot, []);
+    for (
+      let attempt = 0;
+      attempt < 50 && threads.ok && !threads.value.some((l) => l.includes(threadId));
+      attempt++
+    ) {
+      await new Promise((r) => setTimeout(r, 20));
+      threads = await cmdThreads(storageRoot, []);
+    }
+
+    const dataPath = join(storageRoot, "logs", added.value.hash, `${threadId}.data.jsonl`);
+    const runningPath = join(dirname(dataPath), `${threadId}.running`);
+    await waitUntilRunningFileAbsent(runningPath, 120);
+
+    const put = await cmdCasPut(storageRoot, threadId, "keep-after-thread-rm");
+    expect(put.ok).toBe(true);
+    if (!put.ok) {
+      return;
+    }
+    const hash = put.value;
+    const casBlob = join(getGlobalCasDir(storageRoot), `${hash}.txt`);
+
+    const removed = await cmdThreadRemove(storageRoot, threadId);
+    expect(removed.ok).toBe(true);
+
+    const stillThere = await readTextFileIfExists(casBlob);
+    expect(stillThere).toBeNull();
+  });
+
   test("cli entrypoint dispatches threads / ps (spawn)", () => {
     const env = { ...process.env, UNCAGED_WORKFLOW_STORAGE_ROOT: storageRoot };
-    const threads = spawnSync(process.execPath, [cliEntryPath, "threads"], {
+    const threads = spawnSync(process.execPath, [cliEntryPath, "thread", "list"], {
       env,
       encoding: "utf8",
     });
     expect(threads.status).toBe(0);
 
-    const ps = spawnSync(process.execPath, [cliEntryPath, "ps"], { env, encoding: "utf8" });
+    const ps = spawnSync(process.execPath, [cliEntryPath, "thread", "ps"], {
+      env,
+      encoding: "utf8",
+    });
     expect(ps.status).toBe(0);
   });
 
@@ -250,7 +326,7 @@ describe("cli thread commands", () => {
       .trim()
       .split("\n")
       .filter((l) => l !== "");
-    expect(lines.length).toBe(2);
+    expect(lines.length).toBe(3);
 
     const runningPath = join(dirname(dataPath), `${threadId}.running`);
     expect(await pathExists(runningPath)).toBe(false);
@@ -289,8 +365,8 @@ describe("cli thread commands", () => {
     const resumed = await cmdResume(storageRoot, threadId);
     expect(resumed.ok).toBe(true);
 
-    await waitUntilMinDataLines(dataPath, 3, 120);
-    expect(await countDataJsonlLines(dataPath)).toBe(3);
+    await waitUntilMinDataLines(dataPath, 4, 120);
+    expect(await countDataJsonlLines(dataPath)).toBe(4);
 
     const runningPath = join(dirname(dataPath), `${threadId}.running`);
     await waitUntilRunningFileAbsent(runningPath, 100);

packages/cli-workflow/src/cli-dispatch.ts

@@ -2,9 +2,13 @@ import { printCliError, printCliLine, printCliWarn } from "./cli-output.js";
 import { cmdAdd, formatAddSuccess, parseAddArgv } from "./cmd-add.js";
 import { cmdCasGet, cmdCasList, cmdCasPut, cmdCasRm } from "./cmd-cas.js";
 import { cmdFork, parseForkArgv } from "./cmd-fork.js";
+import { cmdGc } from "./cmd-gc.js";
+import { formatSkillDoc, formatSkillIndex, formatSkillTopic } from "./cmd-help.js";
 import { cmdHistory } from "./cmd-history.js";
+import { cmdInitTemplate, cmdInitWorkspace } from "./cmd-init.js";
 import { cmdKill } from "./cmd-kill.js";
 import { cmdList, formatListLines } from "./cmd-list.js";
+import { cmdLive } from "./cmd-live.js";
 import { cmdPause } from "./cmd-pause.js";
 import { cmdPs } from "./cmd-ps.js";
 import { cmdRemove } from "./cmd-remove.js";
@@ -14,37 +18,58 @@ import { cmdRun } from "./cmd-run.js";
 import { cmdShow, formatShowYaml } from "./cmd-show.js";
 import { cmdThreadRemove, cmdThreadShow } from "./cmd-thread.js";
 import { cmdThreads } from "./cmd-threads.js";
+import { parseLiveArgv } from "./live-argv.js";
 import { parseRunArgv } from "./run-argv.js";
 
-function usage(): string {
-  return [
-    "Usage:",
-    "  uncaged-workflow add <name> <file.esm.js> [--types <path>]",
-    "  uncaged-workflow list",
-    "  uncaged-workflow show <name>",
-    "  uncaged-workflow remove <name>",
-    "  uncaged-workflow run <name> [--prompt <text>] [--max-rounds N]",
-    "  uncaged-workflow ps",
-    "  uncaged-workflow kill <thread-id>",
-    "  uncaged-workflow history <name>",
-    "  uncaged-workflow rollback <name> [hash]",
-    "  uncaged-workflow pause <thread-id>",
-    "  uncaged-workflow resume <thread-id>",
-    "  uncaged-workflow threads [name]",
-    "  uncaged-workflow thread <id>",
-    "  uncaged-workflow thread rm <id>",
-    "  uncaged-workflow fork <thread-id> [--from-role <role>]",
-    "  uncaged-workflow cas get <thread-id> <hash>",
-    "  uncaged-workflow cas put <thread-id> <content>",
-    "  uncaged-workflow cas list <thread-id>",
-    "  uncaged-workflow cas rm <thread-id> <hash>",
-  ].join("\n");
+type DispatchFn = (storageRoot: string, argv: string[]) => Promise<number>;
+
+type CommandEntry = {
+  handler: DispatchFn;
+  args: string;
+  description: string;
+};
+
+type CommandGroup = {
+  name: string;
+  commands: ReadonlyArray<{ name: string; args: string; description: string }>;
+};
+
+// ── Individual dispatch functions ──────────────────────────────────────
+
+async function dispatchInitWorkspace(_storageRoot: string, argv: string[]): Promise<number> {
+  const name = argv[0];
+  if (name === undefined || argv.length > 1) {
+    printCliError(`${formatCliUsage()}\n\nerror: init workspace requires <name>`);
+    return 1;
+  }
+  const result = await cmdInitWorkspace(process.cwd(), name);
+  if (!result.ok) {
+    printCliError(result.error);
+    return 1;
+  }
+  printCliLine(`initialized workflow workspace at ${result.value.rootPath}`);
+  return 0;
+}
+
+async function dispatchInitTemplate(_storageRoot: string, argv: string[]): Promise<number> {
+  const name = argv[0];
+  if (name === undefined || argv.length > 1) {
+    printCliError(`${formatCliUsage()}\n\nerror: init template requires <name>`);
+    return 1;
+  }
+  const result = await cmdInitTemplate(process.cwd(), name);
+  if (!result.ok) {
+    printCliError(result.error);
+    return 1;
+  }
+  printCliLine(`initialized template at ${result.value.templatePath}`);
+  return 0;
 }
 
 async function dispatchAdd(storageRoot: string, argv: string[]): Promise<number> {
   const parsed = parseAddArgv(argv);
   if (!parsed.ok) {
-    printCliError(`${usage()}\n\nerror: ${parsed.error}`);
+    printCliError(`${formatCliUsage()}\n\nerror: ${parsed.error}`);
     return 1;
   }
   const result = await cmdAdd(storageRoot, parsed.value);
@@ -61,7 +86,7 @@ async function dispatchAdd(storageRoot: string, argv: string[]): Promise<number>
 
 async function dispatchList(storageRoot: string, argv: string[]): Promise<number> {
   if (argv.length > 0) {
-    printCliError(`${usage()}\n\nerror: list takes no arguments`);
+    printCliError(`${formatCliUsage()}\n\nerror: list takes no arguments`);
     return 1;
   }
   const result = await cmdList(storageRoot);
@@ -78,7 +103,7 @@ async function dispatchList(storageRoot: string, argv: string[]): Promise<number
 async function dispatchShow(storageRoot: string, argv: string[]): Promise<number> {
   const name = argv[0];
   if (name === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: show requires <name>`);
+    printCliError(`${formatCliUsage()}\n\nerror: show requires <name>`);
     return 1;
   }
   const result = await cmdShow(storageRoot, name);
@@ -93,7 +118,7 @@ async function dispatchShow(storageRoot: string, argv: string[]): Promise<number
 async function dispatchRemove(storageRoot: string, argv: string[]): Promise<number> {
   const name = argv[0];
   if (name === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: remove requires <name>`);
+    printCliError(`${formatCliUsage()}\n\nerror: remove requires <name>`);
     return 1;
   }
   const result = await cmdRemove(storageRoot, name);
@@ -108,7 +133,7 @@ async function dispatchRemove(storageRoot: string, argv: string[]): Promise<numb
 async function dispatchRun(storageRoot: string, argv: string[]): Promise<number> {
   const parsed = parseRunArgv(argv);
   if (!parsed.ok) {
-    printCliError(`${usage()}\n\nerror: ${parsed.error}`);
+    printCliError(`${formatCliUsage()}\n\nerror: ${parsed.error}`);
     return 1;
   }
 
@@ -129,7 +154,7 @@ async function dispatchRun(storageRoot: string, argv: string[]): Promise<number>
 
 async function dispatchPs(storageRoot: string, argv: string[]): Promise<number> {
   if (argv.length > 0) {
-    printCliError(`${usage()}\n\nerror: ps takes no arguments`);
+    printCliError(`${formatCliUsage()}\n\nerror: ps takes no arguments`);
     return 1;
   }
   for (const line of await cmdPs(storageRoot)) {
@@ -141,7 +166,7 @@ async function dispatchPs(storageRoot: string, argv: string[]): Promise<number>
 async function dispatchKill(storageRoot: string, argv: string[]): Promise<number> {
   const threadId = argv[0];
   if (threadId === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: kill requires <thread-id>`);
+    printCliError(`${formatCliUsage()}\n\nerror: kill requires <thread-id>`);
     return 1;
   }
   const result = await cmdKill(storageRoot, threadId);
@@ -153,10 +178,19 @@ async function dispatchKill(storageRoot: string, argv: string[]): Promise<number
   return 0;
 }
 
+async function dispatchLive(storageRoot: string, argv: string[]): Promise<number> {
+  const parsed = parseLiveArgv(argv);
+  if (!parsed.ok) {
+    printCliError(`${formatCliUsage()}\n\nerror: ${parsed.error}`);
+    return 1;
+  }
+  return cmdLive(storageRoot, parsed.value);
+}
+
 async function dispatchHistory(storageRoot: string, argv: string[]): Promise<number> {
   const name = argv[0];
   if (name === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: history requires <name>`);
+    printCliError(`${formatCliUsage()}\n\nerror: history requires <name>`);
     return 1;
   }
   const result = await cmdHistory(storageRoot, name);
@@ -173,7 +207,7 @@ async function dispatchHistory(storageRoot: string, argv: string[]): Promise<num
 async function dispatchRollback(storageRoot: string, argv: string[]): Promise<number> {
   const name = argv[0];
   if (name === undefined || argv.length > 2) {
-    printCliError(`${usage()}\n\nerror: rollback requires <name> [hash]`);
+    printCliError(`${formatCliUsage()}\n\nerror: rollback requires <name> [hash]`);
     return 1;
   }
   const hashArg = argv[1];
@@ -189,7 +223,7 @@ async function dispatchRollback(storageRoot: string, argv: string[]): Promise<nu
 async function dispatchPause(storageRoot: string, argv: string[]): Promise<number> {
   const threadId = argv[0];
   if (threadId === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: pause requires <thread-id>`);
+    printCliError(`${formatCliUsage()}\n\nerror: pause requires <thread-id>`);
     return 1;
   }
   const result = await cmdPause(storageRoot, threadId);
@@ -204,7 +238,7 @@ async function dispatchPause(storageRoot: string, argv: string[]): Promise<numbe
 async function dispatchResume(storageRoot: string, argv: string[]): Promise<number> {
   const threadId = argv[0];
   if (threadId === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: resume requires <thread-id>`);
+    printCliError(`${formatCliUsage()}\n\nerror: resume requires <thread-id>`);
     return 1;
   }
   const result = await cmdResume(storageRoot, threadId);
@@ -216,7 +250,7 @@ async function dispatchResume(storageRoot: string, argv: string[]): Promise<numb
   return 0;
 }
 
-async function dispatchThreads(storageRoot: string, argv: string[]): Promise<number> {
+async function dispatchThreadList(storageRoot: string, argv: string[]): Promise<number> {
   const result = await cmdThreads(storageRoot, argv);
   if (!result.ok) {
     printCliError(result.error);
@@ -228,10 +262,10 @@ async function dispatchThreads(storageRoot: string, argv: string[]): Promise<num
   return 0;
 }
 
-async function dispatchThread(storageRoot: string, argv: string[]): Promise<number> {
+async function dispatchThreadShow(storageRoot: string, argv: string[]): Promise<number> {
   const id = argv[0];
   if (id === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: thread requires <id>`);
+    printCliError(`${formatCliUsage()}\n\nerror: thread show requires <id>`);
     return 1;
   }
   const result = await cmdThreadShow(storageRoot, id);
@@ -246,7 +280,7 @@ async function dispatchThread(storageRoot: string, argv: string[]): Promise<numb
 async function dispatchThreadRm(storageRoot: string, argv: string[]): Promise<number> {
   const id = argv[0];
   if (id === undefined || argv.length > 1) {
-    printCliError(`${usage()}\n\nerror: thread rm requires <id>`);
+    printCliError(`${formatCliUsage()}\n\nerror: thread rm requires <id>`);
     return 1;
   }
   const result = await cmdThreadRemove(storageRoot, id);
@@ -258,18 +292,27 @@ async function dispatchThreadRm(storageRoot: string, argv: string[]): Promise<nu
   return 0;
 }
 
-async function dispatchThreadBranch(storageRoot: string, rest: string[]): Promise<number> {
-  const sub = rest[0];
-  if (sub === "rm") {
-    return dispatchThreadRm(storageRoot, rest.slice(1));
+async function dispatchGc(storageRoot: string, argv: string[]): Promise<number> {
+  if (argv.length > 0) {
+    printCliError(`${formatCliUsage()}\n\nerror: gc takes no arguments`);
+    return 1;
   }
-  return dispatchThread(storageRoot, rest);
+  const result = await cmdGc(storageRoot);
+  if (!result.ok) {
+    printCliError(result.error);
+    return 1;
+  }
+  const stats = result.value;
+  printCliLine(
+    `scanned ${stats.scannedThreads} threads, ${stats.activeRefs} active refs, deleted ${stats.deletedEntries} entries`,
+  );
+  return 0;
 }
 
 async function dispatchFork(storageRoot: string, argv: string[]): Promise<number> {
   const parsed = parseForkArgv(argv);
   if (!parsed.ok) {
-    printCliError(`${usage()}\n\nerror: ${parsed.error}`);
+    printCliError(`${formatCliUsage()}\n\nerror: ${parsed.error}`);
     return 1;
   }
   const result = await cmdFork(storageRoot, parsed.value.threadId, parsed.value.fromRole);
@@ -281,11 +324,13 @@ async function dispatchFork(storageRoot: string, argv: string[]): Promise<number
   return 0;
 }
 
+// ── CAS subcommand table ───────────────────────────────────────────────
+
 async function dispatchCasGet(storageRoot: string, rest: string[]): Promise<number> {
   const threadId = rest[0];
   const hash = rest[1];
   if (threadId === undefined || hash === undefined || rest.length > 2) {
-    printCliError(`${usage()}\n\nerror: cas get requires <thread-id> <hash>`);
+    printCliError(`${formatCliUsage()}\n\nerror: cas get requires <thread-id> <hash>`);
     return 1;
   }
   const result = await cmdCasGet(storageRoot, threadId, hash);
@@ -301,7 +346,7 @@ async function dispatchCasPut(storageRoot: string, rest: string[]): Promise<numb
   const threadId = rest[0];
   const content = rest[1];
   if (threadId === undefined || content === undefined || rest.length > 2) {
-    printCliError(`${usage()}\n\nerror: cas put requires <thread-id> <content>`);
+    printCliError(`${formatCliUsage()}\n\nerror: cas put requires <thread-id> <content>`);
     return 1;
   }
   const result = await cmdCasPut(storageRoot, threadId, content);
@@ -316,7 +361,7 @@ async function dispatchCasPut(storageRoot: string, rest: string[]): Promise<numb
 async function dispatchCasList(storageRoot: string, rest: string[]): Promise<number> {
   const threadId = rest[0];
   if (threadId === undefined || rest.length > 1) {
-    printCliError(`${usage()}\n\nerror: cas list requires <thread-id>`);
+    printCliError(`${formatCliUsage()}\n\nerror: cas list requires <thread-id>`);
     return 1;
   }
   const result = await cmdCasList(storageRoot, threadId);
@@ -334,7 +379,7 @@ async function dispatchCasRm(storageRoot: string, rest: string[]): Promise<numbe
   const threadId = rest[0];
   const hash = rest[1];
   if (threadId === undefined || hash === undefined || rest.length > 2) {
-    printCliError(`${usage()}\n\nerror: cas rm requires <thread-id> <hash>`);
+    printCliError(`${formatCliUsage()}\n\nerror: cas rm requires <thread-id> <hash>`);
     return 1;
   }
   const result = await cmdCasRm(storageRoot, threadId, hash);
@@ -346,65 +391,335 @@ async function dispatchCasRm(storageRoot: string, rest: string[]): Promise<numbe
   return 0;
 }
 
-const CAS_SUBCOMMAND_TABLE: Record<
-  string,
-  (storageRoot: string, rest: string[]) => Promise<number>
-> = {
-  get: dispatchCasGet,
-  put: dispatchCasPut,
-  list: dispatchCasList,
-  rm: dispatchCasRm,
+// ── Subcommand tables with metadata ────────────────────────────────────
+
+const WORKFLOW_SUBCOMMAND_TABLE: Record<string, CommandEntry> = {
+  add: {
+    handler: dispatchAdd,
+    args: "<name> <file.esm.js> [--types <path>]",
+    description: "Register a workflow bundle in the registry",
+  },
+  list: { handler: dispatchList, args: "", description: "List all registered workflows" },
+  show: {
+    handler: dispatchShow,
+    args: "<name>",
+    description: "Show details of a registered workflow",
+  },
+  rm: {
+    handler: dispatchRemove,
+    args: "<name>",
+    description: "Remove a workflow from the registry",
+  },
+  history: {
+    handler: dispatchHistory,
+    args: "<name>",
+    description: "Show version history of a workflow",
+  },
+  rollback: {
+    handler: dispatchRollback,
+    args: "<name> [hash]",
+    description: "Rollback a workflow to a previous version",
+  },
 };
 
-async function dispatchCas(storageRoot: string, argv: string[]): Promise<number> {
-  const sub = argv[0];
-  if (sub === undefined) {
-    printCliError(`${usage()}\n\nerror: unknown cas subcommand: (none)`);
-    return 1;
-  }
-  const handler = CAS_SUBCOMMAND_TABLE[sub];
-  if (handler === undefined) {
-    printCliError(`${usage()}\n\nerror: unknown cas subcommand: ${sub}`);
-    return 1;
-  }
-  return handler(storageRoot, argv.slice(1));
+const THREAD_SUBCOMMAND_TABLE: Record<string, CommandEntry> = {
+  run: {
+    handler: dispatchRun,
+    args: "<name> [--prompt <text>] [--max-rounds N]",
+    description: "Start a new thread executing a workflow",
+  },
+  list: {
+    handler: dispatchThreadList,
+    args: "[name]",
+    description: "List threads, optionally filtered by workflow name",
+  },
+  show: { handler: dispatchThreadShow, args: "<id>", description: "Show thread details and state" },
+  rm: { handler: dispatchThreadRm, args: "<id>", description: "Remove a thread" },
+  fork: {
+    handler: dispatchFork,
+    args: "<thread-id> [--from-role <role>]",
+    description: "Fork a thread, optionally from a specific role",
+  },
+  ps: { handler: dispatchPs, args: "", description: "List running threads" },
+  kill: { handler: dispatchKill, args: "<thread-id>", description: "Kill a running thread" },
+  live: {
+    handler: dispatchLive,
+    args: "<thread-id> | --latest [--debug] [--role <name>]",
+    description: "Attach to a thread and stream output live",
+  },
+  pause: { handler: dispatchPause, args: "<thread-id>", description: "Pause a running thread" },
+  resume: { handler: dispatchResume, args: "<thread-id>", description: "Resume a paused thread" },
+};
+
+const CAS_SUBCOMMAND_TABLE: Record<string, CommandEntry> = {
+  get: {
+    handler: dispatchCasGet,
+    args: "<thread-id> <hash>",
+    description: "Retrieve content by hash from a thread's CAS",
+  },
+  put: {
+    handler: dispatchCasPut,
+    args: "<thread-id> <content>",
+    description: "Store content in a thread's CAS, returns hash",
+  },
+  list: {
+    handler: dispatchCasList,
+    args: "<thread-id>",
+    description: "List all CAS entries for a thread",
+  },
+  rm: { handler: dispatchCasRm, args: "<thread-id> <hash>", description: "Remove a CAS entry" },
+  gc: { handler: dispatchGc, args: "", description: "Garbage-collect unreferenced CAS entries" },
+};
+
+const INIT_SUBCOMMAND_TABLE: Record<string, CommandEntry> = {
+  workspace: {
+    handler: dispatchInitWorkspace,
+    args: "<name>",
+    description: "Initialize a new workflow workspace",
+  },
+  template: {
+    handler: dispatchInitTemplate,
+    args: "<name>",
+    description: "Initialize a new workflow template",
+  },
+};
+
+// ── Command registry ───────────────────────────────────────────────────
+
+export function getCommandRegistry(): ReadonlyArray<CommandGroup> {
+  return [
+    {
+      name: "workflow",
+      commands: Object.entries(WORKFLOW_SUBCOMMAND_TABLE).map(([name, e]) => ({
+        name,
+        args: e.args,
+        description: e.description,
+      })),
+    },
+    {
+      name: "thread",
+      commands: Object.entries(THREAD_SUBCOMMAND_TABLE).map(([name, e]) => ({
+        name,
+        args: e.args,
+        description: e.description,
+      })),
+    },
+    {
+      name: "cas",
+      commands: Object.entries(CAS_SUBCOMMAND_TABLE).map(([name, e]) => ({
+        name,
+        args: e.args,
+        description: e.description,
+      })),
+    },
+    {
+      name: "init",
+      commands: Object.entries(INIT_SUBCOMMAND_TABLE).map(([name, e]) => ({
+        name,
+        args: e.args,
+        description: e.description,
+      })),
+    },
+  ];
 }
 
-type DispatchFn = (storageRoot: string, argv: string[]) => Promise<number>;
+// ── Auto-generated CLI usage ───────────────────────────────────────────
+
+export function formatCliUsage(): string {
+  const groups = getCommandRegistry();
+  const lines: string[] = ["Usage:"];
+  for (const group of groups) {
+    for (const cmd of group.commands) {
+      const args = cmd.args ? ` ${cmd.args}` : "";
+      lines.push(`  uncaged-workflow ${group.name} ${cmd.name}${args}`);
+    }
+    lines.push("");
+  }
+  lines.push("  uncaged-workflow run <name> [...]          (shortcut for thread run)");
+  lines.push("  uncaged-workflow live <thread-id> [...]    (shortcut for thread live)");
+  lines.push("");
+  lines.push("  uncaged-workflow skill [topic]             agent-consumable reference docs");
+  lines.push("  uncaged-workflow help                      show this usage");
+  lines.push("");
+  lines.push("Environment variables:");
+  lines.push(
+    "  WORKFLOW_STORAGE_ROOT              Override storage directory (default: ~/.uncaged/workflow)",
+  );
+  lines.push(
+    "  UNCAGED_WORKFLOW_STORAGE_ROOT      Internal override (takes priority over WORKFLOW_STORAGE_ROOT)",
+  );
+  return lines.join("\n");
+}
+
+function printDeprecation(oldCmd: string, newCmd: string): void {
+  printCliWarn(`⚠ "${oldCmd}" is deprecated, use "${newCmd}" instead`);
+}
+
+// ── Group dispatchers ──────────────────────────────────────────────────
+
+function dispatchGroup(
+  tableName: string,
+  table: Record<string, CommandEntry>,
+  storageRoot: string,
+  argv: string[],
+): Promise<number> | null {
+  const sub = argv[0];
+  if (sub === undefined || sub === "--help" || sub === "-h") {
+    const entries = Object.entries(table);
+    const lines = [`${tableName} subcommands:\n`];
+    for (const [name, e] of entries) {
+      const args = e.args ? ` ${e.args}` : "";
+      lines.push(`  uncaged-workflow ${tableName} ${name}${args}`);
+      lines.push(`      ${e.description}\n`);
+    }
+    printCliLine(lines.join("\n"));
+    return Promise.resolve(sub === undefined ? 1 : 0);
+  }
+  const entry = table[sub];
+  if (entry === undefined) {
+    return null;
+  }
+  return entry.handler(storageRoot, argv.slice(1));
+}
+
+async function dispatchInit(storageRoot: string, argv: string[]): Promise<number> {
+  const result = dispatchGroup("init", INIT_SUBCOMMAND_TABLE, storageRoot, argv);
+  if (result !== null) {
+    return result;
+  }
+  const sub = argv[0];
+  printCliError(`${formatCliUsage()}\n\nerror: unknown init subcommand: ${sub}`);
+  return 1;
+}
+
+async function dispatchWorkflow(storageRoot: string, argv: string[]): Promise<number> {
+  const result = dispatchGroup("workflow", WORKFLOW_SUBCOMMAND_TABLE, storageRoot, argv);
+  if (result !== null) {
+    return result;
+  }
+  const sub = argv[0];
+  if (sub === "remove") {
+    printDeprecation("workflow remove", "workflow rm");
+    return dispatchRemove(storageRoot, argv.slice(1));
+  }
+  printCliError(`${formatCliUsage()}\n\nerror: unknown workflow subcommand: ${sub}`);
+  return 1;
+}
+
+async function dispatchThread(storageRoot: string, argv: string[]): Promise<number> {
+  const result = dispatchGroup("thread", THREAD_SUBCOMMAND_TABLE, storageRoot, argv);
+  if (result !== null) {
+    return result;
+  }
+  const sub = argv[0];
+  printCliError(`${formatCliUsage()}\n\nerror: unknown thread subcommand: ${sub}`);
+  return 1;
+}
+
+async function dispatchCas(storageRoot: string, argv: string[]): Promise<number> {
+  const result = dispatchGroup("cas", CAS_SUBCOMMAND_TABLE, storageRoot, argv);
+  if (result !== null) {
+    return result;
+  }
+  const sub = argv[0];
+  printCliError(`${formatCliUsage()}\n\nerror: unknown cas subcommand: ${sub}`);
+  return 1;
+}
+
+// ── Help ────────────────────────────────────────────────────────────────
+
+async function dispatchSkill(_storageRoot: string, argv: string[]): Promise<number> {
+  const topic = argv[0];
+  if (topic === undefined) {
+    printCliLine(formatSkillIndex());
+    return 0;
+  }
+  const doc = formatSkillTopic(topic);
+  if (doc === null) {
+    printCliError(`unknown skill topic: ${topic}\n\n${formatSkillIndex()}`);
+    return 1;
+  }
+  printCliLine(doc);
+  return 0;
+}
+
+async function dispatchHelp(_storageRoot: string, argv: string[]): Promise<number> {
+  // Legacy compat: help --skill [topic] → skill [topic]
+  const skillIdx = argv.indexOf("--skill");
+  if (skillIdx !== -1) {
+    const topic = argv[skillIdx + 1];
+    if (topic === undefined) {
+      printCliLine(formatSkillIndex());
+      return 0;
+    }
+    const doc = formatSkillTopic(topic);
+    if (doc === null) {
+      printCliError(`unknown skill topic: ${topic}\n\n${formatSkillIndex()}`);
+      return 1;
+    }
+    printCliLine(doc);
+    return 0;
+  }
+  printCliLine(formatCliUsage());
+  return 0;
+}
+
+// ── Top-level command table (Phase 3) ──────────────────────────────────
 
 const COMMAND_TABLE: Record<string, DispatchFn> = {
-  add: dispatchAdd,
-  list: dispatchList,
-  show: dispatchShow,
-  remove: dispatchRemove,
-  run: dispatchRun,
-  ps: dispatchPs,
-  kill: dispatchKill,
-  history: dispatchHistory,
-  rollback: dispatchRollback,
-  pause: dispatchPause,
-  resume: dispatchResume,
-  threads: dispatchThreads,
-  thread: dispatchThreadBranch,
-  fork: dispatchFork,
+  // Grouped commands (primary)
+  workflow: dispatchWorkflow,
+  thread: dispatchThread,
   cas: dispatchCas,
+  init: dispatchInit,
+  help: dispatchHelp,
+  skill: dispatchSkill,
+
+  // Top-level shortcuts (no deprecation)
+  run: dispatchRun,
+  live: dispatchLive,
+};
+
+// Deprecated flat commands that delegate to grouped commands
+const DEPRECATED_ALIASES: Record<string, { newCmd: string; handler: DispatchFn }> = {
+  add: { newCmd: "workflow add", handler: dispatchAdd },
+  list: { newCmd: "workflow list", handler: dispatchList },
+  show: { newCmd: "workflow show", handler: dispatchShow },
+  remove: { newCmd: "workflow rm", handler: dispatchRemove },
+  ps: { newCmd: "thread ps", handler: dispatchPs },
+  kill: { newCmd: "thread kill", handler: dispatchKill },
+  pause: { newCmd: "thread pause", handler: dispatchPause },
+  resume: { newCmd: "thread resume", handler: dispatchResume },
+  threads: { newCmd: "thread list", handler: dispatchThreadList },
+  fork: { newCmd: "thread fork", handler: dispatchFork },
+  gc: { newCmd: "cas gc", handler: dispatchGc },
+  history: { newCmd: "workflow history", handler: dispatchHistory },
+  rollback: { newCmd: "workflow rollback", handler: dispatchRollback },
 };
 
 export async function runCli(storageRoot: string, argv: string[]): Promise<number> {
   if (argv.length === 0) {
-    printCliError(usage());
+    printCliLine(formatCliUsage());
     return 1;
   }
   const command = argv[0];
   if (command === undefined) {
-    printCliError(usage());
+    printCliLine(formatCliUsage());
     return 1;
   }
   const rest = argv.slice(1);
+
   const dispatch = COMMAND_TABLE[command];
-  if (dispatch === undefined) {
-    printCliError(`${usage()}\n\nerror: unknown command ${command}`);
-    return 1;
+  if (dispatch !== undefined) {
+    return dispatch(storageRoot, rest);
   }
-  return dispatch(storageRoot, rest);
+
+  const deprecated = DEPRECATED_ALIASES[command];
+  if (deprecated !== undefined) {
+    printDeprecation(command, deprecated.newCmd);
+    return deprecated.handler(storageRoot, rest);
+  }
+
+  printCliError(`${formatCliUsage()}\n\nerror: unknown command ${command}`);
+  return 1;
 }

packages/cli-workflow/src/cmd-add.ts

@@ -192,7 +192,7 @@ export async function cmdAdd(
     return validated;
   }
 
-  const extracted = await extractBundleExports(resolvedPath);
+  const extracted = await extractBundleExports(resolvedPath, { storageRoot });
   if (!extracted.ok) {
     return extracted;
   }

packages/cli-workflow/src/cmd-cas.ts

@@ -1,23 +1,11 @@
-import { dirname, join } from "node:path";
-
-import { createThreadCas, err, ok, type Result } from "@uncaged/workflow";
-
-import { resolveThreadDataPath } from "./thread-scan.js";
-
-function resolveCasDir(threadDataPath: string, threadId: string): string {
-  return join(dirname(threadDataPath), `${threadId}.cas`);
-}
+import { createCasStore, err, getGlobalCasDir, ok, type Result } from "@uncaged/workflow";
 
 export async function cmdCasGet(
   storageRoot: string,
-  threadId: string,
+  _threadId: string,
   hash: string,
 ): Promise<Result<string, string>> {
-  const dataPath = await resolveThreadDataPath(storageRoot, threadId);
-  if (dataPath === null) {
-    return err(`thread not found: ${threadId}`);
-  }
-  const cas = createThreadCas(resolveCasDir(dataPath, threadId));
+  const cas = createCasStore(getGlobalCasDir(storageRoot));
   const content = await cas.get(hash);
   if (content === null) {
     return err(`cas entry not found: ${hash}`);
@@ -27,41 +15,29 @@ export async function cmdCasGet(
 
 export async function cmdCasPut(
   storageRoot: string,
-  threadId: string,
+  _threadId: string,
   content: string,
 ): Promise<Result<string, string>> {
-  const dataPath = await resolveThreadDataPath(storageRoot, threadId);
-  if (dataPath === null) {
-    return err(`thread not found: ${threadId}`);
-  }
-  const cas = createThreadCas(resolveCasDir(dataPath, threadId));
+  const cas = createCasStore(getGlobalCasDir(storageRoot));
   const hash = await cas.put(content);
   return ok(hash);
 }
 
 export async function cmdCasList(
   storageRoot: string,
-  threadId: string,
+  _threadId: string,
 ): Promise<Result<string[], string>> {
-  const dataPath = await resolveThreadDataPath(storageRoot, threadId);
-  if (dataPath === null) {
-    return err(`thread not found: ${threadId}`);
-  }
-  const cas = createThreadCas(resolveCasDir(dataPath, threadId));
+  const cas = createCasStore(getGlobalCasDir(storageRoot));
   const hashes = await cas.list();
   return ok(hashes);
 }
 
 export async function cmdCasRm(
   storageRoot: string,
-  threadId: string,
+  _threadId: string,
   hash: string,
 ): Promise<Result<void, string>> {
-  const dataPath = await resolveThreadDataPath(storageRoot, threadId);
-  if (dataPath === null) {
-    return err(`thread not found: ${threadId}`);
-  }
-  const cas = createThreadCas(resolveCasDir(dataPath, threadId));
+  const cas = createCasStore(getGlobalCasDir(storageRoot));
   await cas.delete(hash);
   return ok(undefined);
 }

packages/cli-workflow/src/cmd-fork.ts

@@ -65,8 +65,9 @@ export async function cmdFork(
   const newThreadId = generateUlid(Date.now());
   const stepsOnWire = plan.value.historicalSteps.map((s) => ({
     role: s.role,
-    content: s.content,
+    contentHash: s.contentHash,
     meta: s.meta,
+    refs: s.refs,
     timestamp: s.timestamp,
   }));
 

packages/cli-workflow/src/cmd-gc.ts

@@ -0,0 +1,5 @@
+import { type GcResult, garbageCollectCas, type Result } from "@uncaged/workflow";
+
+export async function cmdGc(storageRoot: string): Promise<Result<GcResult, string>> {
+  return garbageCollectCas(storageRoot);
+}

packages/cli-workflow/src/cmd-help.ts

@@ -0,0 +1,238 @@
+import { getCommandRegistry } from "./cli-dispatch.js";
+
+type SkillTopic = {
+  name: string;
+  description: string;
+  format: () => string;
+};
+
+const SKILL_TOPICS: ReadonlyArray<SkillTopic> = [
+  { name: "cli", description: "Full CLI command reference", format: formatSkillCli },
+  {
+    name: "develop",
+    description: "Guide for agents executing roles inside a workflow",
+    format: formatSkillDevelop,
+  },
+  {
+    name: "author",
+    description: "Guide for building and publishing workflow bundles",
+    format: formatSkillAuthor,
+  },
+];
+
+export function getSkillTopics(): ReadonlyArray<{ name: string; description: string }> {
+  return SKILL_TOPICS.map((t) => ({ name: t.name, description: t.description }));
+}
+
+export function formatSkillTopic(topic: string): string | null {
+  const entry = SKILL_TOPICS.find((t) => t.name === topic);
+  if (entry === undefined) {
+    return null;
+  }
+  return entry.format();
+}
+
+export function formatSkillIndex(): string {
+  const rows = SKILL_TOPICS.map((t) => `| \`${t.name}\` | ${t.description} |`);
+  return `# uncaged-workflow help --skill
+
+Available topics:
+
+| Topic | Description |
+|-------|-------------|
+${rows.join("\n")}
+
+Usage: \`uncaged-workflow skill <topic>\`
+`;
+}
+
+// ── cli topic (existing full reference) ────────────────────────────────
+
+function formatSkillCli(): string {
+  const groups = getCommandRegistry();
+
+  const commandSections: string[] = [];
+  for (const group of groups) {
+    const rows = group.commands.map((cmd) => {
+      const args = cmd.args ? `\`${cmd.args}\`` : "(none)";
+      return `| \`${group.name} ${cmd.name}\` | ${args} | ${cmd.description} |`;
+    });
+    commandSections.push(
+      `### ${group.name}\n\n| Command | Args | Description |\n|---------|------|-------------|\n${rows.join("\n")}`,
+    );
+  }
+
+  return `# uncaged-workflow CLI Reference
+
+## Core Concepts
+
+| Concept | Description |
+|---------|-------------|
+| **Workflow** | A single-file ESM bundle (\`.esm.js\`) that exports \`run\` and \`descriptor\`. Identified by name and XXH64 hash. |
+| **Bundle** | The physical \`.esm.js\` file stored in the bundles directory. Immutable once written. |
+| **Thread** | A single execution of a workflow, identified by a ULID. Persists state as JSONL files. |
+| **CAS** | Content-Addressable Storage. Per-thread key-value store keyed by content hash. |
+| **Registry** | \`workflow.yaml\` — maps workflow names to their current and historical bundle hashes. |
+
+## Commands
+
+${commandSections.join("\n\n")}
+
+### Top-level shortcuts
+
+| Command | Equivalent | Description |
+|---------|------------|-------------|
+| \`run\` | \`thread run\` | Shortcut to start a thread |
+| \`live\` | \`thread live\` | Shortcut to attach to a thread |
+
+## Typical Workflow
+
+1. \`uncaged-workflow workflow add my-wf ./my-wf.esm.js\` — register a workflow
+2. \`uncaged-workflow run my-wf --prompt "do the thing"\` — start a thread
+3. \`uncaged-workflow live --latest\` — attach and watch output
+4. \`uncaged-workflow thread show <thread-id>\` — inspect completed thread
+
+## Exit Codes
+
+| Code | Meaning |
+|------|---------|
+| 0 | Success |
+| 1 | Error |
+
+## Environment Variables
+
+| Variable | Description |
+|----------|-------------|
+| \`UNCAGED_WORKFLOW_STORAGE_ROOT\` | Override the default storage directory for all workflow data |
+`;
+}
+
+// ── develop topic (for agents inside a workflow) ───────────────────────
+
+function formatSkillDevelop(): string {
+  return `# Workflow Role Guide
+
+Reference for agents executing roles (planner, coder, reviewer, etc.) inside a running workflow thread.
+
+## Thread ID
+
+Every thread has a 26-character Crockford Base32 ULID (e.g. \`06F03H5V6JTMDST6P3TVH42RWM\`).
+
+It appears in the **first message** of the conversation. If unsure:
+
+\`\`\`
+uncaged-workflow thread list
+\`\`\`
+
+## CAS (Content-Addressable Storage)
+
+Store and retrieve content by hash, scoped to the current thread.
+
+| Operation | Command |
+|-----------|---------|
+| **Store** | \`uncaged-workflow cas put <THREAD_ID> '<content>'\` → prints hash |
+| **Read** | \`uncaged-workflow cas get <THREAD_ID> <HASH>\` → prints content |
+| **List** | \`uncaged-workflow cas list <THREAD_ID>\` |
+
+CAS is the **only** supported way to persist structured data (phase plans, review notes, etc.) within a thread. Do not use temp files.
+
+## Meta Output
+
+Each role must produce structured output that the moderator extracts. The exact schema depends on the role, but the pattern is:
+
+1. Do your work (write code, run tests, etc.)
+2. Output a compact JSON object matching the role's schema
+3. The moderator extracts and validates it automatically
+
+## Thread Context
+
+The conversation history contains outputs from previous roles. Read it to understand:
+- What task was requested (from the initial prompt)
+- What previous roles produced (plans, code changes, review results)
+- What the moderator decided (which phase to work on, whether to retry)
+`;
+}
+
+// ── author topic (for workflow developers) ─────────────────────────────
+
+function formatSkillAuthor(): string {
+  return `# Workflow Authoring Guide
+
+How to build, test, and publish workflow bundles for uncaged-workflow.
+
+## Bundle Structure
+
+A workflow bundle is a single ESM file (\`.esm.js\`) that exports:
+
+\`\`\`typescript
+// Required exports
+export const descriptor: WorkflowDescriptor;
+export const run: WorkflowRun;
+\`\`\`
+
+## WorkflowDescriptor
+
+Defines the workflow's metadata and role sequence:
+
+\`\`\`typescript
+type WorkflowDescriptor = {
+  name: string;           // verb-first kebab-case, e.g. "solve-issue"
+  description: string;    // one-line summary
+  roles: string[];        // ordered role names, e.g. ["planner", "coder", "reviewer"]
+};
+\`\`\`
+
+## WorkflowRun
+
+The main function that creates and returns a moderator:
+
+\`\`\`typescript
+type WorkflowRun = (ctx: WorkflowContext) => Moderator;
+\`\`\`
+
+The **Moderator** controls the flow — it decides which role runs next, handles retries, and determines when the workflow is complete.
+
+## Role Definition
+
+Each role has:
+
+| Field | Type | Purpose |
+|-------|------|---------|
+| \`description\` | string | What the role does |
+| \`systemPrompt\` | string | System prompt for the agent |
+| \`extractPrompt\` | string | Instruction for extracting structured meta |
+| \`schema\` | ZodSchema | Validates the extracted meta |
+| \`extractRefs\` | fn or null | Extracts CAS hashes from meta for DAG linking |
+| \`extractMode\` | "single" | Extraction mode |
+
+## Development Workflow
+
+\`\`\`bash
+# 1. Initialize a workspace
+uncaged-workflow init workspace my-workflow
+
+# 2. Write your template (roles + moderator + descriptor)
+
+# 3. Build the ESM bundle
+bun run build
+
+# 4. Register locally
+uncaged-workflow workflow add my-workflow ./dist/my-workflow.esm.js
+
+# 5. Test
+uncaged-workflow run my-workflow --prompt "test task"
+uncaged-workflow live --latest
+\`\`\`
+
+## Versioning
+
+Bundles are immutable and identified by XXH64 hash. Re-registering a workflow with a new bundle creates a new version. Use \`workflow history\` and \`workflow rollback\` to manage versions.
+`;
+}
+
+// ── Legacy compat ──────────────────────────────────────────────────────
+
+/** @deprecated Use formatSkillTopic("cli") instead */
+export function formatSkillDoc(): string {
+  return formatSkillCli();
+}

packages/cli-workflow/src/cmd-init.ts

@@ -0,0 +1,415 @@
+import { mkdir, readFile, writeFile } from "node:fs/promises";
+import { dirname, join, resolve } from "node:path";
+
+import { err, ok, type Result } from "@uncaged/workflow";
+
+import { pathExists } from "./fs-utils.js";
+
+export type CmdInitWorkspaceSuccess = {
+  rootPath: string;
+};
+
+export type CmdInitTemplateSuccess = {
+  templatePath: string;
+};
+
+function validateWorkspaceSegment(name: string): Result<void, string> {
+  if (name.length === 0) {
+    return err("workspace name must not be empty");
+  }
+  if (name === "." || name === "..") {
+    return err("invalid workspace name");
+  }
+  if (name.includes("/") || name.includes("\\")) {
+    return err("workspace name must not contain path separators");
+  }
+  return ok(undefined);
+}
+
+function rootPackageJson(workspaceName: string): string {
+  return `${JSON.stringify(
+    {
+      name: workspaceName,
+      private: true,
+      type: "module",
+      workspaces: ["templates/*", "workflows"],
+    },
+    null,
+    2,
+  )}\n`;
+}
+
+function workflowsPackageJson(): string {
+  return `${JSON.stringify(
+    {
+      name: "workflows",
+      version: "0.0.0",
+      private: true,
+      type: "module",
+      dependencies: {
+        "@uncaged/workflow": "^0.1.0",
+        zod: "^4.0.0",
+      },
+    },
+    null,
+    2,
+  )}\n`;
+}
+
+function biomeJson(): string {
+  return `${JSON.stringify(
+    {
+      $schema: "https://biomejs.dev/schemas/2.4.14/schema.json",
+      files: {
+        includes: ["**", "!**/node_modules", "!**/dist"],
+      },
+      formatter: {
+        indentWidth: 2,
+      },
+      linter: {
+        enabled: true,
+        rules: {
+          recommended: true,
+        },
+      },
+    },
+    null,
+    2,
+  )}\n`;
+}
+
+function tsconfigJson(): string {
+  return `${JSON.stringify(
+    {
+      compilerOptions: {
+        strict: true,
+        target: "ESNext",
+        module: "ESNext",
+        moduleResolution: "Bundler",
+        skipLibCheck: true,
+      },
+    },
+    null,
+    2,
+  )}\n`;
+}
+
+function agentsMd(): string {
+  return `# AGENTS — Workflow 工作区开发指南
+
+面向在本仓库中编写 workflow 的 coding agent。引擎层术语与架构细节与 **@uncaged/workflow** 上游文档一致，编写时可对照 \`CLAUDE.md\` 与 \`docs/architecture.md\`。
+
+## 1. 项目结构（workspace / template / workflow instance）
+
+| 层级 | 目录 / 产物 | 职责 |
+|------|----------------|------|
+| **Workspace** | 仓库根（\`package.json\` 含 \`workspaces: ["templates/*", "workflows"]\`） | Bun monorepo：统一管理本地模板包与 workflow 实例 |
+| **Template** | \`templates/<name>/\`（如 \`src/roles.ts\`、\`src/moderator.ts\`、\`src/index.ts\`） | 纯数据：**WorkflowDefinition**（各 **RoleDefinition** + **Moderator**），**不绑定**具体 Agent |
+| **Workflow instance** | \`workflows/\`（或单独包） | 把模板与运行时 **AgentFn** / **ExtractFn** 组合，产出可注册的 **单文件 ESM bundle**（\`run\` + \`descriptor\` 命名导出） |
+
+Init 生成的骨架：\`templates/\` 下放可复用定义，\`workflows/\` 下放绑定与打包入口。
+
+## 2. 核心概念
+
+- **RoleMeta**：\`Record<string, Record<string, unknown>>\`，角色名 → 该角色结构化 meta 的形状约定。
+- **RoleDefinition<Meta>**：纯数据——\`description\`、\`systemPrompt\`、\`extractPrompt\`、\`schema\`（Zod v4）。不含执行逻辑。
+- **WorkflowDefinition<M extends RoleMeta>**：\`description\` + \`roles\`（各角色定义）+ **Moderator**。
+- **Moderator**：\`(ctx: ModeratorContext<M>) => (角色名) | END\`。同步、纯函数，只做路由。
+- **AgentFn**：\`(ctx: AgentContext) => Promise<string>\`，原始文本输出；从上下文读取当前角色的 \`systemPrompt\`。
+- **ExtractFn**：从上下文与 prompt 解析结构化数据（引擎与 Agent 都可使用）。
+
+引擎循环简述：**Moderator** → 选角色 → **Agent** 产出文本 → **Extract** 写入 **meta** → 追加 step，重复直至 **END**。详见 \`docs/architecture.md\` 中的三阶段说明。
+
+## 3. 开发流程
+
+1. **定义 RoleMeta**：为每个角色约定 meta 的 TypeScript 类型（与 Zod schema 对齐）。
+2. **编写 RoleDefinition**：为每个角色写 Zod \`schema\`，补齐 \`systemPrompt\` / \`extractPrompt\` / \`description\`。
+3. **编写 Moderator**：根据 \`ctx.steps\` 与业务状态返回下一个角色名或 \`END\`。
+4. **组装 WorkflowDefinition**：在模板 \`index\` 中导出 definition（以及必要的角色 / moderator 导出）。
+5. **实例化**：在 workflow 包中使用 \`createWorkflow(def, binding, extract)\`（或项目约定的封装）绑定 **AgentFn** / **ExtractFn**。
+6. **构建**：打包为单个 **.esm.js** bundle，使用 **uncaged-workflow add** 注册。
+
+## 4. 编码规范
+
+与 **CLAUDE.md** 对齐，摘要如下：
+
+- **Functional-first**：优先 \`function\` + \`type\`，避免面向对象业务模型。
+- **type 而非 interface**：类型别名一律用 \`type\`，不要使用 \`interface\`。
+- **显式可空**：不要用 \`?:\`；可空字段写成 \`T | null\`。
+- **function 而非 class**：不用 class（第三方库要求或 \`Error\` 子类除外）。
+- **Crockford Base32**：日志 tag、bundle hash、thread id 等标识约定（引擎侧）；工作区内自定义日志若沿用引擎 logger，tag 为 8 字符 Crockford Base32，且每个调用点唯一。
+- **Named exports only**：不要使用 **default export**；workflow bundle 须 **export const run** 与 **export const descriptor**。
+- **No console.log**：库代码用结构化 logger；CLI 用户输出可按项目 Biome 规则例外标注。
+- **No dynamic import**：业务与 bundle 内禁止 \`import()\`；例外仅限「运行时路径由用户提供的 bundle 加载器」（引擎内部）。
+
+## 5. Template 复用
+
+- **已发布模板**：可通过 npm 依赖 \`@uncaged/workflow-template-*\` 等包，在 workflow 实例中 import 其 **WorkflowDefinition** 再绑定 Agent。
+- **本地模板**：放在本仓库 \`templates/<name>/\`，由 workspace 协议引用（如 \`"template-foo": "workspace:*"\` 或相对路径），便于同源修改与版本控制。
+
+选择模板时保持 **definition 与 agent 绑定分离**：模板只描述「做什么、顺序如何」，实例决定「谁执行、如何抽取 meta」。
+
+## 6. Build and Test
+
+日常命令：
+
+\`\`\`sh
+bun install
+bun run check    # Biome：lint + format
+bun test
+bun build        # 若包内配置了 build 脚本则用于产出 dist / bundle
+uncaged-workflow add <name> <path/to/bundle.esm.js>
+\`\`\`
+
+提交前至少运行 **bun run check** 与 **bun test**；registry 与本地运行流参见 README 与 CLI 文档。
+
+## 7. 常见陷阱
+
+- **No dynamic import**：bundle 须静态可分析；动态 \`import()\` 会破坏哈希与加载约束。
+- **No default export**：引擎只接受命名导出 \`run\` / \`descriptor\`。
+- **No console.log**：避免在可被 Biome \`noConsole\` 规则覆盖的代码路径直接使用 console。
+- **Single-file ESM bundle**：交付物是单一 \`.esm.js\`；静态 import 仅限 Node 内置（见 architecture 文档中的 Bundle Contract）。
+
+---
+
+编写新 workflow 时，先对齐 **RoleMeta → RoleDefinition（Zod）→ Moderator → 绑定 → 单文件 bundle**，再对照本节规范自检。
+`;
+}
+
+function readmeMd(workspaceName: string): string {
+  return `# ${workspaceName}
+
+Local workflow development workspace (Bun monorepo).
+
+## Layout
+
+- \`templates/\` — reusable workflow definition packages (roles + moderator), no agent binding
+- \`workflows/\` — workflow instances that bind templates to agents and export \`run\` + \`descriptor\`
+
+## Commands
+
+\`\`\`sh
+bun install
+bun run check   # after you add scripts / Biome
+uncaged-workflow add <name> <bundle.esm.js>
+uncaged-workflow run <name>
+\`\`\`
+
+Create this skeleton with:
+
+\`\`\`sh
+uncaged-workflow init workspace ${workspaceName}
+\`\`\`
+`;
+}
+
+export async function cmdInitWorkspace(
+  parentDir: string,
+  workspaceName: string,
+): Promise<Result<CmdInitWorkspaceSuccess, string>> {
+  const validated = validateWorkspaceSegment(workspaceName);
+  if (!validated.ok) {
+    return validated;
+  }
+
+  const rootPath = join(parentDir, workspaceName);
+  if (await pathExists(rootPath)) {
+    return err(`directory already exists: ${rootPath}`);
+  }
+
+  await mkdir(rootPath, { recursive: false });
+  await mkdir(join(rootPath, "templates"), { recursive: false });
+  await mkdir(join(rootPath, "workflows"), { recursive: false });
+
+  await Promise.all([
+    writeFile(join(rootPath, "package.json"), rootPackageJson(workspaceName), "utf8"),
+    writeFile(join(rootPath, "biome.json"), biomeJson(), "utf8"),
+    writeFile(join(rootPath, "tsconfig.json"), tsconfigJson(), "utf8"),
+    writeFile(join(rootPath, "AGENTS.md"), agentsMd(), "utf8"),
+    writeFile(join(rootPath, "README.md"), readmeMd(workspaceName), "utf8"),
+    writeFile(join(rootPath, "templates", ".gitkeep"), "", "utf8"),
+    writeFile(join(rootPath, "workflows", "package.json"), workflowsPackageJson(), "utf8"),
+  ]);
+
+  return ok({ rootPath });
+}
+
+function hasTemplatesWorkspaceGlob(workspaces: unknown): boolean {
+  return Array.isArray(workspaces) && workspaces.includes("templates/*");
+}
+
+async function readPackageJsonWorkspaces(dir: string): Promise<unknown | null> {
+  const pkgPath = join(dir, "package.json");
+  if (!(await pathExists(pkgPath))) {
+    return null;
+  }
+  let raw: string;
+  try {
+    raw = await readFile(pkgPath, "utf8");
+  } catch {
+    return null;
+  }
+  let parsed: unknown;
+  try {
+    parsed = JSON.parse(raw) as unknown;
+  } catch {
+    return null;
+  }
+  if (typeof parsed !== "object" || parsed === null || !("workspaces" in parsed)) {
+    return null;
+  }
+  return (parsed as { workspaces: unknown }).workspaces;
+}
+
+/** Resolve uncaged-workflow workspace root (package.json with `templates/*` in `workspaces`). */
+async function findWorkflowWorkspaceRoot(startDir: string): Promise<Result<string, string>> {
+  let dir = resolve(startDir);
+  for (;;) {
+    const workspaces = await readPackageJsonWorkspaces(dir);
+    if (workspaces !== null && hasTemplatesWorkspaceGlob(workspaces)) {
+      return ok(dir);
+    }
+    const parent = dirname(dir);
+    if (parent === dir) {
+      return err(
+        'not inside a workflow workspace (no package.json with workspaces containing "templates/*")',
+      );
+    }
+    dir = parent;
+  }
+}
+
+function templatePackageJson(templateName: string): string {
+  return `${JSON.stringify(
+    {
+      name: `template-${templateName}`,
+      version: "0.0.0",
+      private: true,
+      type: "module",
+      dependencies: {
+        "@uncaged/workflow": "^0.1.0",
+        zod: "^4.0.0",
+      },
+    },
+    null,
+    2,
+  )}\n`;
+}
+
+function templateTsconfigJson(): string {
+  return `${JSON.stringify(
+    {
+      extends: "../../tsconfig.json",
+      compilerOptions: {
+        rootDir: "src",
+        outDir: "dist",
+      },
+      include: ["src/**/*.ts"],
+    },
+    null,
+    2,
+  )}\n`;
+}
+
+function templateRolesTs(): string {
+  return `import type { RoleDefinition } from "@uncaged/workflow";
+import * as z from "zod/v4";
+
+export const HELLO_TEMPLATE_DESCRIPTION =
+  "Minimal starter template: one greeter role, then END.";
+
+export type HelloTemplateMeta = {
+  greeter: {
+    message: string;
+  };
+};
+
+const greeterMetaSchema = z.object({
+  message: z.string(),
+});
+
+export const greeterRole: RoleDefinition<HelloTemplateMeta["greeter"]> = {
+  description: "Says hello — replace with your first role.",
+  systemPrompt: "You are a helpful assistant. Reply with one short friendly sentence.",
+  extractPrompt: "Extract the assistant's greeting as message.",
+  schema: greeterMetaSchema,
+  extractRefs: null,
+};
+`;
+}
+
+function templateModeratorTs(): string {
+  return `import { END, type Moderator, type ModeratorContext } from "@uncaged/workflow";
+
+import type { HelloTemplateMeta } from "./roles.js";
+
+export const helloTemplateModerator: Moderator<HelloTemplateMeta> = (
+  ctx: ModeratorContext<HelloTemplateMeta>,
+) => {
+  if (ctx.steps.length === 0) {
+    return "greeter";
+  }
+  return END;
+};
+`;
+}
+
+function templateIndexTs(): string {
+  return `import type { WorkflowDefinition } from "@uncaged/workflow";
+
+import { helloTemplateModerator } from "./moderator.js";
+import {
+  HELLO_TEMPLATE_DESCRIPTION,
+  type HelloTemplateMeta,
+  greeterRole,
+} from "./roles.js";
+
+export {
+  HELLO_TEMPLATE_DESCRIPTION,
+  type HelloTemplateMeta,
+  greeterRole,
+} from "./roles.js";
+export { helloTemplateModerator } from "./moderator.js";
+
+export const helloTemplateWorkflowDefinition: WorkflowDefinition<HelloTemplateMeta> = {
+  description: HELLO_TEMPLATE_DESCRIPTION,
+  roles: {
+    greeter: greeterRole,
+  },
+  moderator: helloTemplateModerator,
+};
+`;
+}
+
+export async function cmdInitTemplate(
+  startDir: string,
+  templateName: string,
+): Promise<Result<CmdInitTemplateSuccess, string>> {
+  const validated = validateWorkspaceSegment(templateName);
+  if (!validated.ok) {
+    return validated;
+  }
+
+  const rootResult = await findWorkflowWorkspaceRoot(startDir);
+  if (!rootResult.ok) {
+    return rootResult;
+  }
+
+  const workspaceRoot = rootResult.value;
+  const templateDir = join(workspaceRoot, "templates", templateName);
+  if (await pathExists(templateDir)) {
+    return err(`template already exists: ${templateDir}`);
+  }
+
+  await mkdir(join(templateDir, "src"), { recursive: true });
+
+  await Promise.all([
+    writeFile(join(templateDir, "package.json"), templatePackageJson(templateName), "utf8"),
+    writeFile(join(templateDir, "tsconfig.json"), templateTsconfigJson(), "utf8"),
+    writeFile(join(templateDir, "src", "roles.ts"), templateRolesTs(), "utf8"),
+    writeFile(join(templateDir, "src", "moderator.ts"), templateModeratorTs(), "utf8"),
+    writeFile(join(templateDir, "src", "index.ts"), templateIndexTs(), "utf8"),
+  ]);
+
+  return ok({ templatePath: templateDir });
+}

packages/cli-workflow/src/cmd-live.ts

@@ -0,0 +1,463 @@
+import { watch } from "node:fs";
+import { readFile } from "node:fs/promises";
+import { dirname, join } from "node:path";
+
+import {
+  type CasStore,
+  createCasStore,
+  getContentMerklePayload,
+  getGlobalCasDir,
+  tryParseRoleStepRecord,
+  tryParseWorkflowResultRecord,
+  type WorkflowCompletion,
+} from "@uncaged/workflow";
+
+import { printCliError, printCliLine } from "./cli-output.js";
+import { pathExists } from "./fs-utils.js";
+import type { ParsedLiveArgv } from "./live-argv.js";
+import { findLatestThreadDataPath, resolveThreadDataPath } from "./thread-scan.js";
+
+export const LIVE_CONTENT_MAX_LINES = 10;
+
+export type LiveRoleRow = {
+  role: string;
+  content: string;
+  meta: Record<string, unknown>;
+  timestamp: number;
+};
+
+export function formatLiveTimeLabel(timestampMs: number): string {
+  const d = new Date(timestampMs);
+  const hh = String(d.getHours()).padStart(2, "0");
+  const mm = String(d.getMinutes()).padStart(2, "0");
+  const ss = String(d.getSeconds()).padStart(2, "0");
+  return `${hh}:${mm}:${ss}`;
+}
+
+function shouldUseColor(): boolean {
+  return process.stdout.isTTY === true && process.env.NO_COLOR === undefined;
+}
+
+function highlightLiveRole(name: string): string {
+  if (!shouldUseColor()) {
+    return name;
+  }
+  return `\x1b[1m\x1b[36m${name}\x1b[0m`;
+}
+
+function dimGreyLine(line: string): string {
+  if (!shouldUseColor()) {
+    return line;
+  }
+  return `\x1b[2m\x1b[90m${line}\x1b[0m`;
+}
+
+export function formatLiveDebugLine(timestampMs: number, tag: string, message: string): string {
+  const label = `[${formatLiveTimeLabel(timestampMs)}] [${tag}] ${message.replace(/\n/g, " ")}`;
+  return dimGreyLine(label);
+}
+
+export function renderLiveRoleStepLines(row: LiveRoleRow, roleDisplay: string): string[] {
+  const header = `[${formatLiveTimeLabel(row.timestamp)}] ▶ ${roleDisplay}`;
+  const lines: string[] = [header];
+  const parts = row.content.split("\n");
+  const shown = parts.slice(0, LIVE_CONTENT_MAX_LINES);
+  for (const ln of shown) {
+    lines.push(`  ${ln}`);
+  }
+  const omitted = parts.length - shown.length;
+  if (omitted > 0) {
+    lines.push(`  … (${omitted} more line${omitted === 1 ? "" : "s"})`);
+  }
+  lines.push(`  meta: ${JSON.stringify(row.meta)}`);
+  return lines;
+}
+
+function printSummary(result: WorkflowCompletion): void {
+  printCliLine(`completed: returnCode=${result.returnCode} — ${result.summary}`);
+}
+
+type LiveSessionState = {
+  sawStart: boolean;
+  completed: boolean;
+  carry: string;
+  contentOffset: number;
+};
+
+type InfoLiveState = {
+  carry: string;
+  contentOffset: number;
+};
+
+function tryParseInfoRecord(obj: Record<string, unknown>): {
+  tag: string;
+  content: string;
+  timestamp: number;
+} | null {
+  const tag = obj.tag;
+  const content = obj.content;
+  const timestamp = obj.timestamp;
+  if (
+    typeof tag !== "string" ||
+    typeof content !== "string" ||
+    typeof timestamp !== "number" ||
+    !Number.isFinite(timestamp)
+  ) {
+    return null;
+  }
+  return { tag, content, timestamp };
+}
+
+async function handleJsonlLine(
+  rawLine: string,
+  state: LiveSessionState,
+  roleFilter: string | null,
+  cas: CasStore,
+): Promise<{ parseError: string | null; workflowResult: WorkflowCompletion | null }> {
+  const trimmed = rawLine.trim();
+  if (trimmed === "") {
+    return { parseError: null, workflowResult: null };
+  }
+
+  let rec: unknown;
+  try {
+    rec = JSON.parse(trimmed) as unknown;
+  } catch {
+    return { parseError: "invalid JSON in thread data file", workflowResult: null };
+  }
+  if (rec === null || typeof rec !== "object") {
+    return { parseError: "invalid record in thread data file", workflowResult: null };
+  }
+  const obj = rec as Record<string, unknown>;
+
+  if (!state.sawStart) {
+    state.sawStart = true;
+    return { parseError: null, workflowResult: null };
+  }
+
+  const wf = tryParseWorkflowResultRecord(obj);
+  if (wf !== null) {
+    state.completed = true;
+    return { parseError: null, workflowResult: wf };
+  }
+
+  const roleRow = tryParseRoleStepRecord(obj);
+  if (roleRow === null) {
+    return {
+      parseError: "unrecognized record in thread data (expected role step or result)",
+      workflowResult: null,
+    };
+  }
+
+  if (roleFilter !== null && roleRow.role !== roleFilter) {
+    return { parseError: null, workflowResult: null };
+  }
+
+  const payload = await getContentMerklePayload(cas, roleRow.contentHash);
+  const content =
+    payload !== null ? payload : `(content not in CAS; contentHash=${roleRow.contentHash})`;
+
+  const row: LiveRoleRow = {
+    role: roleRow.role,
+    content,
+    meta: roleRow.meta,
+    timestamp: roleRow.timestamp,
+  };
+  for (const outLine of renderLiveRoleStepLines(row, highlightLiveRole(row.role))) {
+    printCliLine(outLine);
+  }
+  return { parseError: null, workflowResult: null };
+}
+
+async function pumpNewContent(
+  dataPath: string,
+  state: LiveSessionState,
+  roleFilter: string | null,
+  cas: CasStore,
+): Promise<number | null> {
+  let text: string;
+  try {
+    text = await readFile(dataPath, "utf8");
+  } catch {
+    return null;
+  }
+
+  if (text.length < state.contentOffset) {
+    state.contentOffset = 0;
+    state.carry = "";
+  }
+
+  const chunk = text.slice(state.contentOffset);
+  state.contentOffset = text.length;
+  state.carry += chunk;
+
+  const parts = state.carry.split("\n");
+  state.carry = parts.pop() ?? "";
+
+  for (const line of parts) {
+    const { parseError, workflowResult } = await handleJsonlLine(line, state, roleFilter, cas);
+    if (parseError !== null) {
+      printCliError(parseError);
+      return 1;
+    }
+    if (workflowResult !== null) {
+      printSummary(workflowResult);
+      return 0;
+    }
+  }
+
+  return null;
+}
+
+async function pumpNewInfoContent(infoPath: string, state: InfoLiveState): Promise<void> {
+  let text: string;
+  try {
+    text = await readFile(infoPath, "utf8");
+  } catch {
+    return;
+  }
+
+  if (text.length < state.contentOffset) {
+    state.contentOffset = 0;
+    state.carry = "";
+  }
+
+  const chunk = text.slice(state.contentOffset);
+  state.contentOffset = text.length;
+  state.carry += chunk;
+
+  const parts = state.carry.split("\n");
+  state.carry = parts.pop() ?? "";
+
+  for (const line of parts) {
+    const trimmed = line.trim();
+    if (trimmed === "") {
+      continue;
+    }
+    let rec: unknown;
+    try {
+      rec = JSON.parse(trimmed) as unknown;
+    } catch {
+      continue;
+    }
+    if (rec === null || typeof rec !== "object") {
+      continue;
+    }
+    const parsed = tryParseInfoRecord(rec as Record<string, unknown>);
+    if (parsed === null) {
+      continue;
+    }
+    printCliLine(formatLiveDebugLine(parsed.timestamp, parsed.tag, parsed.content));
+  }
+}
+
+type WatchPumpTask = {
+  path: string;
+  pump: () => Promise<number | null>;
+};
+
+async function runWatchPumpStep(
+  settled: () => boolean,
+  pump: () => Promise<number | null>,
+  closeAll: () => void,
+  finish: (code: number) => void,
+): Promise<void> {
+  if (settled()) {
+    return;
+  }
+  try {
+    const code = await pump();
+    if (code !== null) {
+      closeAll();
+      finish(code);
+    }
+  } catch (e) {
+    closeAll();
+    throw e instanceof Error ? e : new Error(String(e));
+  }
+}
+
+function watchLivePaths(params: { tasks: WatchPumpTask[]; signal: AbortSignal }): Promise<number> {
+  const { tasks, signal } = params;
+
+  return new Promise((resolve, reject) => {
+    let settled = false;
+    const finish = (code: number): void => {
+      if (settled) {
+        return;
+      }
+      settled = true;
+      resolve(code);
+    };
+
+    const pumpChains = new Map<string, Promise<void>>();
+    for (const t of tasks) {
+      pumpChains.set(t.path, Promise.resolve());
+    }
+
+    const watchers: ReturnType<typeof watch>[] = [];
+
+    const closeAll = (): void => {
+      for (const w of watchers) {
+        w.close();
+      }
+    };
+
+    function schedulePump(path: string, pump: () => Promise<number | null>): void {
+      const prev = pumpChains.get(path) ?? Promise.resolve();
+      const next = (async () => {
+        await prev;
+        await runWatchPumpStep(() => settled, pump, closeAll, finish);
+      })();
+      pumpChains.set(path, next);
+    }
+
+    for (const { path, pump } of tasks) {
+      const watcher = watch(path, (eventType) => {
+        if (eventType === "rename") {
+          return;
+        }
+        schedulePump(path, pump);
+      });
+      watchers.push(watcher);
+      watcher.on("error", (err: Error) => {
+        closeAll();
+        reject(err);
+      });
+    }
+
+    const onAbort = (): void => {
+      closeAll();
+      finish(0);
+    };
+    signal.addEventListener("abort", onAbort, { once: true });
+
+    for (const { path, pump } of tasks) {
+      schedulePump(path, pump);
+    }
+  });
+}
+
+type LiveThreadTarget = {
+  threadId: string;
+  dataPath: string;
+};
+
+async function resolveLiveThreadTarget(
+  storageRoot: string,
+  parsed: ParsedLiveArgv,
+): Promise<LiveThreadTarget | null> {
+  if (parsed.latest) {
+    const found = await findLatestThreadDataPath(storageRoot);
+    if (found === null) {
+      printCliError("live: no threads found");
+      return null;
+    }
+    return found;
+  }
+
+  const id = parsed.threadId;
+  if (id === null) {
+    printCliError("live: internal error: missing thread id");
+    return null;
+  }
+  const resolved = await resolveThreadDataPath(storageRoot, id);
+  if (resolved === null) {
+    printCliError(`thread not found: ${id}`);
+    return null;
+  }
+  return { threadId: id, dataPath: resolved };
+}
+
+async function buildLiveWatchTasks(params: {
+  dataPath: string;
+  infoPath: string;
+  debug: boolean;
+  dataState: LiveSessionState;
+  infoState: InfoLiveState;
+  roleFilter: string | null;
+  cas: CasStore;
+}): Promise<WatchPumpTask[]> {
+  const { dataPath, infoPath, debug, dataState, infoState, roleFilter, cas } = params;
+  const tasks: WatchPumpTask[] = [
+    {
+      path: dataPath,
+      pump: () => pumpNewContent(dataPath, dataState, roleFilter, cas),
+    },
+  ];
+
+  if (debug && (await pathExists(infoPath))) {
+    tasks.push({
+      path: infoPath,
+      pump: async () => {
+        await pumpNewInfoContent(infoPath, infoState);
+        return null;
+      },
+    });
+  }
+
+  return tasks;
+}
+
+export async function cmdLive(storageRoot: string, parsed: ParsedLiveArgv): Promise<number> {
+  const target = await resolveLiveThreadTarget(storageRoot, parsed);
+  if (target === null) {
+    return 1;
+  }
+
+  const { threadId, dataPath } = target;
+  const roleFilter = parsed.role;
+  const infoPath = join(dirname(dataPath), `${threadId}.info.jsonl`);
+  const cas = createCasStore(getGlobalCasDir(storageRoot));
+
+  const dataState: LiveSessionState = {
+    sawStart: false,
+    completed: false,
+    carry: "",
+    contentOffset: 0,
+  };
+
+  const infoState: InfoLiveState = {
+    carry: "",
+    contentOffset: 0,
+  };
+
+  const controller = new AbortController();
+  const onSigInt = (): void => {
+    controller.abort();
+  };
+  process.on("SIGINT", onSigInt);
+
+  try {
+    const firstData = await pumpNewContent(dataPath, dataState, roleFilter, cas);
+    if (firstData === 1) {
+      return 1;
+    }
+
+    if (parsed.debug && (await pathExists(infoPath))) {
+      await pumpNewInfoContent(infoPath, infoState);
+    }
+
+    if (firstData === 0 || dataState.completed) {
+      return 0;
+    }
+
+    const tasks = await buildLiveWatchTasks({
+      dataPath,
+      infoPath,
+      debug: parsed.debug,
+      dataState,
+      infoState,
+      roleFilter,
+      cas,
+    });
+
+    return await watchLivePaths({ tasks, signal: controller.signal });
+  } catch (e) {
+    const message = e instanceof Error ? e.message : String(e);
+    printCliError(`live: ${message}`);
+    return 1;
+  } finally {
+    process.off("SIGINT", onSigInt);
+  }
+}

packages/cli-workflow/src/cmd-rollback.ts

@@ -44,6 +44,7 @@ export async function cmdRollback(
   }
 
   const nextRegistry = {
+    config: reg.value.config,
     workflows: { ...reg.value.workflows, [name]: rolled.value },
   };
   const written = await writeWorkflowRegistry(storageRoot, nextRegistry);

packages/cli-workflow/src/cmd-run.ts

@@ -46,7 +46,7 @@ export async function cmdRun(
       threadId,
       workflowName: name,
       prompt,
-      options: { maxRounds },
+      options: { maxRounds, depth: 0 },
     },
     { awaitResponseLine: false },
   );

packages/cli-workflow/src/cmd-thread.ts

@@ -1,7 +1,7 @@
-import { rm, unlink } from "node:fs/promises";
+import { unlink } from "node:fs/promises";
 import { dirname, join } from "node:path";
 
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, garbageCollectCas, ok, type Result } from "@uncaged/workflow";
 
 import { readTextFileIfExists } from "./fs-utils.js";
 import { resolveThreadDataPath } from "./thread-scan.js";
@@ -33,12 +33,12 @@ export async function cmdThreadRemove(
   const dir = dirname(dataPath);
   const infoPath = join(dir, `${threadId}.info.jsonl`);
   const runningPath = join(dir, `${threadId}.running`);
-  const casPath = join(dir, `${threadId}.cas`);
 
   await unlink(dataPath);
   await unlink(infoPath).catch(() => {});
   await unlink(runningPath).catch(() => {});
-  await rm(casPath, { recursive: true, force: true });
+
+  await garbageCollectCas(storageRoot);
 
   return ok(undefined);
 }

packages/cli-workflow/src/live-argv.ts

@@ -0,0 +1,75 @@
+import { err, ok, type Result } from "@uncaged/workflow";
+
+export type ParsedLiveArgv = {
+  threadId: string | null;
+  latest: boolean;
+  debug: boolean;
+  role: string | null;
+};
+
+type LiveArgvScan = {
+  latest: boolean;
+  debug: boolean;
+  role: string | null;
+  threadId: string | null;
+};
+
+function applyLiveArgvToken(argv: string[], i: number, s: LiveArgvScan): Result<number, string> {
+  const a = argv[i];
+  if (a === "--latest") {
+    s.latest = true;
+    return ok(i + 1);
+  }
+  if (a === "--debug") {
+    s.debug = true;
+    return ok(i + 1);
+  }
+  if (a === "--role") {
+    const v = argv[i + 1];
+    if (v === undefined || v.startsWith("--")) {
+      return err("missing value for --role");
+    }
+    s.role = v;
+    return ok(i + 2);
+  }
+  if (a.startsWith("--")) {
+    return err(`unknown live flag: ${a}`);
+  }
+  if (s.threadId !== null) {
+    return err("unexpected extra argument");
+  }
+  s.threadId = a;
+  return ok(i + 1);
+}
+
+export function parseLiveArgv(argv: string[]): Result<ParsedLiveArgv, string> {
+  const s: LiveArgvScan = {
+    latest: false,
+    debug: false,
+    role: null,
+    threadId: null,
+  };
+
+  let i = 0;
+  while (i < argv.length) {
+    const step = applyLiveArgvToken(argv, i, s);
+    if (!step.ok) {
+      return step;
+    }
+    i = step.value;
+  }
+
+  if (s.latest && s.threadId !== null) {
+    return err("live --latest does not take <thread-id>");
+  }
+  if (!s.latest && s.threadId === null) {
+    return err("live requires <thread-id> or --latest");
+  }
+
+  return ok({
+    threadId: s.threadId,
+    latest: s.latest,
+    debug: s.debug,
+    role: s.role,
+  });
+}

packages/cli-workflow/src/storage-env.ts

@@ -1,10 +1,21 @@
 import { getDefaultWorkflowStorageRoot } from "@uncaged/workflow";
 
-/** Resolve storage root, honoring `UNCAGED_WORKFLOW_STORAGE_ROOT` for tests/tools. */
+/**
+ * Resolve storage root with env var override support.
+ *
+ * Priority (highest first):
+ *  1. `UNCAGED_WORKFLOW_STORAGE_ROOT` — internal/test override
+ *  2. `WORKFLOW_STORAGE_ROOT` — user-facing override
+ *  3. Default (`~/.uncaged/workflow`)
+ */
 export function resolveWorkflowStorageRoot(): string {
-  const override = process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
-  if (override !== undefined && override !== "") {
-    return override;
+  const internal = process.env.UNCAGED_WORKFLOW_STORAGE_ROOT;
+  if (internal !== undefined && internal !== "") {
+    return internal;
+  }
+  const userOverride = process.env.WORKFLOW_STORAGE_ROOT;
+  if (userOverride !== undefined && userOverride !== "") {
+    return userOverride;
   }
   return getDefaultWorkflowStorageRoot();
 }

packages/cli-workflow/src/thread-scan.ts

@@ -1,4 +1,4 @@
-import { readdir } from "node:fs/promises";
+import { readdir, stat } from "node:fs/promises";
 import { join } from "node:path";
 
 import { pathExists, readTextFileIfExists } from "./fs-utils.js";
@@ -15,6 +15,28 @@ export type HistoricalThreadRow = {
   workflowName: string | null;
 };
 
+async function readThreadStartTimestampMs(dataPath: string): Promise<number | null> {
+  const text = await readTextFileIfExists(dataPath);
+  if (text === null) {
+    return null;
+  }
+  const firstLine = text.split("\n")[0];
+  if (firstLine === undefined || firstLine.trim() === "") {
+    return null;
+  }
+  let parsed: unknown;
+  try {
+    parsed = JSON.parse(firstLine) as unknown;
+  } catch {
+    return null;
+  }
+  if (parsed === null || typeof parsed !== "object") {
+    return null;
+  }
+  const ts = (parsed as Record<string, unknown>).timestamp;
+  return typeof ts === "number" && Number.isFinite(ts) ? ts : null;
+}
+
 async function readWorkflowNameFromDataJsonl(dataPath: string): Promise<string | null> {
   const text = await readTextFileIfExists(dataPath);
   if (text === null) {
@@ -124,6 +146,50 @@ export async function listHistoricalThreads(
   return out;
 }
 
+/**
+ * Picks the thread whose `.data.jsonl` is newest by start-record `timestamp`,
+ * falling back to file `mtime` when the timestamp is missing.
+ * Tie-breaker: larger `mtime` wins when start timestamps are equal.
+ */
+export async function findLatestThreadDataPath(
+  storageRoot: string,
+): Promise<{ threadId: string; dataPath: string } | null> {
+  const threads = await listHistoricalThreads(storageRoot, null);
+  if (threads.length === 0) {
+    return null;
+  }
+
+  let best: {
+    threadId: string;
+    dataPath: string;
+    primary: number;
+    secondary: number;
+  } | null = null;
+
+  for (const t of threads) {
+    const dataPath = join(storageRoot, "logs", t.hash, `${t.threadId}.data.jsonl`);
+    let mtimeMs = 0;
+    try {
+      const st = await stat(dataPath);
+      mtimeMs = st.mtimeMs;
+    } catch {
+      continue;
+    }
+    const startTs = await readThreadStartTimestampMs(dataPath);
+    const primary = startTs !== null ? startTs : mtimeMs;
+    const secondary = mtimeMs;
+    if (
+      best === null ||
+      primary > best.primary ||
+      (primary === best.primary && secondary > best.secondary)
+    ) {
+      best = { threadId: t.threadId, dataPath, primary, secondary };
+    }
+  }
+
+  return best === null ? null : { threadId: best.threadId, dataPath: best.dataPath };
+}
+
 export async function resolveThreadDataPath(
   storageRoot: string,
   threadId: string,

packages/workflow-agent-cursor/src/index.ts

@@ -54,7 +54,7 @@ export function createCursorAgent(config: CursorAgentConfig): AgentFn {
       "From the thread context, determine the absolute filesystem path where the project/repository is located.",
       extractCtx,
     );
-    const fullPrompt = buildAgentPrompt(ctx);
+    const fullPrompt = await buildAgentPrompt(ctx);
     const args = [
       "-p",
       fullPrompt,

packages/workflow-agent-hermes/src/index.ts

@@ -35,7 +35,7 @@ export function createHermesAgent(config: HermesAgentConfig): AgentFn {
   const timeoutMs = config.timeout;
 
   return async (ctx) => {
-    const fullPrompt = buildAgentPrompt(ctx);
+    const fullPrompt = await buildAgentPrompt(ctx);
     const args = [
       "chat",
       "-q",

packages/workflow-agent-llm/__tests__/create-llm-adapter.test.ts

@@ -1,8 +1,14 @@
 import { describe, expect, test } from "bun:test";
-import { START, type ThreadContext } from "@uncaged/workflow";
+import { mkdtempSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { createCasStore, START, type ThreadContext } from "@uncaged/workflow";
 
 import { createLlmAdapter } from "../src/create-llm-adapter.js";
 
+const casDir = mkdtempSync(join(tmpdir(), "wf-llm-adapter-cas-"));
+const testCas = createCasStore(casDir);
+
 function makeCtx(userContent: string): ThreadContext {
   return {
     start: {
@@ -11,9 +17,11 @@ function makeCtx(userContent: string): ThreadContext {
       meta: { maxRounds: 10 },
       timestamp: 1,
     },
+    depth: 0,
     steps: [],
     threadId: "01TEST000000000000000000TR",
     currentRole: { name: "planner", systemPrompt: "system instructions" },
+    cas: testCas,
   };
 }
 

packages/workflow-role-coder/package.json

@@ -1,15 +0,0 @@
-{
-  "name": "@uncaged/workflow-role-coder",
-  "version": "0.1.0",
-  "type": "module",
-  "main": "src/index.ts",
-  "types": "src/index.ts",
-  "scripts": {
-    "build": "echo 'TODO'",
-    "test": "echo no tests"
-  },
-  "dependencies": {
-    "@uncaged/workflow": "workspace:*",
-    "zod": "^4.0.0"
-  }
-}

packages/workflow-role-coder/src/index.ts

@@ -1 +0,0 @@
-export { type CoderMeta, coderMetaSchema, coderRole } from "./coder.js";

packages/workflow-role-committer/package.json

@@ -1,15 +0,0 @@
-{
-  "name": "@uncaged/workflow-role-committer",
-  "version": "0.1.0",
-  "type": "module",
-  "main": "src/index.ts",
-  "types": "src/index.ts",
-  "scripts": {
-    "build": "echo 'TODO'",
-    "test": "bun test"
-  },
-  "dependencies": {
-    "@uncaged/workflow": "workspace:*",
-    "zod": "^4.0.0"
-  }
-}

packages/workflow-role-committer/src/index.ts

@@ -1 +0,0 @@
-export { type CommitterMeta, committerMetaSchema, committerRole } from "./committer.js";

packages/workflow-role-committer/tsconfig.json

@@ -1,10 +0,0 @@
-{
-  "extends": "../../tsconfig.json",
-  "compilerOptions": {
-    "rootDir": "src",
-    "outDir": "dist",
-    "composite": true
-  },
-  "include": ["src/**/*.ts"],
-  "references": [{ "path": "../workflow" }]
-}

packages/workflow-role-planner/package.json

@@ -1,15 +0,0 @@
-{
-  "name": "@uncaged/workflow-role-planner",
-  "version": "0.1.0",
-  "type": "module",
-  "main": "src/index.ts",
-  "types": "src/index.ts",
-  "scripts": {
-    "build": "echo 'TODO'",
-    "test": "echo no tests"
-  },
-  "dependencies": {
-    "@uncaged/workflow": "workspace:*",
-    "zod": "^4.0.0"
-  }
-}

packages/workflow-role-planner/src/index.ts

@@ -1,6 +0,0 @@
-export {
-  type PlannerMeta,
-  phaseSchema,
-  plannerMetaSchema,
-  plannerRole,
-} from "./planner.js";

packages/workflow-role-planner/tsconfig.json

@@ -1,10 +0,0 @@
-{
-  "extends": "../../tsconfig.json",
-  "compilerOptions": {
-    "rootDir": "src",
-    "outDir": "dist",
-    "composite": true
-  },
-  "include": ["src/**/*.ts"],
-  "references": [{ "path": "../workflow" }]
-}

packages/workflow-role-preparer/package.json

@@ -1,15 +0,0 @@
-{
-  "name": "@uncaged/workflow-role-preparer",
-  "version": "0.1.0",
-  "type": "module",
-  "main": "src/index.ts",
-  "types": "src/index.ts",
-  "scripts": {
-    "build": "echo 'TODO'",
-    "test": "echo no tests"
-  },
-  "dependencies": {
-    "@uncaged/workflow": "workspace:*",
-    "zod": "^4.0.0"
-  }
-}

packages/workflow-role-preparer/tsconfig.json

@@ -1,8 +0,0 @@
-{
-  "extends": "../../tsconfig.json",
-  "compilerOptions": {
-    "rootDir": "src",
-    "outDir": "dist"
-  },
-  "include": ["src"]
-}

packages/workflow-role-reviewer/src/index.ts

@@ -1 +0,0 @@
-export { type ReviewerMeta, reviewerMetaSchema, reviewerRole } from "./reviewer.js";

packages/workflow-role-reviewer/tsconfig.json

@@ -1,10 +0,0 @@
-{
-  "extends": "../../tsconfig.json",
-  "compilerOptions": {
-    "rootDir": "src",
-    "outDir": "dist",
-    "composite": true
-  },
-  "include": ["src/**/*.ts"],
-  "references": [{ "path": "../workflow" }]
-}

packages/workflow-template-develop/__tests__/committer.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect, test } from "bun:test";
 
-import { committerMetaSchema, committerRole } from "../src/committer.js";
+import { committerMetaSchema, committerRole } from "../src/roles/committer.js";
 
 describe("committerRole", () => {
   test("committed sample validates against schema", () => {

packages/workflow-template-develop/__tests__/develop-template.test.ts

@@ -0,0 +1,257 @@
+import { describe, expect, test } from "bun:test";
+import {
+  END,
+  type ModeratorContext,
+  type RoleStep,
+  START,
+  validateWorkflowDescriptor,
+} from "@uncaged/workflow";
+import { buildDevelopDescriptor } from "../src/descriptor.js";
+import { developModerator } from "../src/index.js";
+import type { CommitterMeta, PlannerMeta } from "../src/roles/index.js";
+import type { DevelopMeta } from "../src/roles.js";
+
+const DEFAULT_PHASES: PlannerMeta["phases"] = [
+  {
+    hash: "4KNMR2PX",
+    title: "Do the work",
+  },
+];
+
+function makeStart(maxRounds: number): ModeratorContext<DevelopMeta>["start"] {
+  return {
+    role: START,
+    content: "Implement the feature",
+    meta: { maxRounds },
+    timestamp: 0,
+  };
+}
+
+function makeCtx(
+  maxRounds: number,
+  steps: ModeratorContext<DevelopMeta>["steps"],
+): ModeratorContext<DevelopMeta> {
+  return {
+    threadId: "01TEST000000000000000000TR",
+    depth: 0,
+    start: makeStart(maxRounds),
+    steps,
+  };
+}
+
+function plannerStep(phases: PlannerMeta["phases"] = DEFAULT_PHASES): RoleStep<DevelopMeta> {
+  return {
+    role: "planner",
+    contentHash: "STUBHASHPLANNER001",
+    meta: { phases },
+    refs: phases.map((p) => p.hash),
+    timestamp: 1,
+  };
+}
+
+function coderStep(completedPhase = "4KNMR2PX"): RoleStep<DevelopMeta> {
+  return {
+    role: "coder",
+    contentHash: "STUBHASHCODER00001",
+    meta: { completedPhase, filesChanged: ["a.ts"], summary: "implemented" },
+    refs: [completedPhase],
+    timestamp: 2,
+  };
+}
+
+function reviewerStep(approved: boolean): RoleStep<DevelopMeta> {
+  return {
+    role: "reviewer",
+    contentHash: "STUBHASHREVIEWER01",
+    meta: approved
+      ? { status: "approved" as const }
+      : { status: "rejected" as const, issues: ["needs fix"] },
+    refs: [],
+    timestamp: 3,
+  };
+}
+
+function testerStep(passed: boolean): RoleStep<DevelopMeta> {
+  return {
+    role: "tester",
+    contentHash: "STUBHASHTESTER01",
+    meta: passed
+      ? { status: "passed" as const, details: "all checks passed" }
+      : { status: "failed" as const, details: "lint failed" },
+    refs: [],
+    timestamp: 4,
+  };
+}
+
+function committerStep(meta: CommitterMeta): RoleStep<DevelopMeta> {
+  return {
+    role: "committer",
+    contentHash: "STUBHASHCOMMITTER1",
+    meta,
+    refs: [],
+    timestamp: 5,
+  };
+}
+
+describe("developModerator", () => {
+  test("routes initial → planner → coder → reviewer → tester → committer → END", () => {
+    expect(developModerator(makeCtx(20, []))).toBe("planner");
+    expect(developModerator(makeCtx(20, [plannerStep()]))).toBe("coder");
+    expect(developModerator(makeCtx(20, [plannerStep(), coderStep()]))).toBe("reviewer");
+    expect(developModerator(makeCtx(20, [plannerStep(), coderStep(), reviewerStep(true)]))).toBe(
+      "tester",
+    );
+    expect(
+      developModerator(
+        makeCtx(20, [plannerStep(), coderStep(), reviewerStep(true), testerStep(true)]),
+      ),
+    ).toBe("committer");
+    expect(
+      developModerator(
+        makeCtx(20, [
+          plannerStep(),
+          coderStep(),
+          reviewerStep(true),
+          testerStep(true),
+          committerStep({ status: "committed", branch: "feat/x", commitSha: "abc1234" }),
+        ]),
+      ),
+    ).toBe(END);
+  });
+
+  test("reviewer rejects → coder retry when budget allows", () => {
+    const steps: ModeratorContext<DevelopMeta>["steps"] = [
+      plannerStep(),
+      coderStep(),
+      reviewerStep(false),
+    ];
+    expect(developModerator(makeCtx(20, steps))).toBe("coder");
+  });
+
+  test("reviewer rejects → END when max rounds exhausted", () => {
+    const steps: ModeratorContext<DevelopMeta>["steps"] = [
+      plannerStep(),
+      coderStep(),
+      reviewerStep(false),
+    ];
+    expect(developModerator(makeCtx(4, steps))).toBe(END);
+  });
+
+  test("tester failed → coder retry when budget allows", () => {
+    const steps: ModeratorContext<DevelopMeta>["steps"] = [
+      plannerStep(),
+      coderStep(),
+      reviewerStep(true),
+      testerStep(false),
+    ];
+    expect(developModerator(makeCtx(20, steps))).toBe("coder");
+  });
+
+  test("tester failed → END when max rounds exhausted", () => {
+    const steps: ModeratorContext<DevelopMeta>["steps"] = [
+      plannerStep(),
+      coderStep(),
+      reviewerStep(true),
+      testerStep(false),
+    ];
+    expect(developModerator(makeCtx(5, steps))).toBe(END);
+  });
+
+  test("multiple planner phases → coder until all complete, then reviewer", () => {
+    const phases: PlannerMeta["phases"] = [
+      { hash: "AA000001", title: "first phase" },
+      { hash: "AA000002", title: "second phase" },
+    ];
+    expect(developModerator(makeCtx(20, [plannerStep(phases)]))).toBe("coder");
+    expect(developModerator(makeCtx(20, [plannerStep(phases), coderStep("AA000001")]))).toBe(
+      "coder",
+    );
+    expect(
+      developModerator(
+        makeCtx(20, [plannerStep(phases), coderStep("AA000001"), coderStep("AA000002")]),
+      ),
+    ).toBe("reviewer");
+  });
+
+  test("one-shot coder reports only last phase hash → reviewer (moderator treats as all phases done)", () => {
+    const phases: PlannerMeta["phases"] = [
+      { hash: "BB000001", title: "setup branch" },
+      { hash: "BB000002", title: "write tests" },
+      { hash: "BB000003", title: "verify" },
+      { hash: "BB000004", title: "polish" },
+    ];
+    expect(developModerator(makeCtx(20, [plannerStep(phases), coderStep("BB000004")]))).toBe(
+      "reviewer",
+    );
+  });
+
+  test("unrecognised completedPhase hash → coder retry when budget allows", () => {
+    const phases: PlannerMeta["phases"] = [
+      { hash: "CC000001", title: "first phase" },
+      { hash: "CC000002", title: "second phase" },
+    ];
+    expect(developModerator(makeCtx(20, [plannerStep(phases), coderStep("all-done")]))).toBe(
+      "coder",
+    );
+  });
+
+  test("incomplete phases → END when max rounds exhausted", () => {
+    const phases: PlannerMeta["phases"] = [
+      { hash: "DD000001", title: "first phase" },
+      { hash: "DD000002", title: "second phase" },
+    ];
+    const steps: ModeratorContext<DevelopMeta>["steps"] = [
+      plannerStep(phases),
+      coderStep("DD000001"),
+    ];
+    expect(developModerator(makeCtx(3, steps))).toBe(END);
+  });
+
+  test("committer → END for any committer meta status", () => {
+    const committed = committerStep({ status: "committed", branch: "f", commitSha: "x" });
+    const recoverable = committerStep({
+      status: "recoverable",
+      error: "merge conflict",
+      logRef: null,
+    });
+    const unrecoverable = committerStep({
+      status: "unrecoverable",
+      error: "repo missing",
+      logRef: "log1",
+    });
+    const base: ModeratorContext<DevelopMeta>["steps"] = [
+      plannerStep(),
+      coderStep(),
+      reviewerStep(true),
+      testerStep(true),
+    ];
+    expect(developModerator(makeCtx(20, [...base, committed]))).toBe(END);
+    expect(developModerator(makeCtx(20, [...base, recoverable]))).toBe(END);
+    expect(developModerator(makeCtx(20, [...base, unrecoverable]))).toBe(END);
+  });
+});
+
+describe("buildDevelopDescriptor", () => {
+  test("lists all roles with schemas that validate", () => {
+    const descriptor = buildDevelopDescriptor();
+    const validated = validateWorkflowDescriptor(descriptor);
+    expect(validated.ok).toBe(true);
+    if (!validated.ok) {
+      throw new Error(validated.error);
+    }
+    expect(Object.keys(validated.value.roles).sort()).toEqual([
+      "coder",
+      "committer",
+      "planner",
+      "reviewer",
+      "tester",
+    ]);
+    for (const key of ["planner", "coder", "reviewer", "tester", "committer"] as const) {
+      const role = validated.value.roles[key];
+      expect(role).toBeDefined();
+      expect(typeof role.schema).toBe("object");
+      expect(role.schema).not.toBeNull();
+      expect(Array.isArray(role.schema)).toBe(false);
+    }
+  });
+});

packages/workflow-template-develop/__tests__/reviewer.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect, test } from "bun:test";
 
-import { reviewerMetaSchema, reviewerRole } from "../src/reviewer.js";
+import { reviewerMetaSchema, reviewerRole } from "../src/roles/reviewer.js";
 
 describe("reviewerRole", () => {
   test("approved sample validates against schema", () => {

packages/workflow-template-develop/package.json

@@ -1,5 +1,5 @@
 {
-  "name": "@uncaged/workflow-role-reviewer",
+  "name": "@uncaged/workflow-template-develop",
   "version": "0.1.0",
   "type": "module",
   "main": "src/index.ts",

packages/workflow-template-develop/src/descriptor.ts

@@ -0,0 +1,12 @@
+import { buildDescriptor } from "@uncaged/workflow";
+
+import { developModerator } from "./moderator.js";
+import { DEVELOP_WORKFLOW_DESCRIPTION, developRoles } from "./roles.js";
+
+export function buildDevelopDescriptor() {
+  return buildDescriptor({
+    description: DEVELOP_WORKFLOW_DESCRIPTION,
+    roles: developRoles,
+    moderator: developModerator,
+  });
+}

packages/workflow-template-develop/src/index.ts

@@ -0,0 +1,52 @@
+import {
+  type AgentBinding,
+  createWorkflow,
+  type ExtractFn,
+  type LlmProvider,
+  type WorkflowDefinition,
+  type WorkflowFn,
+} from "@uncaged/workflow";
+
+import { developModerator } from "./moderator.js";
+import { DEVELOP_WORKFLOW_DESCRIPTION, type DevelopMeta, developRoles } from "./roles.js";
+
+export { buildDevelopDescriptor } from "./descriptor.js";
+export { developModerator } from "./moderator.js";
+export {
+  type CoderMeta,
+  type CommitterMeta,
+  coderMetaSchema,
+  coderRole,
+  committerMetaSchema,
+  committerRole,
+  type PlannerMeta,
+  phaseSchema,
+  plannerMetaSchema,
+  plannerRole,
+  type ReviewerMeta,
+  reviewerMetaSchema,
+  reviewerRole,
+  type TesterMeta,
+  testerMetaSchema,
+  testerRole,
+} from "./roles/index.js";
+export {
+  DEVELOP_WORKFLOW_DESCRIPTION,
+  type DevelopMeta,
+  type DevelopRoles,
+  developRoles,
+} from "./roles.js";
+
+export const developWorkflowDefinition: WorkflowDefinition<DevelopMeta> = {
+  description: DEVELOP_WORKFLOW_DESCRIPTION,
+  roles: developRoles,
+  moderator: developModerator,
+};
+
+export function createDevelopRun(
+  binding: AgentBinding,
+  extract: ExtractFn,
+  llmProvider: LlmProvider | null,
+): WorkflowFn {
+  return createWorkflow(developWorkflowDefinition, binding, extract, llmProvider);
+}

packages/workflow-template-develop/src/moderator.ts

@@ -0,0 +1,89 @@
+import type { Moderator, ModeratorContext } from "@uncaged/workflow";
+import { END } from "@uncaged/workflow";
+
+import type { DevelopMeta } from "./roles.js";
+
+function coderFinishedAllPlannedPhases(
+  phases: ReadonlyArray<{ hash: string }>,
+  coderCompletedPhases: ReadonlyArray<string>,
+): boolean {
+  if (phases.length === 0) {
+    return true;
+  }
+  const plannedHashes = new Set(phases.map((p) => p.hash));
+  const lastHash = phases[phases.length - 1].hash;
+  const explicit = new Set(coderCompletedPhases.filter((h) => plannedHashes.has(h)));
+  if (phases.every((p) => explicit.has(p.hash))) {
+    return true;
+  }
+  if (coderCompletedPhases.some((h) => h === lastHash)) {
+    return true;
+  }
+  return false;
+}
+
+function nextAfterCoder(
+  ctx: ModeratorContext<DevelopMeta>,
+  maxRounds: number,
+): (keyof DevelopMeta & string) | typeof END {
+  const plannerStep = ctx.steps.find((s) => s.role === "planner");
+  if (plannerStep === undefined) {
+    return "reviewer";
+  }
+  const phases = plannerStep.meta.phases;
+  const coderCompletedPhases = ctx.steps
+    .filter((s) => s.role === "coder")
+    .map((s) => s.meta.completedPhase);
+  const allDone = coderFinishedAllPlannedPhases(phases, coderCompletedPhases);
+  if (allDone) {
+    return "reviewer";
+  }
+  if (ctx.steps.length < maxRounds - 1) {
+    return "coder";
+  }
+  return END;
+}
+
+export const developModerator: Moderator<DevelopMeta> = (ctx) => {
+  const maxRounds = ctx.start.meta.maxRounds;
+
+  if (ctx.steps.length === 0) {
+    return "planner";
+  }
+
+  const last = ctx.steps[ctx.steps.length - 1];
+
+  if (last.role === "planner") {
+    return "coder";
+  }
+
+  if (last.role === "coder") {
+    return nextAfterCoder(ctx, maxRounds);
+  }
+
+  if (last.role === "reviewer") {
+    if (last.meta.status === "approved") {
+      return "tester";
+    }
+    if (ctx.steps.length < maxRounds - 1) {
+      return "coder";
+    }
+    return END;
+  }
+
+  if (last.role === "tester") {
+    if (last.meta.status === "passed") {
+      return "committer";
+    }
+    if (ctx.steps.length < maxRounds - 1) {
+      return "coder";
+    }
+    return END;
+  }
+
+  if (last.role === "committer") {
+    return END;
+  }
+
+  return END;
+};

packages/workflow-template-develop/src/roles.ts

@@ -0,0 +1,29 @@
+import type { RoleDefinition } from "@uncaged/workflow";
+import { type CoderMeta, coderRole } from "./roles/coder.js";
+import { type CommitterMeta, committerRole } from "./roles/committer.js";
+import { type PlannerMeta, plannerRole } from "./roles/planner.js";
+import { type ReviewerMeta, reviewerRole } from "./roles/reviewer.js";
+import { type TesterMeta, testerRole } from "./roles/tester.js";
+
+export const DEVELOP_WORKFLOW_DESCRIPTION =
+  "Plan phases, implement incrementally, review, verify with tests/build/lint, and commit (planner → coder [repeat per phase] → reviewer → tester → committer).";
+
+export type DevelopMeta = {
+  planner: PlannerMeta;
+  coder: CoderMeta;
+  reviewer: ReviewerMeta;
+  tester: TesterMeta;
+  committer: CommitterMeta;
+};
+
+export type DevelopRoles = {
+  [K in keyof DevelopMeta]: RoleDefinition<DevelopMeta[K]>;
+};
+
+export const developRoles: DevelopRoles = {
+  planner: plannerRole,
+  coder: coderRole,
+  reviewer: reviewerRole,
+  tester: testerRole,
+  committer: committerRole,
+};

packages/workflow-template-develop/src/roles/coder.ts

@@ -11,21 +11,13 @@ export type CoderMeta = z.infer<typeof coderMetaSchema>;
 
 const CODER_SYSTEM = `You are a **coder**. Read the thread for the plan and work on the NEXT incomplete phase only.
 
-## Finding the current thread ID
-
-The thread ID is a 26-character Crockford Base32 string (e.g. \`06F03H5V6JTMDST6P3TVH42RWM\`). It appears in the first message of this conversation. If you are unsure, run:
-
-  uncaged-workflow threads
-
-and use the ID of the active thread.
+Run \`uncaged-workflow skill develop\` for thread ID lookup, CAS commands, and meta output guide.
 
 ## Reading phase details
 
-Each planner phase is identified by a content-hash and a title. To read a phase's full details (name, description, acceptance criteria), run:
+Each planner phase has a content-hash and title. Read full details with \`uncaged-workflow cas get <THREAD_ID> <HASH>\`.
 
-  uncaged-workflow cas get <THREAD_ID> <HASH>
-
-Replace \`<THREAD_ID>\` with the actual thread ID and \`<HASH>\` with the phase hash from the plan.
+The thread ID (26-char Crockford Base32) appears in the first message. If unsure, run \`uncaged-workflow thread list\`.
 
 ## Completing a phase
 
@@ -38,4 +30,6 @@ export const coderRole: RoleDefinition<CoderMeta> = {
   extractPrompt:
     "Extract completedPhase: the planner phase hash finished this round (exact hash string from the plan). If multiple phases were finished in one round, use the last finished phase hash. Extract filesChanged and a summary of the work.",
   schema: coderMetaSchema,
+  extractRefs: (meta) => [meta.completedPhase],
+  extractMode: "single",
 };

packages/workflow-template-develop/src/roles/committer.ts

@@ -21,14 +21,16 @@ export const committerMetaSchema = z.discriminatedUnion("status", [
 
 export type CommitterMeta = z.infer<typeof committerMetaSchema>;
 
-const COMMITTER_SYSTEM = `You are the git committer. Create a branch, commit the changes, and push.
+const COMMITTER_SYSTEM = `You are the git committer. Create a branch and commit the changes.
 Report the branch name and commit SHA. On failure, classify as recoverable or unrecoverable.
 Do not attempt to fix failures yourself.`;
 
 export const committerRole: RoleDefinition<CommitterMeta> = {
-  description: "Creates branch, commits, and pushes when review passes.",
+  description: "Creates a branch and commits changes.",
   systemPrompt: COMMITTER_SYSTEM,
   extractPrompt:
     "Extract the commit result: committed (with branch and SHA), recoverable failure, or unrecoverable failure. Include error details and log references if applicable.",
   schema: committerMetaSchema,
+  extractRefs: null,
+  extractMode: "single",
 };

packages/workflow-template-develop/src/roles/index.ts

@@ -0,0 +1,10 @@
+export { type CoderMeta, coderMetaSchema, coderRole } from "./coder.js";
+export { type CommitterMeta, committerMetaSchema, committerRole } from "./committer.js";
+export {
+  type PlannerMeta,
+  phaseSchema,
+  plannerMetaSchema,
+  plannerRole,
+} from "./planner.js";
+export { type ReviewerMeta, reviewerMetaSchema, reviewerRole } from "./reviewer.js";
+export { type TesterMeta, testerMetaSchema, testerRole } from "./tester.js";

packages/workflow-template-develop/src/roles/planner.ts

@@ -14,27 +14,25 @@ export type PlannerMeta = z.infer<typeof plannerMetaSchema>;
 
 const PLANNER_SYSTEM = `You are a **planner** for a software task. Break the work into **sequential phases** the coder will execute one at a time.
 
-## Finding the current thread ID
-
-The thread ID is a 26-character Crockford Base32 string (e.g. \`06F03H5V6JTMDST6P3TVH42RWM\`). It appears in the first message of this conversation. If you are unsure, run:
-
-  uncaged-workflow threads
-
-and use the ID of the active thread.
+Run \`uncaged-workflow skill develop\` for thread ID lookup, CAS commands, and meta output guide.
 
 ## Storing phase details — MANDATORY
 
-For each phase you MUST store its full detail text in CAS using this exact CLI command:
+For each phase, store its full detail text in CAS via \`uncaged-workflow cas put <THREAD_ID> '<content>'\`. The command prints a content-hash — use that as the phase identifier.
 
-  uncaged-workflow cas put <THREAD_ID> '# <name>
+The thread ID (26-char Crockford Base32) appears in the first message. If unsure, run \`uncaged-workflow thread list\`.
 
-Description: <description>
+**Do NOT store phase details in any other way** — the CLI is the only supported storage mechanism.
 
-Acceptance: <acceptance>'
+## Phase granularity
 
-Replace \`<THREAD_ID>\` with the actual thread ID you found above. The command prints a content-hash to stdout — use that hash as the phase identifier.
+Match the number of phases to task complexity:
+- Trivial (add a config option, fix a typo, rename): 1 phase
+- Small (a new feature touching 2-3 files): 1-2 phases
+- Medium (cross-module refactor): 2-3 phases
+- Large (new subsystem, architectural change): 3-5 phases
 
-**Do NOT store phase details in any other way** (no temp files, no invented paths). The CLI command is the only supported storage mechanism.
+Fewer phases is always better. Each phase must justify its existence — if two phases would be tested together anyway, merge them.
 
 ## Output format
 
@@ -49,4 +47,6 @@ export const plannerRole: RoleDefinition<PlannerMeta> = {
   extractPrompt:
     "Extract the implementation phases from the agent's output. Each phase has a hash (the CAS content-hash returned by the cas put command) and a title (one-line summary).",
   schema: plannerMetaSchema,
+  extractRefs: (meta) => meta.phases.map((p) => p.hash),
+  extractMode: "single",
 };

packages/workflow-template-develop/src/roles/reviewer.ts

@@ -21,4 +21,6 @@ export const reviewerRole: RoleDefinition<ReviewerMeta> = {
   extractPrompt:
     "Extract the review verdict: approved or rejected. If rejected, list the blocking issues.",
   schema: reviewerMetaSchema,
+  extractRefs: null,
+  extractMode: "single",
 };

packages/workflow-template-develop/src/roles/tester.ts

@@ -0,0 +1,27 @@
+import type { RoleDefinition } from "@uncaged/workflow";
+import * as z from "zod/v4";
+
+export const testerMetaSchema = z.discriminatedUnion("status", [
+  z.object({
+    status: z.literal("passed"),
+    details: z.string(),
+  }),
+  z.object({
+    status: z.literal("failed"),
+    details: z.string(),
+  }),
+]);
+
+export type TesterMeta = z.infer<typeof testerMetaSchema>;
+
+const TESTER_SYSTEM = `You are a tester. Run the project's test suite, build, and lint commands. Check what commands are available from the preparer's output in the thread. Report pass/fail with details of what failed.`;
+
+export const testerRole: RoleDefinition<TesterMeta> = {
+  description: "Runs test, build, and lint commands and reports pass or fail with details.",
+  systemPrompt: TESTER_SYSTEM,
+  extractPrompt:
+    "Extract the verification result: passed with summary details, or failed with details of what broke.",
+  schema: testerMetaSchema,
+  extractRefs: null,
+  extractMode: "single",
+};

packages/workflow-template-solve-issue/__tests__/solve-issue-template.test.ts

@@ -1,5 +1,9 @@
 import { afterEach, describe, expect, test } from "bun:test";
+import { mkdtemp, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
 import {
+  createCasStore,
   createExtract,
   END,
   type ModeratorContext,
@@ -7,42 +11,69 @@ import {
   START,
   validateWorkflowDescriptor,
 } from "@uncaged/workflow";
-
-import type { CoderMeta } from "@uncaged/workflow-role-coder";
-import type { PlannerMeta } from "@uncaged/workflow-role-planner";
-import type { PreparerMeta } from "@uncaged/workflow-role-preparer";
-
 import { buildSolveIssueDescriptor } from "../src/descriptor.js";
+import type { DeveloperMeta } from "../src/developer.js";
 import { createSolveIssueRun, solveIssueModerator } from "../src/index.js";
+import type { PreparerMeta, SubmitterMeta } from "../src/roles/index.js";
 import type { SolveIssueMeta } from "../src/roles.js";
 
-const DEFAULT_PHASES: PlannerMeta["phases"] = [
-  {
-    hash: "4KNMR2PX",
-    title: "Do the work",
-  },
-];
+function jsonResponse(payload: Record<string, unknown>): Response {
+  return new Response(JSON.stringify(payload), {
+    status: 200,
+    headers: { "Content-Type": "application/json" },
+  });
+}
 
-const EXPECT_PLANNER_META: PlannerMeta = {
-  phases: [
-    {
-      hash: "7BQST3VW",
-      title: "placeholder phase",
-    },
-  ],
-};
+function readToolListFromBody(init: RequestInit | undefined): readonly Record<string, unknown>[] {
+  if (init === undefined || init.body === undefined || init.body === null) {
+    return [];
+  }
+  const body = JSON.parse(String(init.body)) as Record<string, unknown>;
+  const tools = body.tools;
+  if (!Array.isArray(tools)) {
+    return [];
+  }
+  return tools.filter((t): t is Record<string, unknown> => t !== null && typeof t === "object");
+}
 
-const EXPECT_CODER_META: CoderMeta = {
-  completedPhase: "7BQST3VW",
-  filesChanged: [],
-  summary: "",
-};
+function singleToolName(tools: readonly Record<string, unknown>[]): string {
+  if (tools.length === 0) {
+    return "extract";
+  }
+  const fn = tools[0].function as Record<string, unknown> | undefined;
+  return typeof fn?.name === "string" ? fn.name : "extract";
+}
+
+function buildSingleModeResponse(args: Record<string, unknown>, toolName: string): Response {
+  return jsonResponse({
+    choices: [
+      {
+        message: {
+          tool_calls: [
+            {
+              type: "function",
+              function: { name: toolName, arguments: JSON.stringify(args) },
+            },
+          ],
+        },
+      },
+    ],
+  });
+}
+
+function buildReactModeResponse(args: Record<string, unknown>): Response {
+  // reactExtract accepts a plain-JSON assistant message and validates it
+  // directly against the schema, so we skip the cas_get / extract tool dance.
+  return jsonResponse({
+    choices: [{ message: { content: JSON.stringify(args) } }],
+  });
+}
 
 function installMockChatCompletions(sequence: ReadonlyArray<Record<string, unknown>>): () => void {
   const origFetch = globalThis.fetch;
   let i = 0;
   const mockFetch = async (
-    input: Parameters<typeof fetch>[0],
+    _input: Parameters<typeof fetch>[0],
     init?: RequestInit,
   ): Promise<Response> => {
     const args = sequence[i] ?? sequence[sequence.length - 1];
@@ -50,36 +81,11 @@ function installMockChatCompletions(sequence: ReadonlyArray<Record<string, unkno
       throw new Error("installMockChatCompletions: empty sequence");
     }
     i += 1;
-    void input;
-    const body = init?.body ? (JSON.parse(String(init.body)) as Record<string, unknown>) : {};
-    const tools = body.tools;
-    const firstTool =
-      Array.isArray(tools) && tools.length > 0 && tools[0] !== null && typeof tools[0] === "object"
-        ? (tools[0] as Record<string, unknown>)
-        : null;
-    const fn =
-      firstTool !== null ? (firstTool.function as Record<string, unknown> | undefined) : undefined;
-    const toolName = typeof fn?.name === "string" ? fn.name : "extract";
-    return new Response(
-      JSON.stringify({
-        choices: [
-          {
-            message: {
-              tool_calls: [
-                {
-                  type: "function",
-                  function: {
-                    name: toolName,
-                    arguments: JSON.stringify(args),
-                  },
-                },
-              ],
-            },
-          },
-        ],
-      }),
-      { status: 200, headers: { "Content-Type": "application/json" } },
-    );
+    const tools = readToolListFromBody(init);
+    if (tools.length > 1) {
+      return buildReactModeResponse(args);
+    }
+    return buildSingleModeResponse(args, singleToolName(tools));
   };
   globalThis.fetch = Object.assign(mockFetch, {
     preconnect: origFetch.preconnect.bind(origFetch),
@@ -104,6 +110,7 @@ function makeCtx(
 ): ModeratorContext<SolveIssueMeta> {
   return {
     threadId: "01TEST000000000000000000TR",
+    depth: 0,
     start: makeStart(maxRounds),
     steps,
   };
@@ -112,7 +119,7 @@ function makeCtx(
 function preparerStep(): RoleStep<SolveIssueMeta> {
   return {
     role: "preparer",
-    content: "prepared",
+    contentHash: "STUBHASHPREPARER01",
     meta: {
       repoPath: "/home/user/repos/test",
       defaultBranch: "main",
@@ -124,164 +131,108 @@ function preparerStep(): RoleStep<SolveIssueMeta> {
         buildCommand: "bun run build",
       },
     },
+    refs: [],
     timestamp: 0,
   };
 }
 
-function plannerStep(phases: PlannerMeta["phases"] = DEFAULT_PHASES): RoleStep<SolveIssueMeta> {
+function developerStep(): RoleStep<SolveIssueMeta> {
   return {
-    role: "planner",
-    content: "plan",
-    meta: { phases },
+    role: "developer",
+    contentHash: "STUBHASHDEVELOPER1",
+    meta: {
+      branch: "feat/issue-1",
+      commitSha: "abc1234",
+      filesChanged: ["src/login.ts"],
+      summary: "Fixed flaky login test by stabilising async setup.",
+    },
+    refs: [],
     timestamp: 1,
   };
 }
 
-function coderStep(completedPhase = "4KNMR2PX"): RoleStep<SolveIssueMeta> {
+function submitterStep(meta: SubmitterMeta): RoleStep<SolveIssueMeta> {
   return {
-    role: "coder",
-    content: "code",
-    meta: { completedPhase, filesChanged: ["a.ts"], summary: "fixed" },
+    role: "submitter",
+    contentHash: "STUBHASHSUBMITTER1",
+    meta,
+    refs: [],
     timestamp: 2,
   };
 }
 
-function reviewerStep(approved: boolean): RoleStep<SolveIssueMeta> {
-  return {
-    role: "reviewer",
-    content: "rev",
-    meta: approved
-      ? { status: "approved" as const }
-      : { status: "rejected" as const, issues: ["needs fix"] },
-    timestamp: 3,
-  };
-}
-
-function committerStep(): RoleStep<SolveIssueMeta> {
-  return {
-    role: "committer",
-    content: "commit",
-    meta: { status: "committed", branch: "feat/issue-1", commitSha: "abc1234" },
-    timestamp: 4,
-  };
-}
-
 const stubExtract = createExtract({
   baseUrl: "http://127.0.0.1:9",
   apiKey: "",
   model: "test",
 });
 
+const stubLlmProvider = {
+  baseUrl: "http://127.0.0.1:9",
+  apiKey: "",
+  model: "test",
+};
+
 describe("solveIssueModerator", () => {
-  test("routes preparer → planner → coder → reviewer → committer → END", () => {
+  test("routes initial → preparer → developer → submitter → END", () => {
     expect(solveIssueModerator(makeCtx(20, []))).toBe("preparer");
-    expect(solveIssueModerator(makeCtx(20, [preparerStep()]))).toBe("planner");
-    expect(solveIssueModerator(makeCtx(20, [preparerStep(), plannerStep()]))).toBe("coder");
-    expect(solveIssueModerator(makeCtx(20, [preparerStep(), plannerStep(), coderStep()]))).toBe(
-      "reviewer",
-    );
-    expect(
-      solveIssueModerator(
-        makeCtx(20, [preparerStep(), plannerStep(), coderStep(), reviewerStep(true)]),
-      ),
-    ).toBe("committer");
+    expect(solveIssueModerator(makeCtx(20, [preparerStep()]))).toBe("developer");
+    expect(solveIssueModerator(makeCtx(20, [preparerStep(), developerStep()]))).toBe("submitter");
     expect(
       solveIssueModerator(
         makeCtx(20, [
           preparerStep(),
-          plannerStep(),
-          coderStep(),
-          reviewerStep(true),
-          committerStep(),
+          developerStep(),
+          submitterStep({
+            status: "submitted",
+            prUrl: "https://github.com/example/repo/pull/1",
+          }),
         ]),
       ),
     ).toBe(END);
   });
 
-  test("reviewer rejects → coder retry when budget allows", () => {
-    const steps: ModeratorContext<SolveIssueMeta>["steps"] = [
-      plannerStep(),
-      coderStep(),
-      reviewerStep(false),
-    ];
-    expect(solveIssueModerator(makeCtx(20, steps))).toBe("coder");
-  });
-
-  test("reviewer rejects → END when max rounds exhausted", () => {
-    const steps: ModeratorContext<SolveIssueMeta>["steps"] = [
-      plannerStep(),
-      coderStep(),
-      reviewerStep(false),
-    ];
-    expect(solveIssueModerator(makeCtx(4, steps))).toBe(END);
-  });
-
-  test("multiple planner phases → coder until all complete, then reviewer", () => {
-    const phases: PlannerMeta["phases"] = [
-      {
-        hash: "AA000001",
-        title: "first phase",
-      },
-      {
-        hash: "AA000002",
-        title: "second phase",
-      },
-    ];
-    expect(solveIssueModerator(makeCtx(20, [plannerStep(phases)]))).toBe("coder");
-    expect(solveIssueModerator(makeCtx(20, [plannerStep(phases), coderStep("AA000001")]))).toBe(
-      "coder",
-    );
+  test("submitter failed → END", () => {
     expect(
       solveIssueModerator(
-        makeCtx(20, [plannerStep(phases), coderStep("AA000001"), coderStep("AA000002")]),
+        makeCtx(20, [
+          preparerStep(),
+          developerStep(),
+          submitterStep({ status: "failed", error: "gh not authenticated" }),
+        ]),
       ),
-    ).toBe("reviewer");
+    ).toBe(END);
   });
 
-  test("one-shot coder reports only last phase hash → reviewer (moderator treats as all phases done)", () => {
-    const phases: PlannerMeta["phases"] = [
-      { hash: "BB000001", title: "setup branch" },
-      { hash: "BB000002", title: "write tests" },
-      { hash: "BB000003", title: "verify" },
-      { hash: "BB000004", title: "commit and pr" },
-    ];
-    expect(solveIssueModerator(makeCtx(20, [plannerStep(phases), coderStep("BB000004")]))).toBe(
-      "reviewer",
-    );
-  });
-
-  test("unrecognised completedPhase hash → coder retry when budget allows", () => {
-    const phases: PlannerMeta["phases"] = [
-      { hash: "CC000001", title: "first phase" },
-      { hash: "CC000002", title: "second phase" },
-    ];
-    expect(solveIssueModerator(makeCtx(20, [plannerStep(phases), coderStep("all-done")]))).toBe(
-      "coder",
-    );
-  });
-
-  test("incomplete phases → END when max rounds exhausted", () => {
-    const phases: PlannerMeta["phases"] = [
-      { hash: "DD000001", title: "first phase" },
-      { hash: "DD000002", title: "second phase" },
-    ];
-    const steps: ModeratorContext<SolveIssueMeta>["steps"] = [
-      plannerStep(phases),
-      coderStep("DD000001"),
-    ];
-    expect(solveIssueModerator(makeCtx(3, steps))).toBe(END);
+  test("returns END for any unexpected last step (defensive)", () => {
+    // A submitter step with a pseudo-unknown future status would still be
+    // routed to END, since the moderator is a closed switch over known roles.
+    expect(
+      solveIssueModerator(
+        makeCtx(20, [
+          preparerStep(),
+          developerStep(),
+          submitterStep({ status: "submitted", prUrl: "https://example.com/pr/1" }),
+        ]),
+      ),
+    ).toBe(END);
   });
 });
 
 describe("createSolveIssueRun", () => {
   let restoreFetch: (() => void) | null = null;
+  let casDir: string | undefined;
 
-  afterEach(() => {
+  afterEach(async () => {
     restoreFetch?.();
     restoreFetch = null;
+    if (casDir !== undefined) {
+      await rm(casDir, { recursive: true, force: true }).catch(() => {});
+      casDir = undefined;
+    }
   });
 
-  test("structured extraction yields preparer then planner meta from mocked chat completions", async () => {
+  test("structured extraction yields preparer meta from mocked chat completions", async () => {
     const EXPECT_PREPARER_META: PreparerMeta = {
       repoPath: "/home/user/repos/test",
       defaultBranch: "main",
@@ -293,12 +244,23 @@ describe("createSolveIssueRun", () => {
         buildCommand: "bun run build",
       },
     };
-    restoreFetch = installMockChatCompletions([EXPECT_PREPARER_META, EXPECT_PLANNER_META]);
+    restoreFetch = installMockChatCompletions([EXPECT_PREPARER_META]);
 
-    const run = createSolveIssueRun({ agent: async () => "" }, stubExtract);
+    casDir = await mkdtemp(join(tmpdir(), "solve-issue-cas-"));
+    const cas = createCasStore(casDir);
+
+    // Override developer so the test does not spin up a child workflow.
+    const run = createSolveIssueRun(
+      {
+        agent: async () => "",
+        overrides: { developer: async () => "stub-root-hash" },
+      },
+      stubExtract,
+      stubLlmProvider,
+    );
     const gen = run(
       { prompt: "task", steps: [] },
-      { threadId: "01TEST000000000000000000TR", maxRounds: 20 },
+      { threadId: "01TEST000000000000000000TR", maxRounds: 20, depth: 0, cas },
     );
     const first = await gen.next();
     expect(first.done).toBe(false);
@@ -307,14 +269,6 @@ describe("createSolveIssueRun", () => {
     }
     expect(first.value.role).toBe("preparer");
     expect(first.value.meta).toEqual(EXPECT_PREPARER_META);
-
-    const second = await gen.next();
-    expect(second.done).toBe(false);
-    if (second.done) {
-      throw new Error("expected yield");
-    }
-    expect(second.value.role).toBe("planner");
-    expect(second.value.meta).toEqual(EXPECT_PLANNER_META);
   });
 
   test("per-role agent overrides default", async () => {
@@ -324,11 +278,20 @@ describe("createSolveIssueRun", () => {
       conventions: null,
       toolchain: { packageManager: null, testCommand: null, lintCommand: null, buildCommand: null },
     };
-    restoreFetch = installMockChatCompletions([
-      PREPARER_META,
-      EXPECT_PLANNER_META,
-      EXPECT_CODER_META,
-    ]);
+    const DEVELOPER_META: DeveloperMeta = {
+      branch: "feat/x",
+      commitSha: "abc1234",
+      filesChanged: ["a.ts"],
+      summary: "did the work",
+    };
+    const SUBMITTER_META: SubmitterMeta = {
+      status: "submitted",
+      prUrl: "https://github.com/example/repo/pull/2",
+    };
+    restoreFetch = installMockChatCompletions([PREPARER_META, DEVELOPER_META, SUBMITTER_META]);
+
+    casDir = await mkdtemp(join(tmpdir(), "solve-issue-cas-"));
+    const cas = createCasStore(casDir);
 
     const calls: string[] = [];
     const run = createSolveIssueRun(
@@ -342,37 +305,87 @@ describe("createSolveIssueRun", () => {
             calls.push("preparer");
             return "";
           },
-          planner: async () => {
-            calls.push("planner");
-            return "";
+          developer: async () => {
+            calls.push("developer");
+            return "stub-root-hash";
           },
-          coder: async () => {
-            calls.push("coder");
+          submitter: async () => {
+            calls.push("submitter");
             return "";
           },
         },
       },
       stubExtract,
+      stubLlmProvider,
     );
     const gen = run(
       { prompt: "task", steps: [] },
-      { threadId: "01TEST000000000000000000TR", maxRounds: 20 },
+      { threadId: "01TEST000000000000000000TR", maxRounds: 20, depth: 0, cas },
     );
     await gen.next();
     expect(calls).toEqual(["preparer"]);
 
     calls.length = 0;
     await gen.next();
-    expect(calls).toEqual(["planner"]);
+    expect(calls).toEqual(["developer"]);
 
     calls.length = 0;
     await gen.next();
-    expect(calls).toEqual(["coder"]);
+    expect(calls).toEqual(["submitter"]);
+  });
+
+  test("developer defaults to workflowAsAgent override (caller override still wins)", async () => {
+    const PREPARER_META: PreparerMeta = {
+      repoPath: "/tmp/r",
+      defaultBranch: "main",
+      conventions: null,
+      toolchain: { packageManager: null, testCommand: null, lintCommand: null, buildCommand: null },
+    };
+    const DEVELOPER_META: DeveloperMeta = {
+      branch: "feat/y",
+      commitSha: "def5678",
+      filesChanged: ["b.ts"],
+      summary: "more work",
+    };
+    restoreFetch = installMockChatCompletions([PREPARER_META, DEVELOPER_META]);
+
+    casDir = await mkdtemp(join(tmpdir(), "solve-issue-cas-"));
+    const cas = createCasStore(casDir);
+
+    let developerInvocations = 0;
+    const run = createSolveIssueRun(
+      {
+        agent: async () => "",
+        overrides: {
+          developer: async () => {
+            developerInvocations += 1;
+            return "stub-root-hash";
+          },
+        },
+      },
+      stubExtract,
+      stubLlmProvider,
+    );
+    const gen = run(
+      { prompt: "task", steps: [] },
+      { threadId: "01TEST000000000000000000TR", maxRounds: 20, depth: 0, cas },
+    );
+    // preparer
+    await gen.next();
+    // developer (caller override should be invoked, NOT workflowAsAgent default)
+    const devYield = await gen.next();
+    expect(devYield.done).toBe(false);
+    if (devYield.done) {
+      throw new Error("expected yield");
+    }
+    expect(devYield.value.role).toBe("developer");
+    expect(devYield.value.meta).toEqual(DEVELOPER_META);
+    expect(developerInvocations).toBe(1);
   });
 });
 
 describe("buildSolveIssueDescriptor", () => {
-  test("lists all roles with schemas that validate", () => {
+  test("lists preparer, developer, submitter with schemas that validate", () => {
     const descriptor = buildSolveIssueDescriptor();
     const validated = validateWorkflowDescriptor(descriptor);
     expect(validated.ok).toBe(true);
@@ -380,13 +393,11 @@ describe("buildSolveIssueDescriptor", () => {
       throw new Error(validated.error);
     }
     expect(Object.keys(validated.value.roles).sort()).toEqual([
-      "coder",
-      "committer",
-      "planner",
+      "developer",
       "preparer",
-      "reviewer",
+      "submitter",
     ]);
-    for (const key of ["preparer", "planner", "coder", "reviewer", "committer"] as const) {
+    for (const key of ["preparer", "developer", "submitter"] as const) {
       const role = validated.value.roles[key];
       expect(role).toBeDefined();
       expect(typeof role.schema).toBe("object");

packages/workflow-template-solve-issue/__tests__/submitter.test.ts

@@ -0,0 +1,39 @@
+import { describe, expect, test } from "bun:test";
+
+import { submitterMetaSchema, submitterRole } from "../src/roles/submitter.js";
+
+describe("submitterRole", () => {
+  test("submitted sample validates against schema", () => {
+    const parsed = submitterMetaSchema.safeParse({
+      status: "submitted" as const,
+      prUrl: "https://github.com/example/repo/pull/42",
+    });
+    expect(parsed.success).toBe(true);
+  });
+
+  test("failed sample validates against schema", () => {
+    const parsed = submitterMetaSchema.safeParse({
+      status: "failed" as const,
+      error: "gh not authenticated",
+    });
+    expect(parsed.success).toBe(true);
+  });
+
+  test("rejects unknown status discriminant", () => {
+    const parsed = submitterMetaSchema.safeParse({
+      status: "queued",
+      prUrl: "https://example.com",
+    });
+    expect(parsed.success).toBe(false);
+  });
+
+  test("exposes submitter system prompt", () => {
+    expect(submitterRole.systemPrompt).toContain("submitter");
+    expect(submitterRole.systemPrompt).toContain("pull request");
+  });
+
+  test("uses single extract mode without refs", () => {
+    expect(submitterRole.extractMode).toBe("single");
+    expect(submitterRole.extractRefs).toBeNull();
+  });
+});

packages/workflow-template-solve-issue/package.json

@@ -10,10 +10,6 @@
   },
   "dependencies": {
     "@uncaged/workflow": "workspace:*",
-    "@uncaged/workflow-role-committer": "workspace:*",
-    "@uncaged/workflow-role-coder": "workspace:*",
-    "@uncaged/workflow-role-planner": "workspace:*",
-    "@uncaged/workflow-role-preparer": "workspace:*",
-    "@uncaged/workflow-role-reviewer": "workspace:*"
+    "zod": "^4.0.0"
   }
 }

packages/workflow-template-solve-issue/src/developer.ts

@@ -0,0 +1,37 @@
+import type { RoleDefinition } from "@uncaged/workflow";
+import * as z from "zod/v4";
+
+export const developerMetaSchema = z.object({
+  branch: z.string(),
+  commitSha: z.string(),
+  filesChanged: z.array(z.string()),
+  summary: z.string(),
+});
+
+export type DeveloperMeta = z.infer<typeof developerMetaSchema>;
+
+const DEVELOPER_SYSTEM = `You are the **developer**. You delegate the implementation work to the \`develop\` workflow.
+
+The actual implementation (planning → coding → reviewing → testing → committing) is handled by a child workflow that runs in your place. Your output is the Merkle DAG root hash of that child thread.
+
+Pass through the task and let the child workflow do the work.`;
+
+const DEVELOPER_EXTRACT_PROMPT = `The agent output is the root CAS hash of a child workflow thread. Use the cas_get tool to traverse the Merkle DAG and extract the developer summary.
+
+Procedure:
+1. cas_get(<rootHash>) — the root node lists all child step hashes (planner, coder, reviewer, tester, committer).
+2. Find the committer step. cas_get its hash to read the committer's meta — extract branch and commitSha from there.
+3. Find every coder step. cas_get each to read the coder's filesChanged. Union all filesChanged across coder steps.
+4. Compose a short human-readable summary describing what the develop child workflow accomplished (drawn from the coder summaries, or a synthesis of them).
+
+Return: { branch, commitSha, filesChanged, summary }.`;
+
+export const developerRole: RoleDefinition<DeveloperMeta> = {
+  description:
+    "Delegates the actual implementation to the develop workflow (workflow-as-agent). Produces a summary by traversing the child thread's Merkle DAG.",
+  systemPrompt: DEVELOPER_SYSTEM,
+  extractPrompt: DEVELOPER_EXTRACT_PROMPT,
+  schema: developerMetaSchema,
+  extractRefs: () => [],
+  extractMode: "react",
+};

packages/workflow-template-solve-issue/src/index.ts

@@ -2,41 +2,30 @@ import {
   type AgentBinding,
   createWorkflow,
   type ExtractFn,
+  type LlmProvider,
   type WorkflowDefinition,
   type WorkflowFn,
+  workflowAsAgent,
 } from "@uncaged/workflow";
 
 import { solveIssueModerator } from "./moderator.js";
 import { SOLVE_ISSUE_WORKFLOW_DESCRIPTION, type SolveIssueMeta, solveIssueRoles } from "./roles.js";
 
+export { buildSolveIssueDescriptor } from "./descriptor.js";
 export {
-  type CoderMeta,
-  coderMetaSchema,
-  coderRole,
-} from "@uncaged/workflow-role-coder";
-export {
-  type CommitterMeta,
-  committerMetaSchema,
-  committerRole,
-} from "@uncaged/workflow-role-committer";
-export {
-  type PlannerMeta,
-  phaseSchema,
-  plannerMetaSchema,
-  plannerRole,
-} from "@uncaged/workflow-role-planner";
+  type DeveloperMeta,
+  developerMetaSchema,
+  developerRole,
+} from "./developer.js";
+export { solveIssueModerator } from "./moderator.js";
 export {
   type PreparerMeta,
   preparerMetaSchema,
   preparerRole,
-} from "@uncaged/workflow-role-preparer";
-export {
-  type ReviewerMeta,
-  reviewerMetaSchema,
-  reviewerRole,
-} from "@uncaged/workflow-role-reviewer";
-export { buildSolveIssueDescriptor } from "./descriptor.js";
-export { solveIssueModerator } from "./moderator.js";
+  type SubmitterMeta,
+  submitterMetaSchema,
+  submitterRole,
+} from "./roles/index.js";
 export {
   SOLVE_ISSUE_WORKFLOW_DESCRIPTION,
   type SolveIssueMeta,
@@ -50,6 +39,25 @@ export const solveIssueWorkflowDefinition: WorkflowDefinition<SolveIssueMeta> =
   moderator: solveIssueModerator,
 };
 
-export function createSolveIssueRun(binding: AgentBinding, extract: ExtractFn): WorkflowFn {
-  return createWorkflow(solveIssueWorkflowDefinition, binding, extract);
+/**
+ * Build the solve-issue {@link WorkflowFn}.
+ *
+ * The `developer` role always delegates to the registered `develop` workflow via
+ * {@link workflowAsAgent}; if the caller supplies their own `developer` override in
+ * `binding.overrides`, it takes precedence so tests and custom hosts can stub it.
+ */
+export function createSolveIssueRun(
+  binding: AgentBinding,
+  extract: ExtractFn,
+  llmProvider: LlmProvider | null,
+): WorkflowFn {
+  const developerOverride = binding.overrides?.developer ?? workflowAsAgent("develop");
+  const mergedBinding: AgentBinding = {
+    agent: binding.agent,
+    overrides: {
+      ...(binding.overrides ?? {}),
+      developer: developerOverride,
+    },
+  };
+  return createWorkflow(solveIssueWorkflowDefinition, mergedBinding, extract, llmProvider);
 }

packages/workflow-template-solve-issue/src/moderator.ts

@@ -1,52 +1,9 @@
-import type { Moderator, ModeratorContext } from "@uncaged/workflow";
+import type { Moderator } from "@uncaged/workflow";
 import { END } from "@uncaged/workflow";
 
 import type { SolveIssueMeta } from "./roles.js";
 
-function coderFinishedAllPlannedPhases(
-  phases: ReadonlyArray<{ hash: string }>,
-  coderCompletedPhases: ReadonlyArray<string>,
-): boolean {
-  if (phases.length === 0) {
-    return true;
-  }
-  const plannedHashes = new Set(phases.map((p) => p.hash));
-  const lastHash = phases[phases.length - 1].hash;
-  const explicit = new Set(coderCompletedPhases.filter((h) => plannedHashes.has(h)));
-  if (phases.every((p) => explicit.has(p.hash))) {
-    return true;
-  }
-  if (coderCompletedPhases.some((h) => h === lastHash)) {
-    return true;
-  }
-  return false;
-}
-
-function nextAfterCoder(
-  ctx: ModeratorContext<SolveIssueMeta>,
-  maxRounds: number,
-): (keyof SolveIssueMeta & string) | typeof END {
-  const plannerStep = ctx.steps.find((s) => s.role === "planner");
-  if (plannerStep === undefined) {
-    return "reviewer";
-  }
-  const phases = plannerStep.meta.phases;
-  const coderCompletedPhases = ctx.steps
-    .filter((s) => s.role === "coder")
-    .map((s) => s.meta.completedPhase);
-  const allDone = coderFinishedAllPlannedPhases(phases, coderCompletedPhases);
-  if (allDone) {
-    return "reviewer";
-  }
-  if (ctx.steps.length < maxRounds - 1) {
-    return "coder";
-  }
-  return END;
-}
-
 export const solveIssueModerator: Moderator<SolveIssueMeta> = (ctx) => {
-  const maxRounds = ctx.start.meta.maxRounds;
-
   if (ctx.steps.length === 0) {
     return "preparer";
   }
@@ -54,31 +11,14 @@ export const solveIssueModerator: Moderator<SolveIssueMeta> = (ctx) => {
   const last = ctx.steps[ctx.steps.length - 1];
 
   if (last.role === "preparer") {
-    return "planner";
+    return "developer";
   }
 
-  if (last.role === "planner") {
-    return "coder";
+  if (last.role === "developer") {
+    return "submitter";
   }
 
-  if (last.role === "coder") {
-    return nextAfterCoder(ctx, maxRounds);
-  }
-
-  if (last.role === "reviewer") {
-    if (last.meta.status === "approved") {
-      return "committer";
-    }
-    if (ctx.steps.length < maxRounds - 1) {
-      return "coder";
-    }
-    return END;
-  }
-
-  if (last.role === "committer") {
-    if (last.meta.status === "recoverable" && ctx.steps.length < maxRounds - 1) {
-      return "coder";
-    }
+  if (last.role === "submitter") {
     return END;
   }
 

packages/workflow-template-solve-issue/src/roles.ts

@@ -1,19 +1,15 @@
 import type { RoleDefinition } from "@uncaged/workflow";
-import { type CoderMeta, coderRole } from "@uncaged/workflow-role-coder";
-import { type CommitterMeta, committerRole } from "@uncaged/workflow-role-committer";
-import { type PlannerMeta, plannerRole } from "@uncaged/workflow-role-planner";
-import { type PreparerMeta, preparerRole } from "@uncaged/workflow-role-preparer";
-import { type ReviewerMeta, reviewerRole } from "@uncaged/workflow-role-reviewer";
+import { type DeveloperMeta, developerRole } from "./developer.js";
+import { type PreparerMeta, preparerRole } from "./roles/preparer.js";
+import { type SubmitterMeta, submitterRole } from "./roles/submitter.js";
 
 export const SOLVE_ISSUE_WORKFLOW_DESCRIPTION =
-  "Prepare repo context, plan phases, implement incrementally, review, and commit to resolve an issue end-to-end (preparer → planner → coder [repeat per phase] → reviewer → committer).";
+  "Resolve an issue end-to-end by preparing the repo, delegating implementation to the develop workflow, and opening a pull request (preparer → developer → submitter).";
 
 export type SolveIssueMeta = {
   preparer: PreparerMeta;
-  planner: PlannerMeta;
-  coder: CoderMeta;
-  reviewer: ReviewerMeta;
-  committer: CommitterMeta;
+  developer: DeveloperMeta;
+  submitter: SubmitterMeta;
 };
 
 export type SolveIssueRoles = {
@@ -22,8 +18,6 @@ export type SolveIssueRoles = {
 
 export const solveIssueRoles: SolveIssueRoles = {
   preparer: preparerRole,
-  planner: plannerRole,
-  coder: coderRole,
-  reviewer: reviewerRole,
-  committer: committerRole,
+  developer: developerRole,
+  submitter: submitterRole,
 };

packages/workflow-template-solve-issue/src/roles/index.ts

@@ -3,3 +3,4 @@ export {
   preparerMetaSchema,
   preparerRole,
 } from "./preparer.js";
+export { type SubmitterMeta, submitterMetaSchema, submitterRole } from "./submitter.js";

packages/workflow-template-solve-issue/src/roles/preparer.ts

@@ -47,4 +47,6 @@ export const preparerRole: RoleDefinition<PreparerMeta> = {
   extractPrompt:
     "Extract repoPath (absolute path), defaultBranch, conventions (summary string or null), and toolchain (packageManager, testCommand, lintCommand, buildCommand — each string or null).",
   schema: preparerMetaSchema,
+  extractRefs: null,
+  extractMode: "single",
 };

packages/workflow-template-solve-issue/src/roles/submitter.ts

@@ -0,0 +1,44 @@
+import type { RoleDefinition } from "@uncaged/workflow";
+import * as z from "zod/v4";
+
+export const submitterMetaSchema = z.discriminatedUnion("status", [
+  z.object({
+    status: z.literal("submitted"),
+    prUrl: z.string(),
+  }),
+  z.object({
+    status: z.literal("failed"),
+    error: z.string(),
+  }),
+]);
+
+export type SubmitterMeta = z.infer<typeof submitterMetaSchema>;
+
+const SUBMITTER_SYSTEM = `You are the **submitter**. Your job is to push the work branch to the remote and open a pull request.
+
+## Inputs
+
+Read the thread for context:
+- The **preparer**'s output gives you the absolute repo path and the default branch (and remote URL by inspecting the repo).
+- The **developer**'s output gives you the branch name that was committed and a list of files changed plus a summary of the work.
+
+## Procedure
+
+1. \`cd\` into the repo path from the preparer's output.
+2. Push the developer's branch to the remote: \`git push -u origin <branch>\`.
+3. Open a pull request (e.g. via \`gh pr create\`) targeting the default branch. The PR title should be short and describe the change. The PR description should summarize what changed (drawing from the developer's summary and filesChanged) and reference the original issue/task if applicable.
+4. Report the resulting PR URL.
+
+On any failure (push rejected, gh not authenticated, PR creation failed, etc.), report status="failed" with a short error message. Do not retry — surface the error so the moderator can decide.`;
+
+const SUBMITTER_EXTRACT_PROMPT =
+  "Extract the submission result. status='submitted' with prUrl on success, or status='failed' with a short error message on failure.";
+
+export const submitterRole: RoleDefinition<SubmitterMeta> = {
+  description: "Pushes the developer's branch to the remote and opens a pull request.",
+  systemPrompt: SUBMITTER_SYSTEM,
+  extractPrompt: SUBMITTER_EXTRACT_PROMPT,
+  schema: submitterMetaSchema,
+  extractRefs: null,
+  extractMode: "single",
+};

packages/workflow-template-solve-issue/tsconfig.json

@@ -6,11 +6,5 @@
     "composite": true
   },
   "include": ["src/**/*.ts"],
-  "references": [
-    { "path": "../workflow" },
-    { "path": "../workflow-role-coder" },
-    { "path": "../workflow-role-committer" },
-    { "path": "../workflow-role-planner" },
-    { "path": "../workflow-role-reviewer" }
-  ]
+  "references": [{ "path": "../workflow" }]
 }

packages/workflow-util-agent/__tests__/build-agent-prompt.test.ts

@@ -1,5 +1,8 @@
-import { describe, expect, test } from "bun:test";
-import { START, type ThreadContext } from "@uncaged/workflow";
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { mkdtemp, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { createCasStore, putContentMerkleNode, START, type ThreadContext } from "@uncaged/workflow";
 
 import { buildAgentPrompt } from "../src/index.js";
 
@@ -13,35 +16,53 @@ function startTask(content: string): ThreadContext["start"] {
 }
 
 describe("buildAgentPrompt", () => {
-  test("includes system prompt and full task; omits tools when there are no steps", () => {
+  let casRoot: string;
+
+  beforeEach(async () => {
+    casRoot = await mkdtemp(join(tmpdir(), "wf-build-prompt-cas-"));
+  });
+
+  afterEach(async () => {
+    await rm(casRoot, { recursive: true, force: true });
+  });
+
+  test("includes system prompt and full task; omits tools when there are no steps", async () => {
+    const cas = createCasStore(casRoot);
     const ctx: ThreadContext = {
       start: startTask("fix the bug"),
+      depth: 0,
       steps: [],
       threadId: "01TEST000000000000000000TR",
       currentRole: { name: START, systemPrompt: "You are an agent." },
+      cas,
     };
-    const text = buildAgentPrompt(ctx);
+    const text = await buildAgentPrompt(ctx);
     expect(text).toContain("You are an agent.");
     expect(text).toContain("## Task");
     expect(text).toContain("fix the bug");
     expect(text).not.toContain("## Tools");
   });
 
-  test("single step shows full content and meta, and includes tools", () => {
+  test("single step shows full content and meta, and includes tools", async () => {
+    const cas = createCasStore(casRoot);
+    const onlyHash = await putContentMerkleNode(cas, "only step full body");
     const ctx: ThreadContext = {
       start: startTask("user task"),
+      depth: 0,
       threadId: "01TEST000000000000000000TR",
       currentRole: { name: "coder", systemPrompt: "Be helpful." },
+      cas,
       steps: [
         {
           role: "coder",
-          content: "only step full body",
+          contentHash: onlyHash,
           meta: { files: ["a.ts"] },
+          refs: [onlyHash],
           timestamp: 2,
         },
       ],
     };
-    const text = buildAgentPrompt(ctx);
+    const text = await buildAgentPrompt(ctx);
     expect(text).toContain("## Task");
     expect(text).toContain("user task");
     expect(text).toContain("## Step: coder");
@@ -51,27 +72,34 @@ describe("buildAgentPrompt", () => {
     expect(text).toContain("uncaged-workflow thread 01TEST000000000000000000TR");
   });
 
-  test("two or more steps: previous steps are meta-only; latest step is full", () => {
+  test("two or more steps: previous steps are meta-only; latest step is full", async () => {
+    const cas = createCasStore(casRoot);
+    const plannerHash = await putContentMerkleNode(cas, "PLANNER_SECRET_FULL_TEXT");
+    const coderHash = await putContentMerkleNode(cas, "last step full content");
     const ctx: ThreadContext = {
       start: startTask("first message full: task content here"),
+      depth: 0,
       threadId: "01TEST000000000000000000TR",
       currentRole: { name: "coder", systemPrompt: "System." },
+      cas,
       steps: [
         {
           role: "planner",
-          content: "PLANNER_SECRET_FULL_TEXT",
+          contentHash: plannerHash,
           meta: { plan: "short" },
+          refs: [plannerHash],
           timestamp: 2,
         },
         {
           role: "coder",
-          content: "last step full content",
+          contentHash: coderHash,
           meta: { done: true },
+          refs: [coderHash],
           timestamp: 3,
         },
       ],
     };
-    const text = buildAgentPrompt(ctx);
+    const text = await buildAgentPrompt(ctx);
     expect(text).toContain("first message full: task content here");
     expect(text).toContain("## Previous Steps");
     expect(text).toContain("### Step 1: planner");
@@ -84,33 +112,42 @@ describe("buildAgentPrompt", () => {
     expect(text).toContain("uncaged-workflow thread 01TEST000000000000000000TR");
   });
 
-  test("middle steps show meta summary only, not full content", () => {
+  test("middle steps show meta summary only, not full content", async () => {
+    const cas = createCasStore(casRoot);
+    const ha = await putContentMerkleNode(cas, "HIDDEN_A");
+    const hb = await putContentMerkleNode(cas, "HIDDEN_B_MIDDLE");
+    const hc = await putContentMerkleNode(cas, "VISIBLE_LAST");
     const ctx: ThreadContext = {
       start: startTask("start"),
+      depth: 0,
       threadId: "01TEST000000000000000000TR",
       currentRole: { name: "c", systemPrompt: "S" },
+      cas,
       steps: [
         {
           role: "a",
-          content: "HIDDEN_A",
+          contentHash: ha,
           meta: { n: 1 },
+          refs: [ha],
           timestamp: 2,
         },
         {
           role: "b",
-          content: "HIDDEN_B_MIDDLE",
+          contentHash: hb,
           meta: { n: 2 },
+          refs: [hb],
           timestamp: 3,
         },
         {
           role: "c",
-          content: "VISIBLE_LAST",
+          contentHash: hc,
           meta: { n: 3 },
+          refs: [hc],
           timestamp: 4,
         },
       ],
     };
-    const text = buildAgentPrompt(ctx);
+    const text = await buildAgentPrompt(ctx);
     expect(text).not.toContain("HIDDEN_A");
     expect(text).not.toContain("HIDDEN_B_MIDDLE");
     expect(text).toContain('Summary: {"n":1}');

packages/workflow-util-agent/src/build-agent-prompt.ts

@@ -1,7 +1,16 @@
 import type { AgentContext } from "@uncaged/workflow";
+import { getContentMerklePayload } from "@uncaged/workflow";
+
+async function resolveStepText(ctx: AgentContext, contentHash: string): Promise<string> {
+  const text = await getContentMerklePayload(ctx.cas, contentHash);
+  if (text === null) {
+    throw new Error(`buildAgentPrompt: missing CAS blob for ${contentHash}`);
+  }
+  return text;
+}
 
 /** Builds the full agent prompt: system instructions plus summarized thread history. */
-export function buildAgentPrompt(ctx: AgentContext): string {
+export async function buildAgentPrompt(ctx: AgentContext): Promise<string> {
   const lines: string[] = [];
   lines.push(ctx.currentRole.systemPrompt);
   lines.push("");
@@ -15,10 +24,11 @@ export function buildAgentPrompt(ctx: AgentContext): string {
 
   if (steps.length === 1) {
     const s = steps[0];
+    const body = await resolveStepText(ctx, s.contentHash);
     lines.push("");
     lines.push(`## Step: ${s.role}`);
     lines.push("");
-    lines.push(s.content);
+    lines.push(body);
     lines.push("");
     lines.push(`Meta: ${JSON.stringify(s.meta)}`);
   } else {
@@ -31,10 +41,11 @@ export function buildAgentPrompt(ctx: AgentContext): string {
       lines.push(`Summary: ${JSON.stringify(s.meta)}`);
     }
     const last = steps[steps.length - 1];
+    const lastBody = await resolveStepText(ctx, last.contentHash);
     lines.push("");
     lines.push(`## Latest Step: ${last.role}`);
     lines.push("");
-    lines.push(last.content);
+    lines.push(lastBody);
     lines.push("");
     lines.push(`Meta: ${JSON.stringify(last.meta)}`);
   }

packages/workflow/__tests__/build-descriptor.test.ts

@@ -22,6 +22,8 @@ describe("buildDescriptor", () => {
           systemPrompt: "You are an analyst.",
           extractPrompt: "Extract title and count from the analysis.",
           schema,
+          extractRefs: null,
+          extractMode: "single",
         },
       },
       moderator: () => END,

packages/workflow/__tests__/bundle-validator.test.ts

@@ -26,6 +26,19 @@ export const run = async function* (input) {
     expect(r.ok).toBe(true);
   });
 
+  test("allows static import of @uncaged/workflow", () => {
+    const source = `${minimalDescriptor}import { putContentMerkleNode } from "@uncaged/workflow";
+
+export const run = async function* (_input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "x");
+  return { returnCode: 0, summary: h };
+};
+`;
+    const r = validateWorkflowBundle({ filePath: "/tmp/w.esm.js", source });
+    expect(r.ok).toBe(true);
+  });
+
   test("rejects wrong filename suffix", () => {
     const r = validateWorkflowBundle({
       filePath: "/tmp/w.js",

packages/workflow/__tests__/cas.test.ts

@@ -3,10 +3,16 @@ import { mkdtemp, rm } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 
-import { createThreadCas } from "../src/cas.js";
+import { createCasStore, createThreadCas } from "../src/cas.js";
 import { hashString } from "../src/hash.js";
 
-describe("createThreadCas", () => {
+describe("cas module exports", () => {
+  test("createThreadCas is a deprecated alias of createCasStore", () => {
+    expect(createThreadCas).toBe(createCasStore);
+  });
+});
+
+describe("createCasStore", () => {
   let casDir: string;
 
   beforeEach(async () => {
@@ -18,7 +24,7 @@ describe("createThreadCas", () => {
   });
 
   test("put returns consistent hash for same content", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const h1 = await cas.put("hello world");
     const h2 = await cas.put("hello world");
     expect(h1).toBe(h2);
@@ -26,14 +32,14 @@ describe("createThreadCas", () => {
   });
 
   test("put returns hash matching hashString", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const content = "some content to store";
     const h = await cas.put(content);
     expect(h).toBe(hashString(content));
   });
 
   test("get returns stored content", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const content = "line1\nline2\nline3";
     const h = await cas.put(content);
     const retrieved = await cas.get(h);
@@ -41,13 +47,13 @@ describe("createThreadCas", () => {
   });
 
   test("get returns null for missing hash", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const result = await cas.get("0000000000000");
     expect(result).toBeNull();
   });
 
   test("delete removes entry", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const h = await cas.put("to be deleted");
     await cas.delete(h);
     const result = await cas.get(h);
@@ -55,12 +61,12 @@ describe("createThreadCas", () => {
   });
 
   test("delete on missing hash does not throw", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     await cas.delete("0000000000000");
   });
 
   test("list returns all stored hashes", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const h1 = await cas.put("aaa");
     const h2 = await cas.put("bbb");
     const h3 = await cas.put("ccc");
@@ -69,13 +75,13 @@ describe("createThreadCas", () => {
   });
 
   test("list returns empty array when cas dir does not exist", async () => {
-    const cas = createThreadCas(join(casDir, "nonexistent"));
+    const cas = createCasStore(join(casDir, "nonexistent"));
     const hashes = await cas.list();
     expect(hashes).toEqual([]);
   });
 
   test("put is idempotent — same content written twice causes no error", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const h1 = await cas.put("idempotent");
     const h2 = await cas.put("idempotent");
     expect(h1).toBe(h2);
@@ -84,7 +90,7 @@ describe("createThreadCas", () => {
   });
 
   test("different content produces different hashes", async () => {
-    const cas = createThreadCas(casDir);
+    const cas = createCasStore(casDir);
     const h1 = await cas.put("alpha");
     const h2 = await cas.put("beta");
     expect(h1).not.toBe(h2);

packages/workflow/__tests__/engine.test.ts

@@ -4,11 +4,18 @@ import { tmpdir } from "node:os";
 import { join } from "node:path";
 import * as z from "zod/v4";
 
+import { createCasStore } from "../src/cas.js";
 import { createWorkflow } from "../src/create-workflow.js";
 import { executeThread } from "../src/engine.js";
 import { createExtract } from "../src/extract-fn.js";
 import { createLogger } from "../src/logger.js";
-import { END } from "../src/types.js";
+import {
+  createContentMerkleNode,
+  getContentMerklePayload,
+  parseMerkleNode,
+  serializeMerkleNode,
+} from "../src/merkle.js";
+import { END, type LlmProvider } from "../src/types.js";
 
 const plannerMetaSchema = z.object({
   plan: z.string(),
@@ -89,12 +96,16 @@ const demoWorkflow = createWorkflow<DemoMeta>(
         systemPrompt: "You are a planner.",
         extractPrompt: "Extract plan text and affected files list.",
         schema: plannerMetaSchema,
+        extractRefs: null,
+        extractMode: "single",
       },
       coder: {
         description: "Demo coder",
         systemPrompt: "You are a coder.",
         extractPrompt: "Extract the code diff summary.",
         schema: coderMetaSchema,
+        extractRefs: null,
+        extractMode: "single",
       },
     },
     moderator: (ctx) => {
@@ -115,6 +126,7 @@ const demoWorkflow = createWorkflow<DemoMeta>(
     },
   },
   demoExtract,
+  null,
 );
 
 describe("executeThread", () => {
@@ -138,6 +150,7 @@ describe("executeThread", () => {
       const dataPath = join(root, "logs", hash, `${threadId}.data.jsonl`);
       const infoPath = join(root, "logs", hash, `${threadId}.info.jsonl`);
       await mkdir(join(root, "logs", hash), { recursive: true });
+      const cas = createCasStore(join(root, "cas"));
 
       const logger = createLogger({ sink: { kind: "file", path: infoPath } });
       const ac = new AbortController();
@@ -148,16 +161,30 @@ describe("executeThread", () => {
         { prompt: "Fix the login redirect bug in #3", steps: [] },
         {
           maxRounds: 5,
+          depth: 0,
           signal: ac.signal,
           awaitAfterEachYield: async () => {},
           forkSourceThreadId: null,
           prefilledDiskSteps: null,
         },
-        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath },
+        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
         logger,
       );
 
       expect(result.returnCode).toBe(0);
+      expect(typeof result.rootHash).toBe("string");
+      expect(result.rootHash.length).toBeGreaterThan(0);
+
+      const rootYaml = await cas.get(result.rootHash);
+      expect(rootYaml).not.toBeNull();
+      const rootNode = parseMerkleNode(rootYaml ?? "");
+      expect(rootNode.type).toBe("thread");
+      const rootPayload = rootNode.payload as Record<string, unknown>;
+      expect(rootPayload.workflow).toBe("demo-flow");
+      expect(rootPayload.threadId).toBe(threadId);
+      const rootResult = rootPayload.result as Record<string, unknown>;
+      expect(rootResult.returnCode).toBe(0);
+      expect(rootNode.children.length).toBe(2);
 
       const dataText = await readFile(dataPath, "utf8");
       const lines = dataText
@@ -176,16 +203,34 @@ describe("executeThread", () => {
       expect(params.prompt).toBe("Fix the login redirect bug in #3");
       const opts = params.options as Record<string, unknown>;
       expect(opts.maxRounds).toBe(5);
-      expect(Object.keys(opts).sort()).toEqual(["maxRounds"]);
+      expect(opts.depth).toBe(0);
+      expect(Object.keys(opts).sort()).toEqual(["depth", "maxRounds"]);
 
       const role1 = JSON.parse(lines[1] ?? "{}") as Record<string, unknown>;
       expect(role1.role).toBe("planner");
-      expect(role1.content).toBe("plan-body");
+      expect(typeof role1.contentHash).toBe("string");
+      expect(await getContentMerklePayload(cas, String(role1.contentHash))).toBe("plan-body");
       expect(role1.meta).toEqual({ plan: "do-it", files: ["a.ts"] });
+      expect(role1.refs).toEqual([role1.contentHash]);
       expect(typeof role1.timestamp).toBe("number");
 
       const role2 = JSON.parse(lines[2] ?? "{}") as Record<string, unknown>;
       expect(role2.role).toBe("coder");
+      expect(role2.refs).toEqual([role2.contentHash]);
+
+      const step1Yaml = await cas.get(rootNode.children[0] ?? "");
+      const step2Yaml = await cas.get(rootNode.children[1] ?? "");
+      expect(step1Yaml).not.toBeNull();
+      expect(step2Yaml).not.toBeNull();
+      const step1Node = parseMerkleNode(step1Yaml ?? "");
+      const step2Node = parseMerkleNode(step2Yaml ?? "");
+      expect(step1Node.type).toBe("step");
+      expect(step2Node.type).toBe("step");
+      expect(step1Node.children).toEqual([String(role1.contentHash)]);
+      expect(step2Node.children).toEqual([String(role2.contentHash)]);
+      const step1Payload = step1Node.payload as Record<string, unknown>;
+      expect(step1Payload.role).toBe("planner");
+      expect(step1Payload.meta).toEqual({ plan: "do-it", files: ["a.ts"] });
 
       const infoText = await readFile(infoPath, "utf8");
       const infoLines = infoText
@@ -213,11 +258,14 @@ describe("executeThread", () => {
       const dataPath = join(root, "logs", hash, `${threadId}.data.jsonl`);
       const infoPath = join(root, "logs", hash, `${threadId}.info.jsonl`);
       await mkdir(join(root, "logs", hash), { recursive: true });
+      const cas = createCasStore(join(root, "cas"));
+      const plannerHash = await cas.put(serializeMerkleNode(createContentMerkleNode("plan-body")));
 
       const logger = createLogger({ sink: { kind: "file", path: infoPath } });
       const ac = new AbortController();
 
       const histTs = 9_000_000;
+      const mergedPlannerRefs = ["CAS111AAAAAAA", plannerHash];
       const result = await executeThread(
         demoWorkflow,
         "demo-flow",
@@ -226,30 +274,38 @@ describe("executeThread", () => {
           steps: [
             {
               role: "planner",
-              content: "plan-body",
+              contentHash: plannerHash,
               meta: { plan: "do-it", files: ["a.ts"] },
+              refs: mergedPlannerRefs,
             },
           ],
         },
         {
           maxRounds: 5,
+          depth: 0,
           signal: ac.signal,
           awaitAfterEachYield: async () => {},
           forkSourceThreadId: "01SRC1111111111111111111",
           prefilledDiskSteps: [
             {
               role: "planner",
-              content: "plan-body",
+              contentHash: plannerHash,
               meta: { plan: "do-it", files: ["a.ts"] },
+              refs: mergedPlannerRefs,
               timestamp: histTs,
             },
           ],
         },
-        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath },
+        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
         logger,
       );
 
       expect(result.returnCode).toBe(0);
+      expect(typeof result.rootHash).toBe("string");
+
+      const rootYaml = await cas.get(result.rootHash);
+      const rootNode = parseMerkleNode(rootYaml ?? "");
+      expect(rootNode.children.length).toBe(2);
 
       const dataText = await readFile(dataPath, "utf8");
       const lines = dataText
@@ -264,10 +320,11 @@ describe("executeThread", () => {
       const role0 = JSON.parse(lines[1] ?? "{}") as Record<string, unknown>;
       expect(role0.role).toBe("planner");
       expect(role0.timestamp).toBe(histTs);
+      expect(role0.refs).toEqual(mergedPlannerRefs);
 
       const role1 = JSON.parse(lines[2] ?? "{}") as Record<string, unknown>;
       expect(role1.role).toBe("coder");
-      expect(role1.content).toBe("code-body");
+      expect(await getContentMerklePayload(cas, String(role1.contentHash))).toBe("code-body");
     } finally {
       await rm(root, { recursive: true, force: true });
     }
@@ -281,6 +338,7 @@ describe("executeThread", () => {
       const dataPath = join(root, "logs", hash, `${threadId}.data.jsonl`);
       const infoPath = join(root, "logs", hash, `${threadId}.info.jsonl`);
       await mkdir(join(root, "logs", hash), { recursive: true });
+      const cas = createCasStore(join(root, "cas"));
 
       const logger = createLogger({ sink: { kind: "file", path: infoPath } });
       const ac = new AbortController();
@@ -291,16 +349,23 @@ describe("executeThread", () => {
         { prompt: "hello", steps: [] },
         {
           maxRounds: 0,
+          depth: 0,
           signal: ac.signal,
           awaitAfterEachYield: async () => {},
           forkSourceThreadId: null,
           prefilledDiskSteps: null,
         },
-        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath },
+        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
         logger,
       );
 
       expect(result.returnCode).toBe(0);
+      expect(typeof result.rootHash).toBe("string");
+
+      const rootYaml = await cas.get(result.rootHash);
+      const rootNode = parseMerkleNode(rootYaml ?? "");
+      expect(rootNode.type).toBe("thread");
+      expect(rootNode.children.length).toBe(0);
 
       const dataText = await readFile(dataPath, "utf8");
       const lines = dataText
@@ -312,4 +377,240 @@ describe("executeThread", () => {
       await rm(root, { recursive: true, force: true });
     }
   });
+
+  test("Merkle DAG: root → step nodes → content for full thread traversal", async () => {
+    restoreFetch = installMockChatCompletions([
+      { plan: "do-it", files: ["a.ts"] },
+      { diff: "+ok" },
+    ]);
+
+    const root = await mkdtemp(join(tmpdir(), "wf-engine-dag-"));
+    try {
+      const threadId = "01KQXKW18CT8G75T53R8F4G7YG";
+      const hash = "C9NMV6V2TQT81";
+      const dataPath = join(root, "logs", hash, `${threadId}.data.jsonl`);
+      const infoPath = join(root, "logs", hash, `${threadId}.info.jsonl`);
+      await mkdir(join(root, "logs", hash), { recursive: true });
+      const cas = createCasStore(join(root, "cas"));
+
+      const logger = createLogger({ sink: { kind: "file", path: infoPath } });
+      const ac = new AbortController();
+
+      const result = await executeThread(
+        demoWorkflow,
+        "demo-flow",
+        { prompt: "DAG test", steps: [] },
+        {
+          maxRounds: 5,
+          depth: 0,
+          signal: ac.signal,
+          awaitAfterEachYield: async () => {},
+          forkSourceThreadId: null,
+          prefilledDiskSteps: null,
+        },
+        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
+        logger,
+      );
+
+      const dataText = await readFile(dataPath, "utf8");
+      const lines = dataText
+        .trim()
+        .split("\n")
+        .filter((l) => l !== "");
+      expect(lines.length).toBe(3);
+
+      const rolePlanner = JSON.parse(lines[1] ?? "{}") as Record<string, unknown>;
+      const roleCoder = JSON.parse(lines[2] ?? "{}") as Record<string, unknown>;
+
+      const threadYaml = await cas.get(result.rootHash);
+      expect(threadYaml).not.toBeNull();
+      const threadNode = parseMerkleNode(threadYaml ?? "");
+      expect(threadNode.type).toBe("thread");
+
+      const bodies: string[] = [];
+      for (const stepHash of threadNode.children) {
+        const stepYaml = await cas.get(stepHash);
+        expect(stepYaml).not.toBeNull();
+        const stepNode = parseMerkleNode(stepYaml ?? "");
+        expect(stepNode.type).toBe("step");
+        expect(stepNode.children.length).toBe(1);
+        const contentHash = stepNode.children[0];
+        expect(contentHash).toBeDefined();
+        const body = await getContentMerklePayload(cas, contentHash ?? "");
+        expect(body).not.toBeNull();
+        bodies.push(body ?? "");
+      }
+
+      expect(bodies.sort()).toEqual(["code-body", "plan-body"].sort());
+      expect(rolePlanner.role).toBe("planner");
+      expect(roleCoder.role).toBe("coder");
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("extractMode react traverses CAS DAG via cas_get during extraction", async () => {
+    const dagMetaSchema = z.object({ leafPayload: z.string() });
+    type DagDemoMeta = { walker: z.infer<typeof dagMetaSchema> };
+
+    const origFetch = globalThis.fetch;
+    restoreFetch = () => {
+      globalThis.fetch = origFetch;
+    };
+    let fetchRound = 0;
+
+    const root = await mkdtemp(join(tmpdir(), "wf-engine-react-"));
+    try {
+      const cas = createCasStore(join(root, "cas"));
+      const leafYaml = serializeMerkleNode(createContentMerkleNode("needle-from-leaf"));
+      const leafHash = await cas.put(leafYaml);
+      const rootYaml = serializeMerkleNode({
+        type: "thread",
+        payload: {
+          workflow: "dag-demo",
+          threadId: "01DAG00000000000000000001",
+          result: { returnCode: 0, summary: "" },
+        },
+        children: [leafHash],
+      });
+      const dagRootHash = await cas.put(rootYaml);
+
+      globalThis.fetch = Object.assign(
+        async (_input: Parameters<typeof fetch>[0], _init?: RequestInit) => {
+          fetchRound += 1;
+          if (fetchRound === 1) {
+            return new Response(
+              JSON.stringify({
+                choices: [
+                  {
+                    message: {
+                      tool_calls: [
+                        {
+                          id: "c1",
+                          type: "function",
+                          function: {
+                            name: "cas_get",
+                            arguments: JSON.stringify({ hash: dagRootHash }),
+                          },
+                        },
+                      ],
+                    },
+                  },
+                ],
+              }),
+              { status: 200, headers: { "Content-Type": "application/json" } },
+            );
+          }
+          if (fetchRound === 2) {
+            return new Response(
+              JSON.stringify({
+                choices: [
+                  {
+                    message: {
+                      tool_calls: [
+                        {
+                          id: "c2",
+                          type: "function",
+                          function: {
+                            name: "cas_get",
+                            arguments: JSON.stringify({ hash: leafHash }),
+                          },
+                        },
+                      ],
+                    },
+                  },
+                ],
+              }),
+              { status: 200, headers: { "Content-Type": "application/json" } },
+            );
+          }
+          return new Response(
+            JSON.stringify({
+              choices: [
+                {
+                  message: {
+                    tool_calls: [
+                      {
+                        id: "c3",
+                        type: "function",
+                        function: {
+                          name: "extract",
+                          arguments: JSON.stringify({ leafPayload: "needle-from-leaf" }),
+                        },
+                      },
+                    ],
+                  },
+                },
+              ],
+            }),
+            { status: 200, headers: { "Content-Type": "application/json" } },
+          );
+        },
+        { preconnect: origFetch.preconnect.bind(origFetch) },
+      ) as typeof fetch;
+
+      const llm: LlmProvider = { baseUrl: "http://127.0.0.1:9", apiKey: "test", model: "test" };
+      const extractFn = createExtract(llm);
+
+      const dagWorkflow = createWorkflow<DagDemoMeta>(
+        {
+          roles: {
+            walker: {
+              description: "DAG walker",
+              systemPrompt: "Output only the root CAS hash.",
+              extractPrompt:
+                "Set leafPayload to the string payload of the content Merkle node under the root.",
+              schema: dagMetaSchema,
+              extractRefs: null,
+              extractMode: "react",
+            },
+          },
+          moderator: (ctx) => (ctx.steps.length === 0 ? "walker" : END),
+        },
+        { agent: async () => dagRootHash },
+        extractFn,
+        llm,
+      );
+
+      const threadId = "01KQXKW18CT8G75T53R8F4G7YG";
+      const hash = "C9NMV6V2TQT81";
+      const dataPath = join(root, "logs", hash, `${threadId}.data.jsonl`);
+      const infoPath = join(root, "logs", hash, `${threadId}.info.jsonl`);
+      await mkdir(join(root, "logs", hash), { recursive: true });
+
+      const logger = createLogger({ sink: { kind: "file", path: infoPath } });
+      const ac = new AbortController();
+
+      const result = await executeThread(
+        dagWorkflow,
+        "dag-demo",
+        { prompt: "traverse", steps: [] },
+        {
+          maxRounds: 5,
+          depth: 0,
+          signal: ac.signal,
+          awaitAfterEachYield: async () => {},
+          forkSourceThreadId: null,
+          prefilledDiskSteps: null,
+        },
+        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
+        logger,
+      );
+
+      expect(result.returnCode).toBe(0);
+      expect(fetchRound).toBe(3);
+
+      const dataText = await readFile(dataPath, "utf8");
+      const lines = dataText
+        .trim()
+        .split("\n")
+        .filter((l) => l !== "");
+      const roleRec = JSON.parse(lines[1] ?? "{}") as Record<string, unknown>;
+      expect(roleRec.role).toBe("walker");
+      expect(roleRec.meta).toEqual({ leafPayload: "needle-from-leaf" });
+    } finally {
+      globalThis.fetch = origFetch;
+      await rm(root, { recursive: true, force: true });
+    }
+  });
 });

packages/workflow/__tests__/extract-provider.test.ts

@@ -0,0 +1,87 @@
+import { describe, expect, test } from "bun:test";
+import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import { getExtractProvider } from "../src/extract-provider.js";
+
+describe("getExtractProvider", () => {
+  test("returns provider when config.extract is present", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-ext-prov-ok-"));
+    try {
+      await mkdir(root, { recursive: true });
+      await writeFile(
+        join(root, "workflow.yaml"),
+        `config:
+  maxDepth: 3
+  extract:
+    baseUrl: https://dashscope.aliyuncs.com/compatible-mode/v1
+    model: qwen-plus
+    apiKey: literal-key
+workflows: {}
+`,
+        "utf8",
+      );
+      const r = await getExtractProvider(root);
+      expect(r.ok).toBe(true);
+      if (!r.ok) {
+        return;
+      }
+      expect(r.value.baseUrl).toBe("https://dashscope.aliyuncs.com/compatible-mode/v1");
+      expect(r.value.model).toBe("qwen-plus");
+      expect(r.value.apiKey).toBe("literal-key");
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("errs when registry has no config section", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-ext-prov-missing-"));
+    try {
+      await mkdir(root, { recursive: true });
+      await writeFile(join(root, "workflow.yaml"), "workflows: {}\n", "utf8");
+      const r = await getExtractProvider(root);
+      expect(r.ok).toBe(false);
+      if (r.ok) {
+        return;
+      }
+      expect(r.error).toContain("no global config");
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("resolves apiKey from env at registry read time", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-ext-prov-env-"));
+    const prev = process.env.WF_GET_EXTRACT_PROVIDER_KEY;
+    process.env.WF_GET_EXTRACT_PROVIDER_KEY = "resolved-secret";
+    try {
+      await mkdir(root, { recursive: true });
+      await writeFile(
+        join(root, "workflow.yaml"),
+        `config:
+  maxDepth: 1
+  extract:
+    baseUrl: https://example.com
+    model: m
+    apiKey: env:WF_GET_EXTRACT_PROVIDER_KEY
+workflows: {}
+`,
+        "utf8",
+      );
+      const r = await getExtractProvider(root);
+      expect(r.ok).toBe(true);
+      if (!r.ok) {
+        return;
+      }
+      expect(r.value.apiKey).toBe("resolved-secret");
+    } finally {
+      if (prev === undefined) {
+        delete process.env.WF_GET_EXTRACT_PROVIDER_KEY;
+      } else {
+        process.env.WF_GET_EXTRACT_PROVIDER_KEY = prev;
+      }
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+});

packages/workflow/__tests__/fork-thread.test.ts

@@ -7,9 +7,9 @@ import {
 } from "../src/fork-thread.js";
 
 const sampleDataJsonl = `{"name":"demo","hash":"C9NMV6V2TQT81","threadId":"01AAA1111111111111111111","parameters":{"prompt":"hi","options":{"maxRounds":5}},"timestamp":100}
-{"role":"planner","content":"p","meta":{},"timestamp":101}
-{"role":"coder","content":"c","meta":{},"timestamp":102}
-{"role":"reviewer","content":"r","meta":{},"timestamp":103}
+{"role":"planner","contentHash":"HP0000000000000000000001","meta":{},"refs":[],"timestamp":101}
+{"role":"coder","contentHash":"HP0000000000000000000002","meta":{},"refs":[],"timestamp":102}
+{"role":"reviewer","contentHash":"HP0000000000000000000003","meta":{},"refs":[],"timestamp":103}
 `;
 
 describe("fork-thread", () => {
@@ -24,6 +24,7 @@ describe("fork-thread", () => {
     expect(r.value.start.threadId).toBe("01AAA1111111111111111111");
     expect(r.value.start.prompt).toBe("hi");
     expect(r.value.start.maxRounds).toBe(5);
+    expect(r.value.start.depth).toBe(0);
     expect(r.value.roleSteps.length).toBe(3);
     expect(r.value.roleSteps[0]?.role).toBe("planner");
   });
@@ -83,6 +84,44 @@ describe("fork-thread", () => {
     expect(r.value.workflowName).toBe("demo");
     expect(r.value.historicalSteps.length).toBe(1);
     expect(r.value.historicalSteps[0]?.timestamp).toBe(101);
-    expect(r.value.runOptions).toEqual({ maxRounds: 5 });
+    expect(r.value.runOptions).toEqual({ maxRounds: 5, depth: 0 });
+  });
+
+  test("parseThreadDataJsonl ignores trailing WorkflowResult line", () => {
+    const text = `${sampleDataJsonl.trim()}\n{"returnCode":0,"summary":"done"}\n`;
+    const r = parseThreadDataJsonl(text);
+    expect(r.ok).toBe(true);
+    if (!r.ok) {
+      return;
+    }
+    expect(r.value.roleSteps.length).toBe(3);
+    expect(r.value.roleSteps[2]?.role).toBe("reviewer");
+  });
+
+  test("parseThreadDataJsonl errors when WorkflowResult is not last", () => {
+    const text = `{"name":"demo","hash":"H","threadId":"01ZZZZZZZZZZZZZZZZZZZZZZ","parameters":{"prompt":"p","options":{"maxRounds":3}},"timestamp":1}
+{"returnCode":0,"summary":"early"}
+{"role":"planner","content":"x","meta":{},"timestamp":2}
+`;
+    const r = parseThreadDataJsonl(text);
+    expect(r.ok).toBe(false);
+  });
+
+  test("parseThreadDataJsonl reads explicit depth from start record", () => {
+    const text = `{"name":"demo","hash":"H","threadId":"01ZZZZZZZZZZZZZZZZZZZZZZ","parameters":{"prompt":"p","options":{"maxRounds":3,"depth":2}},"timestamp":1}
+{"role":"planner","contentHash":"HP0000000000000000000099","meta":{},"refs":[],"timestamp":2}
+`;
+    const r = parseThreadDataJsonl(text);
+    expect(r.ok).toBe(true);
+    if (!r.ok) {
+      return;
+    }
+    expect(r.value.start.depth).toBe(2);
+    const plan = buildForkPlan(text, null);
+    expect(plan.ok).toBe(true);
+    if (!plan.ok) {
+      return;
+    }
+    expect(plan.value.runOptions).toEqual({ maxRounds: 3, depth: 2 });
   });
 });

packages/workflow/__tests__/merkle.test.ts

@@ -0,0 +1,29 @@
+import { describe, expect, test } from "bun:test";
+
+import { createContentMerkleNode, parseMerkleNode, serializeMerkleNode } from "../src/merkle.js";
+
+describe("merkle", () => {
+  test("content node roundtrips through YAML", () => {
+    const node = createContentMerkleNode("hello\nworld");
+    const yaml = serializeMerkleNode(node);
+    const back = parseMerkleNode(yaml);
+    expect(back).toEqual(node);
+  });
+
+  test("step node with object payload roundtrips", () => {
+    const node = {
+      type: "step" as const,
+      payload: { role: "planner", foo: 1 },
+      children: ["ABC123", "DEF456"],
+    };
+    const yaml = serializeMerkleNode(node);
+    const back = parseMerkleNode(yaml);
+    expect(back.type).toBe("step");
+    expect(back.payload).toEqual({ role: "planner", foo: 1 });
+    expect(back.children).toEqual(["ABC123", "DEF456"]);
+  });
+
+  test("parse rejects invalid YAML root", () => {
+    expect(() => parseMerkleNode("[]")).toThrow();
+  });
+});

packages/workflow/__tests__/react-extract.test.ts

@@ -0,0 +1,209 @@
+import { afterEach, describe, expect, test } from "bun:test";
+import { mkdtemp, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import * as z from "zod/v4";
+
+import { createCasStore } from "../src/cas.js";
+import { createContentMerkleNode, serializeMerkleNode } from "../src/merkle.js";
+import { reactExtract } from "../src/react-extract.js";
+import type { LlmProvider } from "../src/types.js";
+
+const metaSchema = z.object({ seen: z.string() });
+
+const provider: LlmProvider = {
+  baseUrl: "http://127.0.0.1:9",
+  apiKey: "test",
+  model: "test",
+};
+
+describe("reactExtract", () => {
+  let restoreFetch: (() => void) | null = null;
+
+  afterEach(() => {
+    restoreFetch?.();
+    restoreFetch = null;
+  });
+
+  test("cas_get rounds then extract tool yields validated meta", async () => {
+    const casDir = await mkdtemp(join(tmpdir(), "react-extract-"));
+    const cas = createCasStore(casDir);
+    try {
+      const blob = serializeMerkleNode(createContentMerkleNode("needle"));
+      const h = await cas.put(blob);
+
+      const origFetch = globalThis.fetch;
+      let round = 0;
+      restoreFetch = () => {
+        globalThis.fetch = origFetch;
+      };
+      globalThis.fetch = Object.assign(
+        async (_input: Parameters<typeof fetch>[0], _init?: RequestInit) => {
+          round += 1;
+          if (round === 1) {
+            return new Response(
+              JSON.stringify({
+                choices: [
+                  {
+                    message: {
+                      tool_calls: [
+                        {
+                          id: "t1",
+                          type: "function",
+                          function: {
+                            name: "cas_get",
+                            arguments: JSON.stringify({ hash: h }),
+                          },
+                        },
+                      ],
+                    },
+                  },
+                ],
+              }),
+              { status: 200, headers: { "Content-Type": "application/json" } },
+            );
+          }
+          return new Response(
+            JSON.stringify({
+              choices: [
+                {
+                  message: {
+                    tool_calls: [
+                      {
+                        id: "t2",
+                        type: "function",
+                        function: {
+                          name: "extract",
+                          arguments: JSON.stringify({ seen: "needle" }),
+                        },
+                      },
+                    ],
+                  },
+                },
+              ],
+            }),
+            { status: 200, headers: { "Content-Type": "application/json" } },
+          );
+        },
+        { preconnect: origFetch.preconnect.bind(origFetch) },
+      ) as typeof fetch;
+
+      const text = `## Agent Output\n${h}\n## Extraction Instruction\nExtract seen from CAS.`;
+      const result = await reactExtract({
+        text,
+        schema: metaSchema,
+        provider,
+        cas,
+      });
+
+      expect(result.ok).toBe(true);
+      if (!result.ok) {
+        return;
+      }
+      expect(result.value).toEqual({ seen: "needle" });
+      expect(round).toBe(2);
+    } finally {
+      await rm(casDir, { recursive: true, force: true });
+    }
+  });
+
+  test("stops after max tool rounds when model keeps calling cas_get", async () => {
+    const casDir = await mkdtemp(join(tmpdir(), "react-extract-max-"));
+    const cas = createCasStore(casDir);
+    try {
+      const blob = serializeMerkleNode(createContentMerkleNode("x"));
+      const h = await cas.put(blob);
+
+      const origFetch = globalThis.fetch;
+      let round = 0;
+      restoreFetch = () => {
+        globalThis.fetch = origFetch;
+      };
+      globalThis.fetch = Object.assign(
+        async (_input: Parameters<typeof fetch>[0], _init?: RequestInit) => {
+          round += 1;
+          return new Response(
+            JSON.stringify({
+              choices: [
+                {
+                  message: {
+                    tool_calls: [
+                      {
+                        id: `loop-${round}`,
+                        type: "function",
+                        function: {
+                          name: "cas_get",
+                          arguments: JSON.stringify({ hash: h }),
+                        },
+                      },
+                    ],
+                  },
+                },
+              ],
+            }),
+            { status: 200, headers: { "Content-Type": "application/json" } },
+          );
+        },
+        { preconnect: origFetch.preconnect.bind(origFetch) },
+      ) as typeof fetch;
+
+      const result = await reactExtract({
+        text: "## Agent Output\nnoop\n## Extraction Instruction\nExtract seen.",
+        schema: metaSchema,
+        provider,
+        cas,
+      });
+
+      expect(result.ok).toBe(false);
+      if (result.ok) {
+        return;
+      }
+      expect(result.error).toBe("max_react_rounds_exceeded");
+      expect(round).toBe(10);
+    } finally {
+      await rm(casDir, { recursive: true, force: true });
+    }
+  });
+
+  test("passthrough JSON assistant message without tool calls", async () => {
+    const casDir = await mkdtemp(join(tmpdir(), "react-extract-pass-"));
+    const cas = createCasStore(casDir);
+    try {
+      const origFetch = globalThis.fetch;
+      restoreFetch = () => {
+        globalThis.fetch = origFetch;
+      };
+      globalThis.fetch = Object.assign(
+        async (_input: Parameters<typeof fetch>[0], _init?: RequestInit) =>
+          new Response(
+            JSON.stringify({
+              choices: [
+                {
+                  message: {
+                    content: '{"seen":"direct"}',
+                  },
+                },
+              ],
+            }),
+            { status: 200, headers: { "Content-Type": "application/json" } },
+          ),
+        { preconnect: origFetch.preconnect.bind(origFetch) },
+      ) as typeof fetch;
+
+      const result = await reactExtract({
+        text: "## Agent Output\nok\n## Extraction Instruction\nExtract.",
+        schema: metaSchema,
+        provider,
+        cas,
+      });
+
+      expect(result.ok).toBe(true);
+      if (!result.ok) {
+        return;
+      }
+      expect(result.value).toEqual({ seen: "direct" });
+    } finally {
+      await rm(casDir, { recursive: true, force: true });
+    }
+  });
+});

packages/workflow/__tests__/refs-tracking.test.ts

@@ -0,0 +1,203 @@
+import { afterEach, describe, expect, test } from "bun:test";
+import { mkdir, mkdtemp, readFile, rm } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import * as z from "zod/v4";
+
+import { createCasStore } from "../src/cas.js";
+import { createWorkflow } from "../src/create-workflow.js";
+import { executeThread } from "../src/engine.js";
+import { createExtract } from "../src/extract-fn.js";
+import { buildForkPlan, parseThreadDataJsonl } from "../src/fork-thread.js";
+import { createLogger } from "../src/logger.js";
+import { END } from "../src/types.js";
+
+const phaseSchema = z.object({
+  hash: z.string(),
+  title: z.string(),
+});
+
+const plannerMetaSchema = z.object({
+  phases: z.array(phaseSchema),
+});
+
+type RefsDemoMeta = {
+  planner: z.infer<typeof plannerMetaSchema>;
+};
+
+function installMockChatCompletions(sequence: ReadonlyArray<Record<string, unknown>>): () => void {
+  const origFetch = globalThis.fetch;
+  let i = 0;
+  const mockFetch = async (
+    input: Parameters<typeof fetch>[0],
+    init?: RequestInit,
+  ): Promise<Response> => {
+    const args = sequence[i] ?? sequence[sequence.length - 1];
+    if (args === undefined) {
+      throw new Error("installMockChatCompletions: empty sequence");
+    }
+    i += 1;
+    void input;
+    const body = init?.body ? (JSON.parse(String(init.body)) as Record<string, unknown>) : {};
+    const tools = body.tools;
+    const firstTool =
+      Array.isArray(tools) && tools.length > 0 && tools[0] !== null && typeof tools[0] === "object"
+        ? (tools[0] as Record<string, unknown>)
+        : null;
+    const fn =
+      firstTool !== null ? (firstTool.function as Record<string, unknown> | undefined) : undefined;
+    const toolName = typeof fn?.name === "string" ? fn.name : "extract";
+    return new Response(
+      JSON.stringify({
+        choices: [
+          {
+            message: {
+              tool_calls: [
+                {
+                  type: "function",
+                  function: {
+                    name: toolName,
+                    arguments: JSON.stringify(args),
+                  },
+                },
+              ],
+            },
+          },
+        ],
+      }),
+      { status: 200, headers: { "Content-Type": "application/json" } },
+    );
+  };
+  globalThis.fetch = Object.assign(mockFetch, {
+    preconnect: origFetch.preconnect.bind(origFetch),
+  }) as typeof fetch;
+  return () => {
+    globalThis.fetch = origFetch;
+  };
+}
+
+const refsDemoExtract = createExtract({
+  baseUrl: "http://127.0.0.1:9",
+  apiKey: "test",
+  model: "test",
+});
+
+const refsDemoWorkflow = createWorkflow<RefsDemoMeta>(
+  {
+    roles: {
+      planner: {
+        description: "Planner with phase hashes",
+        systemPrompt: "Plan.",
+        extractPrompt: "Extract phases with CAS hashes.",
+        schema: plannerMetaSchema,
+        extractRefs: (meta) => meta.phases.map((p) => p.hash),
+        extractMode: "single",
+      },
+    },
+    moderator: (ctx) => (ctx.steps.length === 0 ? "planner" : END),
+  },
+  {
+    agent: async () => "plan-output",
+  },
+  refsDemoExtract,
+  null,
+);
+
+describe("RoleStep refs tracking", () => {
+  let restoreFetch: (() => void) | null = null;
+
+  afterEach(() => {
+    restoreFetch?.();
+    restoreFetch = null;
+  });
+
+  test("parseThreadDataJsonl reads refs and defaults missing refs to []", () => {
+    const text = `{"name":"demo","hash":"C9NMV6V2TQT81","threadId":"01AAA1111111111111111111","parameters":{"prompt":"hi","options":{"maxRounds":5}},"timestamp":100}
+{"role":"planner","contentHash":"HPAYLOAD111111","meta":{},"refs":["H111AAAAAAAAA","H222AAAAAAAAA"],"timestamp":101}
+{"role":"coder","contentHash":"HPAYLOAD222222","meta":{},"timestamp":102}
+`;
+    const r = parseThreadDataJsonl(text);
+    expect(r.ok).toBe(true);
+    if (!r.ok) {
+      return;
+    }
+    expect(r.value.roleSteps[0]?.refs).toEqual(["H111AAAAAAAAA", "H222AAAAAAAAA"]);
+    expect(r.value.roleSteps[1]?.refs).toEqual([]);
+  });
+
+  test("executeThread persists refs from extractRefs on role yields", async () => {
+    restoreFetch = installMockChatCompletions([
+      {
+        phases: [
+          { hash: "C9NMV6V2TQT81", title: "phase-a" },
+          { hash: "C9NMV6V2TQT82", title: "phase-b" },
+        ],
+      },
+    ]);
+
+    const root = await mkdtemp(join(tmpdir(), "wf-refs-"));
+    try {
+      const threadId = "01KQXKW18CT8G75T53R8F4G7YG";
+      const hash = "C9NMV6V2TQT81";
+      const dataPath = join(root, "logs", hash, `${threadId}.data.jsonl`);
+      const infoPath = join(root, "logs", hash, `${threadId}.info.jsonl`);
+      await mkdir(join(root, "logs", hash), { recursive: true });
+      const cas = createCasStore(join(root, "cas"));
+
+      const logger = createLogger({ sink: { kind: "file", path: infoPath } });
+      const ac = new AbortController();
+
+      const result = await executeThread(
+        refsDemoWorkflow,
+        "refs-demo",
+        { prompt: "task", steps: [] },
+        {
+          maxRounds: 5,
+          depth: 0,
+          signal: ac.signal,
+          awaitAfterEachYield: async () => {},
+          forkSourceThreadId: null,
+          prefilledDiskSteps: null,
+        },
+        { threadId, hash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
+        logger,
+      );
+
+      expect(result.returnCode).toBe(0);
+      expect(typeof result.rootHash).toBe("string");
+      expect(result.rootHash.length).toBeGreaterThan(0);
+
+      const dataText = await readFile(dataPath, "utf8");
+      const lines = dataText
+        .trim()
+        .split("\n")
+        .filter((l) => l !== "");
+      expect(lines.length).toBe(2);
+
+      const role1 = JSON.parse(lines[1] ?? "{}") as Record<string, unknown>;
+      expect(role1.role).toBe("planner");
+      const refs = role1.refs as string[];
+      expect(refs).toContain("C9NMV6V2TQT81");
+      expect(refs).toContain("C9NMV6V2TQT82");
+      expect(typeof role1.contentHash).toBe("string");
+      expect(refs).toContain(String(role1.contentHash));
+      expect(refs.length).toBe(3);
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("buildForkPlan carries refs on historical steps", () => {
+    const text = `{"name":"demo","hash":"C9NMV6V2TQT81","threadId":"01AAA1111111111111111111","parameters":{"prompt":"hi","options":{"maxRounds":5}},"timestamp":100}
+{"role":"planner","contentHash":"HP111111111111","meta":{},"refs":["KEEPREFAAAAAA"],"timestamp":101}
+{"role":"coder","contentHash":"HP222222222222","meta":{},"refs":["CODERHASHAAAA"],"timestamp":102}
+`;
+    const plan = buildForkPlan(text, null);
+    expect(plan.ok).toBe(true);
+    if (!plan.ok) {
+      return;
+    }
+    expect(plan.value.historicalSteps.length).toBe(1);
+    expect(plan.value.historicalSteps[0]?.refs).toEqual(["KEEPREFAAAAAA"]);
+  });
+});

packages/workflow/__tests__/registry.test.ts

@@ -4,6 +4,7 @@ import { tmpdir } from "node:os";
 import { join } from "node:path";
 
 import {
+  parseWorkflowRegistryYaml,
   readWorkflowRegistry,
   registerWorkflowVersion,
   rollbackWorkflowToHistoryHash,
@@ -21,6 +22,7 @@ describe("workflow registry", () => {
     if (!empty.ok) {
       return;
     }
+    expect(empty.value.config).toBeNull();
 
     const r1 = registerWorkflowVersion(empty.value, "solve-issue", "AAAAAAAAAAAAA", 100);
     const w1 = await writeWorkflowRegistry(dir, r1);
@@ -68,7 +70,7 @@ describe("workflow registry", () => {
   });
 
   test("rollbackWorkflowToHistoryHash swaps head with a prior version", () => {
-    let reg = registerWorkflowVersion({ workflows: {} }, "solve-issue", "H1", 100);
+    let reg = registerWorkflowVersion({ config: null, workflows: {} }, "solve-issue", "H1", 100);
     reg = registerWorkflowVersion(reg, "solve-issue", "H2", 200);
     reg = registerWorkflowVersion(reg, "solve-issue", "H3", 300);
     const entry = reg.workflows["solve-issue"];
@@ -99,6 +101,85 @@ describe("workflow registry", () => {
     expect(bad.ok).toBe(false);
   });
 
+  test("parses config section and literal apiKey", () => {
+    const yaml = `
+config:
+  maxDepth: 3
+  extract:
+    baseUrl: https://example.com/v1
+    model: qwen-plus
+    apiKey: secret-key
+workflows:
+  solve-issue:
+    hash: SPVR4BDMSGC1W
+    timestamp: 1
+    history: []
+`;
+    const r = parseWorkflowRegistryYaml(yaml);
+    expect(r.ok).toBe(true);
+    if (!r.ok) {
+      return;
+    }
+    expect(r.value.config).not.toBeNull();
+    if (r.value.config === null) {
+      return;
+    }
+    expect(r.value.config.maxDepth).toBe(3);
+    expect(r.value.config.extract.baseUrl).toBe("https://example.com/v1");
+    expect(r.value.config.extract.model).toBe("qwen-plus");
+    expect(r.value.config.extract.apiKey).toBe("secret-key");
+  });
+
+  test("parses config apiKey env: prefix from process.env", () => {
+    const prev = process.env.WF_REGISTRY_TEST_API_KEY;
+    process.env.WF_REGISTRY_TEST_API_KEY = "from-env";
+    try {
+      const yaml = `
+config:
+  maxDepth: 1
+  extract:
+    baseUrl: https://dashscope.aliyuncs.com/compatible-mode/v1
+    model: qwen-plus
+    apiKey: env:WF_REGISTRY_TEST_API_KEY
+workflows: {}
+`;
+      const r = parseWorkflowRegistryYaml(yaml);
+      expect(r.ok).toBe(true);
+      if (!r.ok) {
+        return;
+      }
+      expect(r.value.config?.extract.apiKey).toBe("from-env");
+    } finally {
+      if (prev === undefined) {
+        delete process.env.WF_REGISTRY_TEST_API_KEY;
+      } else {
+        process.env.WF_REGISTRY_TEST_API_KEY = prev;
+      }
+    }
+  });
+
+  test("parse errors when env: apiKey variable is unset", () => {
+    const prev = process.env.WF_REGISTRY_TEST_API_KEY_UNSET;
+    delete process.env.WF_REGISTRY_TEST_API_KEY_UNSET;
+    try {
+      const yaml = `
+config:
+  maxDepth: 1
+  extract:
+    baseUrl: https://example.com
+    model: m
+    apiKey: env:WF_REGISTRY_TEST_API_KEY_UNSET
+workflows: {}
+`;
+      const r = parseWorkflowRegistryYaml(yaml);
+      expect(r.ok).toBe(false);
+    } finally {
+      if (prev !== undefined) {
+        process.env.WF_REGISTRY_TEST_API_KEY_UNSET = prev;
+      }
+    }
+  });
+
   test("parse errors on invalid shape", async () => {
     const dir = join(tmpdir(), `wf-reg3-${process.pid}-${Date.now()}`);
     await mkdir(dir, { recursive: true });

packages/workflow/__tests__/storage-root.test.ts

@@ -0,0 +1,14 @@
+import { describe, expect, test } from "bun:test";
+import { join } from "node:path";
+
+import { getDefaultWorkflowStorageRoot, getGlobalCasDir } from "../src/storage-root.js";
+
+describe("getGlobalCasDir", () => {
+  test("joins cas segment under explicit storage root", () => {
+    expect(getGlobalCasDir("/tmp/wf-root")).toBe(join("/tmp/wf-root", "cas"));
+  });
+
+  test("defaults to default workflow root when storage root is undefined", () => {
+    expect(getGlobalCasDir(undefined)).toBe(join(getDefaultWorkflowStorageRoot(), "cas"));
+  });
+});

packages/workflow/__tests__/thread-jsonl-format.test.ts

@@ -10,6 +10,7 @@ describe("RFC-001 thread JSONL shapes", () => {
         prompt: "Fix the login redirect bug in #3",
         options: {
           maxRounds: 5,
+          depth: 0,
         },
       },
       timestamp: 1714963200000,
@@ -17,15 +18,18 @@ describe("RFC-001 thread JSONL shapes", () => {
 
     const roleRecord = {
       role: "planner",
-      content: "Plan: modify auth middleware...",
+      contentHash: "CPHASH000000000000000001",
       meta: { plan: "...", files: ["src/auth.ts"] },
+      refs: [] as string[],
       timestamp: 1714963201000,
     };
 
     expect(Object.keys(startRecord).sort()).toEqual(
       ["hash", "name", "parameters", "threadId", "timestamp"].sort(),
     );
-    expect(Object.keys(roleRecord).sort()).toEqual(["content", "meta", "role", "timestamp"].sort());
+    expect(Object.keys(roleRecord).sort()).toEqual(
+      ["contentHash", "meta", "refs", "role", "timestamp"].sort(),
+    );
   });
 
   test("documents the `.info.jsonl` debug record keys", () => {

packages/workflow/__tests__/worker.test.ts

@@ -5,22 +5,29 @@ import { createConnection } from "node:net";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 
+import { createCasStore } from "../src/cas.js";
+import { createContentMerkleNode, serializeMerkleNode } from "../src/merkle.js";
 import { getWorkerHostScriptPath } from "../src/worker-entry-path.js";
 
-const bundleSource = `export const descriptor = {
+const bundleSource = `import { putContentMerkleNode } from "@uncaged/workflow";
+
+export const descriptor = {
   description: "worker-test",
   roles: {
     planner: { description: "planner", schema: {} },
     coder: { description: "coder", schema: {} },
   },
 };
-export const run = async function* (input) {
+export const run = async function* (input, options) {
+  const cas = options.cas;
   const has = (r) => input.steps.some((s) => s.role === r);
   if (!has("planner")) {
-    yield { role: "planner", content: "p", meta: { plan: input.prompt } };
+    const h = await putContentMerkleNode(cas, "p");
+    yield { role: "planner", contentHash: h, meta: { plan: input.prompt }, refs: [h] };
   }
   if (!has("coder")) {
-    yield { role: "coder", content: "c", meta: { diff: "y" } };
+    const h = await putContentMerkleNode(cas, "c");
+    yield { role: "coder", contentHash: h, meta: { diff: "y" }, refs: [h] };
   }
   return { returnCode: 0, summary: "completed: moderator returned END" };
 };
@@ -102,7 +109,7 @@ describe("worker process", () => {
         threadId,
         workflowName: "demo-flow",
         prompt: "hello",
-        options: { maxRounds: 5 },
+        options: { maxRounds: 5, depth: 0 },
       });
 
       const exitCode: number = await new Promise((resolve) => {
@@ -118,7 +125,7 @@ describe("worker process", () => {
           .trim()
           .split("\n")
           .filter((l) => l !== "").length,
-      ).toBe(3);
+      ).toBe(4);
     } finally {
       await rm(root, { recursive: true, force: true });
     }
@@ -143,6 +150,11 @@ describe("worker process", () => {
 
       const port = await readReadyPort(child);
 
+      const cas = createCasStore(join(root, "cas"));
+      const plannerReplayHash = await cas.put(
+        serializeMerkleNode(createContentMerkleNode("p-old")),
+      );
+
       const threadId = "01KQXKW18CT8G75T53R8F4G7YG";
       const srcId = "01SRCMMMMMMMMMMMMMMMMMMMM";
       await sendJson(port, {
@@ -150,12 +162,13 @@ describe("worker process", () => {
         threadId,
         workflowName: "demo-flow",
         prompt: "hello",
-        options: { maxRounds: 5 },
+        options: { maxRounds: 5, depth: 0 },
         steps: [
           {
             role: "planner",
-            content: "p-old",
+            contentHash: plannerReplayHash,
             meta: { plan: "z" },
+            refs: [plannerReplayHash],
             timestamp: 555,
           },
         ],
@@ -174,7 +187,7 @@ describe("worker process", () => {
         .trim()
         .split("\n")
         .filter((l) => l !== "");
-      expect(lines.length).toBe(3);
+      expect(lines.length).toBe(4);
       const start = JSON.parse(lines[0] ?? "{}") as Record<string, unknown>;
       expect(start.forkFrom).toEqual({ threadId: srcId });
       const replay = JSON.parse(lines[1] ?? "{}") as Record<string, unknown>;
@@ -182,6 +195,8 @@ describe("worker process", () => {
       expect(replay.timestamp).toBe(555);
       const coder = JSON.parse(lines[2] ?? "{}") as Record<string, unknown>;
       expect(coder.role).toBe("coder");
+      const done = JSON.parse(lines[3] ?? "{}") as Record<string, unknown>;
+      expect(done.returnCode).toBe(0);
     } finally {
       await rm(root, { recursive: true, force: true });
     }

packages/workflow/__tests__/workflow-as-agent-integration.test.ts

@@ -0,0 +1,225 @@
+import { afterEach, describe, expect, test } from "bun:test";
+import { mkdir, mkdtemp, readdir, readFile, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import * as z from "zod/v4";
+
+import { createCasStore } from "../src/cas.js";
+import { createWorkflow } from "../src/create-workflow.js";
+import { executeThread } from "../src/engine.js";
+import { createExtract } from "../src/extract-fn.js";
+import { hashWorkflowBundleBytes } from "../src/hash.js";
+import { createLogger } from "../src/logger.js";
+import { getContentMerklePayload, parseMerkleNode } from "../src/merkle.js";
+import {
+  readWorkflowRegistry,
+  registerWorkflowVersion,
+  writeWorkflowRegistry,
+} from "../src/registry.js";
+import { END } from "../src/types.js";
+import { workflowAsAgent } from "../src/workflow-as-agent.js";
+
+const callerMetaSchema = z.object({ done: z.literal(true) });
+
+type ParentMeta = {
+  caller: z.infer<typeof callerMetaSchema>;
+};
+
+function installMockChatCompletions(sequence: ReadonlyArray<Record<string, unknown>>): () => void {
+  const origFetch = globalThis.fetch;
+  let i = 0;
+  const mockFetch = async (
+    input: Parameters<typeof fetch>[0],
+    init?: RequestInit,
+  ): Promise<Response> => {
+    const args = sequence[i] ?? sequence[sequence.length - 1];
+    if (args === undefined) {
+      throw new Error("installMockChatCompletions: empty sequence");
+    }
+    i += 1;
+    void input;
+    const body = init?.body ? (JSON.parse(String(init.body)) as Record<string, unknown>) : {};
+    const tools = body.tools;
+    const firstTool =
+      Array.isArray(tools) && tools.length > 0 && tools[0] !== null && typeof tools[0] === "object"
+        ? (tools[0] as Record<string, unknown>)
+        : null;
+    const fn =
+      firstTool !== null ? (firstTool.function as Record<string, unknown> | undefined) : undefined;
+    const toolName = typeof fn?.name === "string" ? fn.name : "extract";
+    return new Response(
+      JSON.stringify({
+        choices: [
+          {
+            message: {
+              tool_calls: [
+                {
+                  type: "function",
+                  function: {
+                    name: toolName,
+                    arguments: JSON.stringify(args),
+                  },
+                },
+              ],
+            },
+          },
+        ],
+      }),
+      { status: 200, headers: { "Content-Type": "application/json" } },
+    );
+  };
+  globalThis.fetch = Object.assign(mockFetch, {
+    preconnect: origFetch.preconnect.bind(origFetch),
+  }) as typeof fetch;
+  return () => {
+    globalThis.fetch = origFetch;
+  };
+}
+
+const parentExtract = createExtract({
+  baseUrl: "http://127.0.0.1:9",
+  apiKey: "test",
+  model: "test",
+});
+
+const childBundleSource = `import { putContentMerkleNode } from "@uncaged/workflow";
+
+export const descriptor = {
+  description: "child-integration",
+  roles: {
+    agent: {
+      description: "agent",
+      schema: { type: "object", properties: {}, additionalProperties: true },
+    },
+  },
+};
+export async function* run(input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "child-body");
+  yield { role: "agent", contentHash: h, meta: {}, refs: [h] };
+  return { returnCode: 0, summary: "child-done:" + input.prompt };
+}
+`;
+
+async function installChildWorkflow(storageRoot: string): Promise<{ hash: string }> {
+  const bytes = new TextEncoder().encode(childBundleSource);
+  const hash = hashWorkflowBundleBytes(bytes);
+  await mkdir(join(storageRoot, "bundles"), { recursive: true });
+  await writeFile(join(storageRoot, "bundles", `${hash}.esm.js`), childBundleSource, "utf8");
+  const reg = await readWorkflowRegistry(storageRoot);
+  if (!reg.ok) {
+    throw reg.error;
+  }
+  const next = registerWorkflowVersion(reg.value, "child-wf", hash, Date.now());
+  const wr = await writeWorkflowRegistry(storageRoot, next);
+  if (!wr.ok) {
+    throw wr.error;
+  }
+  return { hash };
+}
+
+describe("workflowAsAgent integration", () => {
+  let restoreFetch: (() => void) | null = null;
+
+  afterEach(() => {
+    restoreFetch?.();
+    restoreFetch = null;
+  });
+
+  test("createWorkflow parent invokes nested workflow via workflowAsAgent", async () => {
+    restoreFetch = installMockChatCompletions([{ done: true }]);
+
+    const root = await mkdtemp(join(tmpdir(), "wf-waa-int-"));
+    try {
+      const { hash: childHash } = await installChildWorkflow(root);
+
+      const parentWorkflow = createWorkflow<ParentMeta>(
+        {
+          roles: {
+            caller: {
+              description: "delegates to child workflow",
+              systemPrompt: "system",
+              extractPrompt: "extract done flag",
+              schema: callerMetaSchema,
+              extractRefs: null,
+              extractMode: "single",
+            },
+          },
+          moderator: (ctx) => (ctx.steps.length === 0 ? "caller" : END),
+        },
+        { agent: workflowAsAgent("child-wf", { storageRoot: root }) },
+        parentExtract,
+        null,
+      );
+
+      const threadId = "01KQXKW18CT8G75T53R8F4G7YG";
+      const parentHash = "C9NMV6V2TQT81";
+      const dataPath = join(root, "logs", parentHash, `${threadId}.data.jsonl`);
+      const infoPath = join(root, "logs", parentHash, `${threadId}.info.jsonl`);
+      await mkdir(join(root, "logs", parentHash), { recursive: true });
+      const cas = createCasStore(join(root, "cas"));
+
+      const logger = createLogger({ sink: { kind: "file", path: infoPath } });
+      const ac = new AbortController();
+
+      const result = await executeThread(
+        parentWorkflow,
+        "parent-wf",
+        { prompt: "from-parent", steps: [] },
+        {
+          maxRounds: 5,
+          depth: 0,
+          signal: ac.signal,
+          awaitAfterEachYield: async () => {},
+          forkSourceThreadId: null,
+          prefilledDiskSteps: null,
+        },
+        { threadId, hash: parentHash, dataJsonlPath: dataPath, infoJsonlPath: infoPath, cas },
+        logger,
+      );
+
+      expect(result.returnCode).toBe(0);
+      expect(typeof result.rootHash).toBe("string");
+
+      const parentText = await readFile(dataPath, "utf8");
+      const parentLines = parentText
+        .trim()
+        .split("\n")
+        .filter((l) => l !== "");
+      expect(parentLines.length).toBe(2);
+      const callerLine = JSON.parse(parentLines[1] ?? "{}") as Record<string, unknown>;
+      expect(callerLine.role).toBe("caller");
+      const childRootHash = await getContentMerklePayload(cas, String(callerLine.contentHash));
+      expect(childRootHash).not.toBeNull();
+      const childThreadYaml = await cas.get(childRootHash ?? "");
+      expect(childThreadYaml).not.toBeNull();
+      const childThreadNode = parseMerkleNode(childThreadYaml ?? "");
+      expect(childThreadNode.type).toBe("thread");
+      const childPayload = childThreadNode.payload as Record<string, unknown>;
+      expect(childPayload.workflow).toBe("child-wf");
+      const childResult = childPayload.result as Record<string, unknown>;
+      expect(childResult.summary).toBe("child-done:from-parent");
+
+      const childDir = join(root, "logs", childHash);
+      const childFiles = await readdir(childDir);
+      const childDataName = childFiles.find((n) => n.endsWith(".data.jsonl"));
+      expect(childDataName).toBeDefined();
+
+      const childText = await readFile(join(childDir, childDataName ?? ""), "utf8");
+      const childStart = JSON.parse(
+        childText
+          .trim()
+          .split("\n")
+          .filter((l) => l !== "")[0] ?? "{}",
+      ) as Record<string, unknown>;
+      expect(childStart.forkFrom).toEqual({ threadId });
+      const childOpts = (childStart.parameters as Record<string, unknown>).options as Record<
+        string,
+        unknown
+      >;
+      expect(childOpts.depth).toBe(1);
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+});

packages/workflow/__tests__/workflow-as-agent.test.ts

@@ -0,0 +1,168 @@
+import { describe, expect, test } from "bun:test";
+import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import { createCasStore } from "../src/cas.js";
+import { hashWorkflowBundleBytes } from "../src/hash.js";
+import { parseMerkleNode } from "../src/merkle.js";
+import {
+  readWorkflowRegistry,
+  registerWorkflowVersion,
+  writeWorkflowRegistry,
+} from "../src/registry.js";
+import { type AgentContext, START } from "../src/types.js";
+import { workflowAsAgent } from "../src/workflow-as-agent.js";
+
+function makeAgentCtx(params: {
+  storageRoot: string;
+  depth: number;
+  prompt: string;
+  maxRounds: number;
+}): AgentContext {
+  const ts = Date.now();
+  return {
+    threadId: "01PARENT000000000000000001AA",
+    depth: params.depth,
+    start: {
+      role: START,
+      content: params.prompt,
+      meta: { maxRounds: params.maxRounds },
+      timestamp: ts,
+    },
+    steps: [],
+    currentRole: {
+      name: "caller",
+      systemPrompt: "caller",
+    },
+    cas: createCasStore(join(params.storageRoot, "agent-ctx-cas")),
+  };
+}
+
+const childBundleSource = `import { putContentMerkleNode } from "@uncaged/workflow";
+
+export const descriptor = {
+  description: "child-test",
+  roles: {
+    agent: {
+      description: "agent",
+      schema: { type: "object", properties: {}, additionalProperties: true },
+    },
+  },
+};
+export async function* run(input, options) {
+  const cas = options.cas;
+  const h = await putContentMerkleNode(cas, "child-body");
+  yield { role: "agent", contentHash: h, meta: {}, refs: [h] };
+  return { returnCode: 0, summary: "child-done:" + input.prompt };
+}
+`;
+
+async function installChildWorkflow(storageRoot: string): Promise<{ hash: string }> {
+  const bytes = new TextEncoder().encode(childBundleSource);
+  const hash = hashWorkflowBundleBytes(bytes);
+  await mkdir(join(storageRoot, "bundles"), { recursive: true });
+  await writeFile(join(storageRoot, "bundles", `${hash}.esm.js`), childBundleSource, "utf8");
+  const reg = await readWorkflowRegistry(storageRoot);
+  if (!reg.ok) {
+    throw reg.error;
+  }
+  const next = registerWorkflowVersion(reg.value, "child-wf", hash, Date.now());
+  const wr = await writeWorkflowRegistry(storageRoot, next);
+  if (!wr.ok) {
+    throw wr.error;
+  }
+  return { hash };
+}
+
+describe("workflowAsAgent", () => {
+  test("returns error when workflow name is not registered", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-waa-missing-"));
+    try {
+      const agent = workflowAsAgent("missing-wf", { storageRoot: root });
+      const out = await agent(
+        makeAgentCtx({ storageRoot: root, depth: 0, prompt: "x", maxRounds: 5 }),
+      );
+      expect(out).toContain("not found in registry");
+      expect(out).toContain("missing-wf");
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("runs registered workflow and returns child thread root CAS hash", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-waa-ok-"));
+    try {
+      await installChildWorkflow(root);
+      const agent = workflowAsAgent("child-wf", { storageRoot: root });
+      const out = await agent(
+        makeAgentCtx({ storageRoot: root, depth: 0, prompt: "hello-parent", maxRounds: 5 }),
+      );
+      const cas = createCasStore(join(root, "cas"));
+      const threadYaml = await cas.get(out);
+      expect(threadYaml).not.toBeNull();
+      const node = parseMerkleNode(threadYaml ?? "");
+      expect(node.type).toBe("thread");
+      const payload = node.payload as Record<string, unknown>;
+      expect(payload.workflow).toBe("child-wf");
+      const resultObj = payload.result as Record<string, unknown>;
+      expect(resultObj.summary).toBe("child-done:hello-parent");
+      expect(node.children.length).toBe(1);
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("enforces depth limit (returns error string, does not throw)", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-waa-depth-"));
+    try {
+      const agent = workflowAsAgent("child-wf", { storageRoot: root });
+      const out = await agent(
+        makeAgentCtx({ storageRoot: root, depth: 3, prompt: "x", maxRounds: 5 }),
+      );
+      expect(out).toContain("depth limit");
+      expect(out).toContain("max 3");
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+
+  test("uses registry config maxDepth when set", async () => {
+    const root = await mkdtemp(join(tmpdir(), "wf-waa-maxdepth-cfg-"));
+    try {
+      await installChildWorkflow(root);
+      const reg = await readWorkflowRegistry(root);
+      expect(reg.ok).toBe(true);
+      if (!reg.ok) {
+        return;
+      }
+      const withCfg = {
+        ...reg.value,
+        config: {
+          maxDepth: 2,
+          extract: {
+            baseUrl: "http://127.0.0.1:9",
+            model: "m",
+            apiKey: "k",
+          },
+        },
+      };
+      const wr = await writeWorkflowRegistry(root, withCfg);
+      expect(wr.ok).toBe(true);
+
+      const agent = workflowAsAgent("child-wf", { storageRoot: root });
+      const okOut = await agent(
+        makeAgentCtx({ storageRoot: root, depth: 1, prompt: "nest-once", maxRounds: 5 }),
+      );
+      expect(okOut).not.toContain("depth limit");
+
+      const badOut = await agent(
+        makeAgentCtx({ storageRoot: root, depth: 2, prompt: "x", maxRounds: 5 }),
+      );
+      expect(badOut).toContain("depth limit");
+      expect(badOut).toContain("max 2");
+    } finally {
+      await rm(root, { recursive: true, force: true });
+    }
+  });
+});

packages/workflow/src/bundle-import-env.ts

@@ -0,0 +1,8 @@
+import { pathToFileURL } from "node:url";
+
+/**
+ * Dynamic-import a workflow bundle path (see {@link extractBundleExports} — symlink must exist first).
+ */
+export async function importWorkflowBundleModule(bundlePath: string): Promise<unknown> {
+  return import(pathToFileURL(bundlePath).href);
+}

packages/workflow/src/bundle-validator.ts

@@ -41,9 +41,12 @@ function isAllowedImportSpecifier(spec: string): boolean {
   if (spec.length === 0) {
     return false;
   }
-  if (spec.startsWith(".") || spec.startsWith("/")) {
+  if (spec.startsWith(".") || spec.startsWith("/") || spec.startsWith("file:")) {
     return false;
   }
+  if (spec === "@uncaged/workflow") {
+    return true;
+  }
   return isBuiltin(spec);
 }
 
@@ -297,7 +300,7 @@ function validateImportDeclaration(node: ImportDeclaration): string | null {
     return "only static string import specifiers are allowed";
   }
   if (!isAllowedImportSpecifier(spec)) {
-    return `disallowed import specifier "${spec}" (only Node built-ins are allowed)`;
+    return `disallowed import specifier "${spec}" (only Node built-ins and "@uncaged/workflow" are allowed)`;
   }
   return null;
 }
@@ -312,7 +315,7 @@ function validateExportSource(
     return staticMessage;
   }
   if (!isAllowedImportSpecifier(spec)) {
-    return `${disallowedPrefix} "${spec}" (only Node built-ins are allowed)`;
+    return `${disallowedPrefix} "${spec}" (only Node built-ins and "@uncaged/workflow" are allowed)`;
   }
   return null;
 }
@@ -365,7 +368,7 @@ function bundleConstraintViolationForNode(node: Node): string | null {
 
 /**
  * Validate RFC-001 bundle rules: single-file ESM shape, named exports `run` + `descriptor`,
- * no default export, no dynamic `import()`, static imports restricted to Node builtins.
+ * no default export, no dynamic `import()`, static imports restricted to Node builtins plus `@uncaged/workflow`.
  */
 export function validateWorkflowBundle(input: WorkflowBundleValidationInput): Result<void, string> {
   if (!endsWithEsmJs(input.filePath)) {

packages/workflow/src/cas.ts

@@ -10,7 +10,7 @@ export type CasStore = {
   list(): Promise<string[]>;
 };
 
-export function createThreadCas(casDir: string): CasStore {
+export function createCasStore(casDir: string): CasStore {
   async function ensureDir(): Promise<void> {
     await mkdir(casDir, { recursive: true });
   }
@@ -68,3 +68,6 @@ export function createThreadCas(casDir: string): CasStore {
     },
   };
 }
+
+/** @deprecated Use {@link createCasStore} — CAS is global, not per-thread. */
+export const createThreadCas = createCasStore;

packages/workflow/src/create-workflow.ts

@@ -1,19 +1,25 @@
-import type { ExtractFn } from "./extract-fn.js";
+import type { CasStore } from "./cas.js";
+import { buildExtractUserContent, type ExtractFn } from "./extract-fn.js";
+import { putContentMerkleNode } from "./merkle.js";
+import { reactExtract } from "./react-extract.js";
+import { mergeRefsWithContentHash } from "./refs-field.js";
 import {
   type AgentBinding,
   type AgentContext,
   END,
   type ExtractContext,
+  type LlmProvider,
   type ModeratorContext,
+  type RoleDefinition,
   type RoleMeta,
   type RoleOutput,
   type RoleStep,
   START,
   type ThreadInput,
+  type WorkflowCompletion,
   type WorkflowDefinition,
   type WorkflowFn,
   type WorkflowFnOptions,
-  type WorkflowResult,
 } from "./types.js";
 
 function isRoleNext<M extends RoleMeta>(
@@ -22,19 +28,67 @@ function isRoleNext<M extends RoleMeta>(
   return next !== END;
 }
 
+function resolveExtractedRefs(
+  roleDef: RoleDefinition<Record<string, unknown>>,
+  meta: unknown,
+): string[] {
+  const extractRefsFn = roleDef.extractRefs;
+  if (extractRefsFn === null || typeof extractRefsFn !== "function") {
+    return [];
+  }
+  return extractRefsFn(meta as Record<string, unknown>);
+}
+
+async function resolveRoleMeta<M extends RoleMeta>(
+  roleDef: RoleDefinition<Record<string, unknown>>,
+  extractCtx: ExtractContext<M>,
+  extract: ExtractFn,
+  llmProvider: LlmProvider | null,
+  cas: CasStore,
+): Promise<Record<string, unknown>> {
+  if (roleDef.extractMode === "react") {
+    if (llmProvider === null) {
+      throw new Error(
+        'createWorkflow: llmProvider is required when a role uses extractMode "react"',
+      );
+    }
+    const text = await buildExtractUserContent(
+      extractCtx as unknown as ExtractContext,
+      roleDef.extractPrompt,
+    );
+    const reactResult = await reactExtract({
+      text,
+      schema: roleDef.schema,
+      provider: llmProvider,
+      cas,
+    });
+    if (!reactResult.ok) {
+      throw new Error(`react extract failed: ${reactResult.error}`);
+    }
+    return reactResult.value as Record<string, unknown>;
+  }
+  return (await extract(
+    roleDef.schema,
+    roleDef.extractPrompt,
+    extractCtx as unknown as ExtractContext,
+  )) as Record<string, unknown>;
+}
+
 /**
  * Binds pure role definitions + moderator to runtime agents and structured extraction.
- * Assign with `export const run = createWorkflow(def, binding, extract)`.
+ * Assign with `export const run = createWorkflow(def, binding, extract, llmProvider)`.
+ * Pass the same {@link LlmProvider} as {@link createExtract} when any role uses `extractMode: "react"`.
  */
 export function createWorkflow<M extends RoleMeta>(
   def: Pick<WorkflowDefinition<M>, "roles" | "moderator">,
   binding: AgentBinding,
   extract: ExtractFn,
+  llmProvider: LlmProvider | null,
 ): WorkflowFn {
   return async function* workflowLoop(
     input: ThreadInput,
     options: WorkflowFnOptions,
-  ): AsyncGenerator<RoleOutput, WorkflowResult> {
+  ): AsyncGenerator<RoleOutput, WorkflowCompletion> {
     const nowMs = Date.now();
     const start: ModeratorContext<M>["start"] = {
       role: START,
@@ -46,8 +100,9 @@ export function createWorkflow<M extends RoleMeta>(
     const baseTs = Date.now();
     let steps: RoleStep<M>[] = input.steps.map((out, i) => ({
       role: out.role,
-      content: out.content,
+      contentHash: out.contentHash,
       meta: out.meta,
+      refs: out.refs,
       timestamp: baseTs + i,
     })) as RoleStep<M>[];
 
@@ -61,6 +116,7 @@ export function createWorkflow<M extends RoleMeta>(
 
       const modCtx: ModeratorContext<M> = {
         threadId: options.threadId,
+        depth: options.depth,
         start,
         steps,
       };
@@ -79,6 +135,7 @@ export function createWorkflow<M extends RoleMeta>(
       const agentCtx: AgentContext<M> = {
         ...modCtx,
         currentRole: { name: next, systemPrompt: roleDef.systemPrompt },
+        cas: options.cas,
       };
 
       const agent = binding.overrides?.[next] ?? binding.agent;
@@ -90,21 +147,36 @@ export function createWorkflow<M extends RoleMeta>(
         agentContent: raw,
       };
 
-      const meta = await extract(
-        roleDef.schema,
-        roleDef.extractPrompt,
-        extractCtx as unknown as ExtractContext,
+      const meta = await resolveRoleMeta(
+        roleDef as unknown as RoleDefinition<Record<string, unknown>>,
+        extractCtx,
+        extract,
+        llmProvider,
+        options.cas,
+      );
+
+      const contentHash = await putContentMerkleNode(options.cas, raw);
+
+      const refs = mergeRefsWithContentHash(
+        resolveExtractedRefs(roleDef as unknown as RoleDefinition<Record<string, unknown>>, meta),
+        contentHash,
       );
 
       const ts = Date.now();
       const step = {
         role: next,
-        content: raw,
+        contentHash,
         meta,
+        refs,
         timestamp: ts,
       } as RoleStep<M>;
 
-      yield { role: step.role, content: step.content, meta: step.meta };
+      yield {
+        role: step.role,
+        contentHash: step.contentHash,
+        meta: step.meta,
+        refs: step.refs,
+      };
 
       steps = [...steps, step];
     }

packages/workflow/src/engine.ts

@@ -1,26 +1,39 @@
 import { appendFile, mkdir } from "node:fs/promises";
 import { dirname } from "node:path";
 
+import type { CasStore } from "./cas.js";
 import type { LogFn } from "./logger.js";
-import type { ThreadInput, WorkflowFn, WorkflowFnOptions, WorkflowResult } from "./types.js";
+import { getContentMerklePayload, putStepMerkleNode, putThreadMerkleNode } from "./merkle.js";
+import { normalizeRefsField } from "./refs-field.js";
+import type {
+  ThreadInput,
+  WorkflowCompletion,
+  WorkflowFn,
+  WorkflowFnOptions,
+  WorkflowResult,
+} from "./types.js";
 
 export type ExecuteThreadIo = {
   threadId: string;
   hash: string;
   dataJsonlPath: string;
   infoJsonlPath: string;
+  cas: CasStore;
 };
 
 /** One persisted role line in `.data.jsonl` (engine adds these for fork replay before running the generator). */
 export type PrefilledDiskStep = {
   role: string;
-  content: string;
+  contentHash: string;
   meta: Record<string, unknown>;
+  refs: string[];
   timestamp: number;
 };
 
 export type ExecuteThreadOptions = {
   maxRounds: number;
+  /** Passed to the bundle as `WorkflowFnOptions.depth`. */
+  depth: number;
   signal: AbortSignal;
   /** Invoked after each successful yield (and outer-loop checks); used for pause/resume. */
   awaitAfterEachYield: () => Promise<void>;
@@ -38,50 +51,123 @@ async function appendDataLine(path: string, record: unknown): Promise<void> {
   await appendFile(path, line, "utf8");
 }
 
+async function finalizeThreadResult(params: {
+  cas: CasStore;
+  workflowName: string;
+  threadId: string;
+  stepMerkleHashes: readonly string[];
+  completion: WorkflowCompletion;
+}): Promise<WorkflowResult> {
+  const rootHash = await putThreadMerkleNode(
+    params.cas,
+    {
+      workflow: params.workflowName,
+      threadId: params.threadId,
+      result: {
+        returnCode: params.completion.returnCode,
+        summary: params.completion.summary,
+      },
+    },
+    params.stepMerkleHashes,
+  );
+  return {
+    returnCode: params.completion.returnCode,
+    summary: params.completion.summary,
+    rootHash,
+  };
+}
+
 async function driveWorkflowGenerator(params: {
   fn: WorkflowFn;
+  workflowName: string;
   input: ThreadInput;
   bundleOptions: WorkflowFnOptions;
   executeOptions: ExecuteThreadOptions;
   dataJsonlPath: string;
   threadId: string;
   logger: LogFn;
+  cas: CasStore;
+  stepMerkleHashes: string[];
 }): Promise<WorkflowResult> {
-  const { fn, input, bundleOptions, executeOptions, dataJsonlPath, threadId, logger } = params;
+  const {
+    fn,
+    workflowName,
+    input,
+    bundleOptions,
+    executeOptions,
+    dataJsonlPath,
+    threadId,
+    logger,
+    cas,
+    stepMerkleHashes,
+  } = params;
   const gen = fn(input, bundleOptions);
   let written = 0;
 
   while (true) {
     if (executeOptions.signal.aborted) {
       logger("V8JX4NP2", `thread ${threadId} aborted`);
-      return { returnCode: 130, summary: "thread aborted" };
+      return await finalizeThreadResult({
+        cas,
+        workflowName,
+        threadId,
+        stepMerkleHashes,
+        completion: { returnCode: 130, summary: "thread aborted" },
+      });
     }
 
     if (written >= executeOptions.maxRounds) {
       logger("R3CW7YBQ", `thread ${threadId} stopped at maxRounds=${executeOptions.maxRounds}`);
-      return {
-        returnCode: 0,
-        summary: `completed: reached maxRounds (${executeOptions.maxRounds})`,
-      };
+      return await finalizeThreadResult({
+        cas,
+        workflowName,
+        threadId,
+        stepMerkleHashes,
+        completion: {
+          returnCode: 0,
+          summary: `completed: reached maxRounds (${executeOptions.maxRounds})`,
+        },
+      });
     }
 
     const iterResult = await gen.next();
 
     if (iterResult.done) {
       logger("F3HN8QKP", `thread ${threadId} generator finished`);
-      return iterResult.value;
+      const completion = iterResult.value;
+      return await finalizeThreadResult({
+        cas,
+        workflowName,
+        threadId,
+        stepMerkleHashes,
+        completion,
+      });
     }
 
     written++;
     const step = iterResult.value;
+    const resolved = await getContentMerklePayload(cas, step.contentHash);
+    if (resolved === null) {
+      throw new Error(
+        `role step ${step.role}: CAS blob missing for contentHash ${step.contentHash}`,
+      );
+    }
     const ts = Date.now();
     await appendDataLine(dataJsonlPath, {
       role: step.role,
-      content: step.content,
+      contentHash: step.contentHash,
       meta: step.meta,
+      refs: normalizeRefsField(step.refs),
       timestamp: ts,
     });
 
+    const stepNodeHash = await putStepMerkleNode(
+      cas,
+      { role: step.role, meta: step.meta },
+      step.contentHash,
+    );
+    stepMerkleHashes.push(stepNodeHash);
+
     logger("N7BW4YHQ", `thread ${threadId} wrote role ${step.role}`);
 
     await Promise.race([
@@ -97,7 +183,13 @@ async function driveWorkflowGenerator(params: {
 
     if (executeOptions.signal.aborted) {
       logger("V8JX4NP4", `thread ${threadId} aborted`);
-      return { returnCode: 130, summary: "thread aborted" };
+      return await finalizeThreadResult({
+        cas,
+        workflowName,
+        threadId,
+        stepMerkleHashes,
+        completion: { returnCode: 130, summary: "thread aborted" },
+      });
     }
   }
 }
@@ -133,6 +225,7 @@ export async function executeThread(
       prompt: input.prompt,
       options: {
         maxRounds: options.maxRounds,
+        depth: options.depth,
       },
     },
     timestamp: nowMs,
@@ -145,37 +238,63 @@ export async function executeThread(
 
   logger("T9HQ2KHM", `thread ${io.threadId} started for workflow ${workflowName}`);
 
+  const stepMerkleHashes: string[] = [];
+
   if (prefilled !== null) {
     for (const row of prefilled) {
+      const prefilledPayload = await getContentMerklePayload(io.cas, row.contentHash);
+      if (prefilledPayload === null) {
+        throw new Error(
+          `prefilled step ${row.role}: CAS blob missing for contentHash ${row.contentHash}`,
+        );
+      }
       await appendDataLine(io.dataJsonlPath, {
         role: row.role,
-        content: row.content,
+        contentHash: row.contentHash,
         meta: row.meta,
+        refs: normalizeRefsField(row.refs),
         timestamp: row.timestamp,
       });
+      const stepNodeHash = await putStepMerkleNode(
+        io.cas,
+        { role: row.role, meta: row.meta },
+        row.contentHash,
+      );
+      stepMerkleHashes.push(stepNodeHash);
     }
   }
 
   if (options.maxRounds <= 0) {
     logger("R3CW7YBQ", `thread ${io.threadId} stopped at maxRounds=${options.maxRounds}`);
-    return {
-      returnCode: 0,
-      summary: `completed: reached maxRounds (${options.maxRounds})`,
-    };
+    return await finalizeThreadResult({
+      cas: io.cas,
+      workflowName,
+      threadId: io.threadId,
+      stepMerkleHashes,
+      completion: {
+        returnCode: 0,
+        summary: `completed: reached maxRounds (${options.maxRounds})`,
+      },
+    });
   }
 
   const bundleOptions: WorkflowFnOptions = {
     threadId: io.threadId,
     maxRounds: options.maxRounds,
+    depth: options.depth,
+    cas: io.cas,
   };
 
   return await driveWorkflowGenerator({
     fn,
+    workflowName,
     input,
     bundleOptions,
     executeOptions: options,
     dataJsonlPath: io.dataJsonlPath,
     threadId: io.threadId,
     logger,
+    cas: io.cas,
+    stepMerkleHashes,
   });
 }

packages/workflow/src/ensure-uncaged-workflow-symlink.ts

@@ -0,0 +1,36 @@
+import { mkdir, readlink, symlink, unlink } from "node:fs/promises";
+import path from "node:path";
+import { fileURLToPath } from "node:url";
+
+/** This module lives in `@uncaged/workflow/src`; parent dir is the package root. */
+function installedWorkflowPackageDir(): string {
+  return fileURLToPath(new URL("..", import.meta.url));
+}
+
+/**
+ * Ensures `<storageRoot>/node_modules/@uncaged/workflow` points at the installed `@uncaged/workflow`
+ * package so workflow bundles loaded from `<storageRoot>/bundles/*.esm.js` can resolve `import "@uncaged/workflow"`.
+ */
+export async function ensureUncagedWorkflowSymlink(storageRoot: string): Promise<void> {
+  const target = installedWorkflowPackageDir();
+  const linkDir = path.join(storageRoot, "node_modules", "@uncaged");
+  const linkPath = path.join(linkDir, "workflow");
+  await mkdir(linkDir, { recursive: true });
+
+  try {
+    const existing = await readlink(linkPath);
+    const normalizedExisting = path.resolve(linkDir, existing);
+    if (normalizedExisting === target) {
+      return;
+    }
+    await unlink(linkPath);
+  } catch (e) {
+    const errObj = e as NodeJS.ErrnoException;
+    if (errObj.code !== "ENOENT" && errObj.code !== "EINVAL") {
+      throw e;
+    }
+  }
+
+  const linkType = process.platform === "win32" ? "junction" : "dir";
+  await symlink(target, linkPath, linkType);
+}

packages/workflow/src/extract-bundle-exports.ts

@@ -1,5 +1,5 @@
-import { pathToFileURL } from "node:url";
-
+import { importWorkflowBundleModule } from "./bundle-import-env.js";
+import { ensureUncagedWorkflowSymlink } from "./ensure-uncaged-workflow-symlink.js";
 import { err, ok, type Result } from "./result.js";
 import type { WorkflowFn } from "./types.js";
 import type { WorkflowDescriptor } from "./workflow-descriptor.js";
@@ -10,14 +10,23 @@ export type ExtractedBundleExports = {
   descriptor: WorkflowDescriptor;
 };
 
+export type ExtractBundleExportsOptions = {
+  /** When set, ensures `node_modules/@uncaged/workflow` exists under this root before import. */
+  storageRoot: string | null;
+};
+
 /** Load a workflow `.esm.js` bundle and read its named exports (`run`, `descriptor`). */
 export async function extractBundleExports(
   bundlePath: string,
+  options: ExtractBundleExportsOptions = { storageRoot: null },
 ): Promise<Result<ExtractedBundleExports, string>> {
   let modUnknown: unknown;
   try {
+    if (options.storageRoot !== null) {
+      await ensureUncagedWorkflowSymlink(options.storageRoot);
+    }
     // Dynamic import required: user bundle path resolved at runtime
-    modUnknown = await import(pathToFileURL(bundlePath).href);
+    modUnknown = await importWorkflowBundleModule(bundlePath);
   } catch (e) {
     const message = e instanceof Error ? e.message : String(e);
     return err(`failed to import bundle: ${message}`);

packages/workflow/src/extract-fn.ts

@@ -1,6 +1,7 @@
 import type * as z from "zod/v4";
 
 import { llmExtractWithRetry } from "./llm-extract.js";
+import { getContentMerklePayload } from "./merkle.js";
 import type { ExtractContext, LlmProvider } from "./types.js";
 
 export type ExtractFn = <T extends Record<string, unknown>>(
@@ -9,6 +10,40 @@ export type ExtractFn = <T extends Record<string, unknown>>(
   ctx: ExtractContext,
 ) => Promise<T>;
 
+/** Builds the user-side extraction prompt (thread + agent output + instruction). */
+export async function buildExtractUserContent(
+  ctx: ExtractContext,
+  prompt: string,
+): Promise<string> {
+  const lines: string[] = [];
+  lines.push(`## Role: ${ctx.currentRole.name}`);
+  lines.push(ctx.currentRole.systemPrompt);
+  lines.push("");
+  lines.push("## Task");
+  lines.push(ctx.start.content);
+  lines.push("");
+  if (ctx.steps.length > 0) {
+    lines.push("## Thread History");
+    for (const step of ctx.steps) {
+      const body = await getContentMerklePayload(ctx.cas, step.contentHash);
+      if (body === null) {
+        throw new Error(`extract: missing CAS blob for step ${step.role}: ${step.contentHash}`);
+      }
+      lines.push(`### ${step.role}`);
+      lines.push(body);
+      lines.push(`Meta: ${JSON.stringify(step.meta)}`);
+      lines.push("");
+    }
+  }
+  lines.push("## Agent Output");
+  lines.push(ctx.agentContent);
+  lines.push("");
+  lines.push("## Extraction Instruction");
+  lines.push(prompt);
+
+  return lines.join("\n");
+}
+
 /**
  * Create an ExtractFn backed by an LLM provider.
  * Builds prompt text from {@link ExtractContext} plus `prompt` and calls structured extraction.
@@ -19,29 +54,7 @@ export function createExtract(provider: LlmProvider): ExtractFn {
     prompt: string,
     ctx: ExtractContext,
   ): Promise<T> => {
-    const lines: string[] = [];
-    lines.push(`## Role: ${ctx.currentRole.name}`);
-    lines.push(ctx.currentRole.systemPrompt);
-    lines.push("");
-    lines.push("## Task");
-    lines.push(ctx.start.content);
-    lines.push("");
-    if (ctx.steps.length > 0) {
-      lines.push("## Thread History");
-      for (const step of ctx.steps) {
-        lines.push(`### ${step.role}`);
-        lines.push(step.content);
-        lines.push(`Meta: ${JSON.stringify(step.meta)}`);
-        lines.push("");
-      }
-    }
-    lines.push("## Agent Output");
-    lines.push(ctx.agentContent);
-    lines.push("");
-    lines.push("## Extraction Instruction");
-    lines.push(prompt);
-
-    const text = lines.join("\n");
+    const text = await buildExtractUserContent(ctx, prompt);
     const result = await llmExtractWithRetry({ text, schema, provider });
     if (!result.ok) {
       throw new Error(`extract failed: ${JSON.stringify(result.error)}`);

packages/workflow/src/extract-provider.ts

@@ -0,0 +1,35 @@
+import { readWorkflowRegistry } from "./registry.js";
+import type { WorkflowConfig } from "./registry-types.js";
+import { err, ok, type Result } from "./result.js";
+import { getDefaultWorkflowStorageRoot } from "./storage-root.js";
+import type { LlmProvider } from "./types.js";
+
+const DEFAULT_WORKFLOW_AS_AGENT_MAX_DEPTH = 3;
+
+export function getWorkflowAsAgentMaxDepth(config: WorkflowConfig | null): number {
+  if (config === null) {
+    return DEFAULT_WORKFLOW_AS_AGENT_MAX_DEPTH;
+  }
+  return config.maxDepth;
+}
+
+/** Loads `config.extract` from workflow.yaml (apiKey already resolved at registry parse time). */
+export async function getExtractProvider(
+  storageRoot: string | undefined,
+): Promise<Result<LlmProvider, string>> {
+  const root = storageRoot ?? getDefaultWorkflowStorageRoot();
+  const regResult = await readWorkflowRegistry(root);
+  if (!regResult.ok) {
+    return err(regResult.error.message);
+  }
+  const cfg = regResult.value.config;
+  if (cfg === null) {
+    return err("workflow registry has no global config section");
+  }
+  const ex = cfg.extract;
+  return ok({
+    baseUrl: ex.baseUrl,
+    apiKey: ex.apiKey,
+    model: ex.model,
+  });
+}

packages/workflow/src/fork-thread.ts

@@ -1,5 +1,6 @@
+import { normalizeRefsField } from "./refs-field.js";
 import { err, ok, type Result } from "./result.js";
-import type { RoleOutput } from "./types.js";
+import type { RoleOutput, WorkflowCompletion } from "./types.js";
 
 /** Role steps replayed from `.data.jsonl`, including persisted timestamps. */
 export type ForkHistoricalStep = RoleOutput & { timestamp: number };
@@ -10,34 +11,59 @@ export type ParsedThreadStartRecord = {
   threadId: string;
   prompt: string;
   maxRounds: number;
+  depth: number;
 };
 
+/** Recognizes a persisted workflow completion line (no `role`; has numeric `returnCode` and string `summary`). Omits `rootHash` when absent. */
+export function tryParseWorkflowResultRecord(
+  obj: Record<string, unknown>,
+): WorkflowCompletion | null {
+  if (obj.role !== undefined) {
+    return null;
+  }
+  const returnCode = obj.returnCode;
+  const summary = obj.summary;
+  if (typeof returnCode !== "number" || typeof summary !== "string") {
+    return null;
+  }
+  return { returnCode, summary };
+}
+
+export function tryParseRoleStepRecord(obj: Record<string, unknown>): ForkHistoricalStep | null {
+  const role = obj.role;
+  const contentHash = obj.contentHash;
+  const meta = obj.meta;
+  const timestamp = obj.timestamp;
+  if (typeof role !== "string") {
+    return null;
+  }
+  if (typeof contentHash !== "string") {
+    return null;
+  }
+  if (meta === null || typeof meta !== "object") {
+    return null;
+  }
+  if (typeof timestamp !== "number") {
+    return null;
+  }
+  return {
+    role,
+    contentHash,
+    meta: meta as Record<string, unknown>,
+    refs: normalizeRefsField(obj.refs),
+    timestamp,
+  };
+}
+
 function parseRoleLine(
   obj: Record<string, unknown>,
   lineIndex: number,
 ): Result<ForkHistoricalStep, string> {
-  const role = obj.role;
-  const content = obj.content;
-  const meta = obj.meta;
-  const timestamp = obj.timestamp;
-  if (typeof role !== "string") {
-    return err(`invalid role record at line ${lineIndex}: missing role`);
+  const parsed = tryParseRoleStepRecord(obj);
+  if (parsed === null) {
+    return err(`invalid role record at line ${lineIndex}`);
   }
-  if (typeof content !== "string") {
-    return err(`invalid role record at line ${lineIndex}: missing content`);
-  }
-  if (meta === null || typeof meta !== "object") {
-    return err(`invalid role record at line ${lineIndex}: missing meta`);
-  }
-  if (typeof timestamp !== "number") {
-    return err(`invalid role record at line ${lineIndex}: missing timestamp`);
-  }
-  return ok({
-    role,
-    content,
-    meta: meta as Record<string, unknown>,
-    timestamp,
-  });
+  return ok(parsed);
 }
 
 function parseStartRecordLine(firstLine: string): Result<ParsedThreadStartRecord, string> {
@@ -76,12 +102,17 @@ function parseStartRecordLine(firstLine: string): Result<ParsedThreadStartRecord
     return err("start record missing parameters.options.maxRounds");
   }
 
+  const depthRaw = optRec.depth;
+  const depth =
+    typeof depthRaw === "number" && Number.isFinite(depthRaw) ? Math.trunc(depthRaw) : 0;
+
   return ok({
     workflowName: name,
     hash,
     threadId,
     prompt,
     maxRounds,
+    depth,
   });
 }
 
@@ -101,7 +132,15 @@ function parseFollowingRoleLines(lines: string[]): Result<ForkHistoricalStep[],
     if (rec === null || typeof rec !== "object") {
       return err(`invalid record at line ${i + 1}`);
     }
-    const parsed = parseRoleLine(rec as Record<string, unknown>, i + 1);
+    const recObj = rec as Record<string, unknown>;
+    const wf = tryParseWorkflowResultRecord(recObj);
+    if (wf !== null) {
+      if (i !== lines.length - 1) {
+        return err("WorkflowResult record must be the final line in `.data.jsonl`");
+      }
+      break;
+    }
+    const parsed = parseRoleLine(recObj, i + 1);
     if (!parsed.ok) {
       return parsed;
     }
@@ -194,7 +233,7 @@ export type ForkPlan = {
   hash: string;
   sourceThreadId: string;
   prompt: string;
-  runOptions: { maxRounds: number };
+  runOptions: { maxRounds: number; depth: number };
   historicalSteps: ForkHistoricalStep[];
 };
 
@@ -219,7 +258,7 @@ export function buildForkPlan(
     hash: start.hash,
     sourceThreadId: start.threadId,
     prompt: start.prompt,
-    runOptions: { maxRounds: start.maxRounds },
+    runOptions: { maxRounds: start.maxRounds, depth: start.depth },
     historicalSteps: selected.value,
   });
 }

packages/workflow/src/gc.ts

@@ -0,0 +1,131 @@
+import { readdir, readFile } from "node:fs/promises";
+import { join } from "node:path";
+
+import { type CasStore, createCasStore } from "./cas.js";
+import { parseThreadDataJsonl } from "./fork-thread.js";
+import { err, ok, type Result } from "./result.js";
+import { getGlobalCasDir } from "./storage-root.js";
+
+export type GcResult = {
+  scannedThreads: number;
+  activeRefs: number;
+  deletedEntries: number;
+  deletedHashes: string[];
+};
+
+async function listThreadDataJsonlPaths(storageRoot: string): Promise<Result<string[], string>> {
+  const logsRoot = join(storageRoot, "logs");
+  const paths: string[] = [];
+  let hashes: string[];
+  try {
+    hashes = await readdir(logsRoot);
+  } catch (e) {
+    const errObj = e as NodeJS.ErrnoException;
+    if (errObj.code === "ENOENT") {
+      return ok([]);
+    }
+    return err(`failed to read logs directory: ${String(e)}`);
+  }
+
+  for (const hash of hashes) {
+    const dir = join(logsRoot, hash);
+    let entries: string[];
+    try {
+      entries = await readdir(dir);
+    } catch {
+      continue;
+    }
+    for (const fileName of entries) {
+      if (fileName.endsWith(".data.jsonl")) {
+        paths.push(join(dir, fileName));
+      }
+    }
+  }
+
+  paths.sort();
+  return ok(paths);
+}
+
+async function collectActiveRefsFromDataPaths(
+  dataPaths: string[],
+): Promise<Result<Set<string>, string>> {
+  const activeRefs = new Set<string>();
+  for (const dataPath of dataPaths) {
+    let text: string;
+    try {
+      text = await readFile(dataPath, "utf8");
+    } catch (e) {
+      return err(`failed to read ${dataPath}: ${String(e)}`);
+    }
+    const parsed = parseThreadDataJsonl(text);
+    if (!parsed.ok) {
+      return err(`${dataPath}: ${parsed.error}`);
+    }
+    for (const step of parsed.value.roleSteps) {
+      for (const ref of step.refs) {
+        activeRefs.add(ref);
+      }
+    }
+  }
+  return ok(activeRefs);
+}
+
+async function deleteCasNotInSet(
+  cas: CasStore,
+  activeRefs: Set<string>,
+): Promise<Result<string[], string>> {
+  let listed: string[];
+  try {
+    listed = await cas.list();
+  } catch (e) {
+    return err(`failed to list cas entries: ${String(e)}`);
+  }
+
+  const deletedHashes: string[] = [];
+  for (const hash of listed) {
+    if (activeRefs.has(hash)) {
+      continue;
+    }
+    try {
+      await cas.delete(hash);
+    } catch (e) {
+      return err(`failed to delete cas ${hash}: ${String(e)}`);
+    }
+    deletedHashes.push(hash);
+  }
+
+  deletedHashes.sort();
+  return ok(deletedHashes);
+}
+
+/**
+ * Mark-and-sweep CAS GC: collect `refs` from all thread `.data.jsonl` files under `storageRoot`,
+ * then delete CAS blobs not referenced by any surviving thread data.
+ */
+export async function garbageCollectCas(storageRoot: string): Promise<Result<GcResult, string>> {
+  const pathsResult = await listThreadDataJsonlPaths(storageRoot);
+  if (!pathsResult.ok) {
+    return pathsResult;
+  }
+  const paths = pathsResult.value;
+
+  const refsResult = await collectActiveRefsFromDataPaths(paths);
+  if (!refsResult.ok) {
+    return refsResult;
+  }
+  const activeRefs = refsResult.value;
+
+  const cas = createCasStore(getGlobalCasDir(storageRoot));
+  const deletedResult = await deleteCasNotInSet(cas, activeRefs);
+  if (!deletedResult.ok) {
+    return deletedResult;
+  }
+  const deletedHashes = deletedResult.value;
+
+  return ok({
+    scannedThreads: paths.length,
+    activeRefs: activeRefs.size,
+    deletedEntries: deletedHashes.length,
+    deletedHashes,
+  });
+}

packages/workflow/src/index.ts

@@ -7,7 +7,7 @@ export {
 } from "./base32.js";
 export { buildDescriptor } from "./build-descriptor.js";
 export { validateWorkflowBundle, type WorkflowBundleValidationInput } from "./bundle-validator.js";
-export { type CasStore, createThreadCas } from "./cas.js";
+export { type CasStore, createCasStore, createThreadCas } from "./cas.js";
 export { createWorkflow } from "./create-workflow.js";
 export {
   type ExecuteThreadIo,
@@ -17,6 +17,7 @@ export {
 } from "./engine.js";
 export { type ExtractedBundleExports, extractBundleExports } from "./extract-bundle-exports.js";
 export { createExtract, type ExtractFn } from "./extract-fn.js";
+export { getExtractProvider } from "./extract-provider.js";
 export {
   buildForkPlan,
   type ForkHistoricalStep,
@@ -24,7 +25,10 @@ export {
   type ParsedThreadStartRecord,
   parseThreadDataJsonl,
   selectForkHistoricalSteps,
+  tryParseRoleStepRecord,
+  tryParseWorkflowResultRecord,
 } from "./fork-thread.js";
+export { type GcResult, garbageCollectCas } from "./gc.js";
 export { stringifyWorkflowDescriptor } from "./generate-descriptor.js";
 export { hashString, hashWorkflowBundleBytes } from "./hash.js";
 export {
@@ -40,6 +44,21 @@ export {
   type LoggerSink,
 } from "./logger.js";
 export {
+  createContentMerkleNode,
+  getContentMerklePayload,
+  type MerkleNode,
+  type MerkleNodeType,
+  parseMerkleNode,
+  putContentMerkleNode,
+  putStepMerkleNode,
+  putThreadMerkleNode,
+  type StepMerklePayload,
+  serializeMerkleNode,
+  type ThreadMerklePayload,
+} from "./merkle.js";
+export { type ReactExtractArgs, reactExtract } from "./react-extract.js";
+export {
+  type ExtractProviderConfig,
   getRegisteredWorkflow,
   listRegisteredWorkflowNames,
   parseWorkflowRegistryYaml,
@@ -48,6 +67,7 @@ export {
   rollbackWorkflowToHistoryHash,
   stringifyWorkflowRegistryYaml,
   unregisterWorkflow,
+  type WorkflowConfig,
   type WorkflowHistoryEntry,
   type WorkflowRegistryEntry,
   type WorkflowRegistryFile,
@@ -55,7 +75,7 @@ export {
   writeWorkflowRegistry,
 } from "./registry.js";
 export { err, ok, type Result } from "./result.js";
-export { getDefaultWorkflowStorageRoot } from "./storage-root.js";
+export { getDefaultWorkflowStorageRoot, getGlobalCasDir } from "./storage-root.js";
 export { createThreadPauseGate, type ThreadPauseGate } from "./thread-pause-gate.js";
 export {
   type AgentBinding,
@@ -63,6 +83,7 @@ export {
   type AgentFn,
   END,
   type ExtractContext,
+  type ExtractMode,
   type LlmProvider,
   type Moderator,
   type ModeratorContext,
@@ -74,6 +95,7 @@ export {
   type StartStep,
   type ThreadContext,
   type ThreadInput,
+  type WorkflowCompletion,
   type WorkflowDefinition,
   type WorkflowFn,
   type WorkflowFnOptions,
@@ -81,6 +103,7 @@ export {
 } from "./types.js";
 export { generateUlid } from "./ulid.js";
 export { getWorkerHostScriptPath } from "./worker-entry-path.js";
+export { type WorkflowAsAgentOptions, workflowAsAgent } from "./workflow-as-agent.js";
 export {
   validateWorkflowDescriptor,
   type WorkflowDescriptor,