feat: Phase 2 — engine write path (CAS nodes + threads.json)

- Engine writes StartNode, StateNode, ContentMerkleNode as CAS blobs - threads.json tracks active threads, completed → history/{date}.jsonl - No more .data.jsonl writes - ancestors skip-list: [parent, ...parentAncestors] capped at 11 - Tests: 4 pass (engine write path) Refs #155, closes #157 小橘 <xiaoju@shazhou.work>
feat: Phase 1 — CAS thread storage types + helpers
2026-05-09 07:53:44 +00:00 · 2026-05-09 07:30:47 +00:00 · 2026-05-09 07:12:29 +00:00 · 2026-05-09 07:08:10 +00:00 · 2026-05-09 06:34:24 +00:00 · 2026-05-09 04:39:57 +00:00
245 changed files with 4763 additions and 5522 deletions
@@ -2,7 +2,7 @@

 ## Project Overview

-**@uncaged/workflow** is a workflow engine that executes single-file ESM bundles. Each workflow is a self-contained `.esm.js` file with an XXH64 hash as its version identifier.
+This monorepo implements a workflow engine that executes single-file ESM bundles. Each workflow is a self-contained `.esm.js` file with an XXH64 hash as its version identifier. Shared types live in `@uncaged/workflow-protocol`; bundle authors typically depend on `@uncaged/workflow-runtime`.

 ### Key Terms

@@ -19,14 +19,27 @@
 ```
 workflow/
  packages/
-    workflow/       # @uncaged/workflow — core lib (types, hash, ULID, JSONL, registry)
-    cli-workflow/   # @uncaged/cli-workflow — CLI (uncaged-workflow command)
+    workflow-protocol/              # @uncaged/workflow-protocol — shared types + Result
+    workflow-runtime/               # @uncaged/workflow-runtime — createWorkflow, type re-exports
+    workflow-util/                  # @uncaged/workflow-util — Base32, ULID, logger, storage paths, refs helpers
+    workflow-reactor/               # @uncaged/workflow-reactor — LLM fn + thread reactor (tool calls)
+    workflow-cas/                   # @uncaged/workflow-cas — CAS store, hash, Merkle
+    workflow-register/              # @uncaged/workflow-register — bundle validation, registry YAML, model resolution
+    workflow-execute/               # @uncaged/workflow-execute — engine, extract, fork, GC, workflowAsAgent
+    cli-workflow/                   # @uncaged/cli-workflow — uncaged-workflow CLI
+    workflow-agent-cursor/          # @uncaged/workflow-agent-cursor
+    workflow-agent-hermes/          # @uncaged/workflow-agent-hermes
+    workflow-agent-llm/             # @uncaged/workflow-agent-llm
+    workflow-util-agent/            # @uncaged/workflow-util-agent — buildAgentPrompt, spawnCli
+    workflow-template-develop/      # @uncaged/workflow-template-develop
+    workflow-template-solve-issue/  # @uncaged/workflow-template-solve-issue
+    workflow-dashboard/             # @uncaged/workflow-dashboard — React dashboard (private app)
  docs/             # RFCs, conventions
  biome.json        # root Biome config
  tsconfig.json     # root TypeScript config
 ```

- `workflow` is the core; `cli-workflow` depends on it
+- Execution stack layers: `workflow-protocol` → (`workflow-runtime`, `workflow-util`, `workflow-reactor`) → (`workflow-cas`, `workflow-register`) → `workflow-execute` → `cli-workflow`
 - Packages use `workspace:*` protocol

 ## Language & Paradigm
@@ -167,10 +180,10 @@ type Result<T, E = Error> = { ok: true; value: T } | { ok: false; error: E };

 Never use `console.log/warn/error` directly — Biome's `noConsole` rule enforces this.

-All logging goes through the structured logger from `@uncaged/workflow`:
+All logging goes through the structured logger from `@uncaged/workflow-util`:

 ```typescript
-import { createLogger } from "@uncaged/workflow";
+import { createLogger } from "@uncaged/workflow-util";

 const log = createLogger();

@@ -1,7 +1,7 @@
 {
  "$schema": "https://biomejs.dev/schemas/2.4.14/schema.json",
  "files": {
-    "includes": ["**", "!**/dist", "!**/node_modules"]
+    "includes": ["**", "!**/dist", "!**/node_modules", "!packages/workflow/workflow"]
  },
  "assist": { "actions": { "source": { "organizeImports": "on" } } },
  "formatter": {
@@ -1,6 +1,6 @@
-# @uncaged/workflow — Architecture
+# Uncaged workflow — Architecture

-**Last updated:** 2026-05-06 by 小橘 🍊（NEKO Team）
+**Last updated:** 2026-05-09

 ---

@@ -8,72 +8,106 @@

 A workflow engine that executes single-file ESM bundles. Each workflow is a self-contained `.esm.js` file identified by its XXH64 hash (Crockford Base32). No daemon — processes start on demand and exit when done.

-## Package Structure
+The implementation lives in **15** Bun workspace packages under `packages/`, using the `workspace:*` protocol.

-| Package | npm Name | Purpose |
-|---------|----------|---------|
-| `workflow` | `@uncaged/workflow` | Core: types, engine, ExtractFn, hash/ULID/registry |
-| `cli-workflow` | `@uncaged/cli-workflow` | CLI: `uncaged-workflow` command |
-| `workflow-agent-cursor` | `@uncaged/workflow-agent-cursor` | Cursor CLI agent (extracts workspace from ctx) |
-| `workflow-agent-hermes` | `@uncaged/workflow-agent-hermes` | Hermes CLI agent |
-| `workflow-agent-llm` | `@uncaged/workflow-agent-llm` | OpenAI-compatible LLM agent |
-| `workflow-template-develop` | `@uncaged/workflow-template-develop` | Develop workflow template (roles in `src/roles/`) |
-| `workflow-template-solve-issue` | `@uncaged/workflow-template-solve-issue` | Solve-issue workflow template (roles in `src/roles/`) |
-| `workflow-util-agent` | `@uncaged/workflow-util-agent` | `buildAgentPrompt` + `spawnCli` utilities |
+## Package map

-Monorepo with **bun workspace**, `workspace:*` protocol.
+Grouped by responsibility (npm name → folder).

-## Core Types
+| Layer | Package | One-line role |
+|-------|---------|----------------|
+| Contract | `@uncaged/workflow-protocol` → `workflow-protocol` | Shared TypeScript types and `Result` helpers; peer `zod` only — no other workspace deps. |
+| Author API | `@uncaged/workflow-runtime` → `workflow-runtime` | `createWorkflow` and re-exports of protocol workflow types for bundle authors. |
+| Shared infra | `@uncaged/workflow-util` → `workflow-util` | Base32/ULID, logger, storage root paths, global CAS dir, ref-field helpers. |
+| LLM plumbing | `@uncaged/workflow-reactor` → `workflow-reactor` | `createLlmFn`, `createThreadReactor`, and related tool-call types for threaded LLM invocation. |
+| CAS | `@uncaged/workflow-cas` → `workflow-cas` | `CasStore` implementation, XXH64 hashing, Merkle helpers over CAS payloads. |
+| Registry / bundles | `@uncaged/workflow-register` → `workflow-register` | Bundle validation & dynamic export extraction, `workflow.yaml` registry I/O, provider/model resolution. |
+| Engine | `@uncaged/workflow-execute` → `workflow-execute` | Thread execution, worker entry path, fork/GC, extract pipeline, `workflowAsAgent`. |
+| CLI | `@uncaged/cli-workflow` → `cli-workflow` | `uncaged-workflow` binary (depends on engine, registry, CAS, protocol, util, runtime). |
+| Agent adapters | `@uncaged/workflow-agent-cursor` → `workflow-agent-cursor` | `AgentFn` via `cursor-agent` CLI + workspace extraction. |
+| | `@uncaged/workflow-agent-hermes` → `workflow-agent-hermes` | `AgentFn` via `hermes chat` CLI. |
+| | `@uncaged/workflow-agent-llm` → `workflow-agent-llm` | `AgentFn` via OpenAI-compatible HTTP (`LlmProvider` from runtime). |
+| Agent shared | `@uncaged/workflow-util-agent` → `workflow-util-agent` | `buildAgentPrompt`, `spawnCli` for CLI-backed agents. |
+| Templates | `@uncaged/workflow-template-develop` → `workflow-template-develop` | Develop workflow definition, roles, descriptor builder. |
+| | `@uncaged/workflow-template-solve-issue` → `workflow-template-solve-issue` | Solve-issue workflow definition, roles, descriptor builder. |
+| Dashboard | `@uncaged/workflow-dashboard` → `workflow-dashboard` | Private Vite + React app (`src/main.tsx`); only `react` / `react-dom` dependencies — no workspace packages. |

-```typescript
-// --- Sentinel values ---
-const START = "__start__";
-const END = "__end__";
+## Dependency graph (workspace packages)

-// --- RoleMeta: maps role names → their meta types ---
-type RoleMeta = Record<string, Record<string, unknown>>;
+Bottom-up layering for the execution stack:

-// --- Role Definition: pure data, no execution logic ---
-type RoleDefinition<Meta> = {
-  description: string;      // human-readable
-  systemPrompt: string;     // given to agent
-  extractPrompt: string;    // given to extractor
-  schema: z.ZodType<Meta>;  // meta shape (Zod v4)
-};
-
-// --- Workflow Definition: pure data, no agent binding ---
-type WorkflowDefinition<M extends RoleMeta> = {
-  description: string;
-  roles: { [K in keyof M & string]: RoleDefinition<M[K]> };
-  moderator: Moderator<M>;
-};
-
-// --- Agent: raw string output, reads role info from context ---
-type AgentFn = (ctx: AgentContext) => Promise<string>;
-
-// --- Agent Binding: runtime assignment ---
-type AgentBinding = {
-  agent: AgentFn;
-  overrides?: Partial<Record<string, AgentFn>>;
-};
-
-// --- Extract: structured data from context ---
-type ExtractFn = <T>(schema: z.ZodType<T>, prompt: string, ctx: ExtractContext) => Promise<T>;
-
-// --- Moderator: pure routing function ---
-type Moderator<M extends RoleMeta> = (ctx: ModeratorContext<M>) => (keyof M & string) | typeof END;
-
-// --- Composition ---
-// createWorkflow(def, binding, extract) => WorkflowFn
+```mermaid
+flowchart BT
+  subgraph L0["Layer 0 — contract"]
+    protocol["@uncaged/workflow-protocol"]
+  end
+  subgraph L1["Layer 1 — on protocol"]
+    runtime["@uncaged/workflow-runtime"]
+    util["@uncaged/workflow-util"]
+    reactor["@uncaged/workflow-reactor"]
+  end
+  subgraph L2["Layer 2 — protocol + util"]
+    cas["@uncaged/workflow-cas"]
+    register["@uncaged/workflow-register"]
+  end
+  subgraph L3["Layer 3 — engine"]
+    execute["@uncaged/workflow-execute"]
+  end
+  subgraph L4["Layer 4 — CLI"]
+    cli["@uncaged/cli-workflow"]
+  end
+  runtime --> protocol
+  util --> protocol
+  reactor --> protocol
+  cas --> protocol
+  cas --> util
+  register --> protocol
+  register --> util
+  execute --> protocol
+  execute --> runtime
+  execute --> util
+  execute --> cas
+  execute --> reactor
+  execute --> register
+  cli --> protocol
+  cli --> util
+  cli --> cas
+  cli --> execute
+  cli --> register
+  cli --> runtime
 ```

-## Three-Phase Engine Loop
+**Adjacent consumers** (not in the main CLI stack):

-Each role execution has three distinct phases with progressive context:
+- `@uncaged/workflow-util-agent` → `@uncaged/workflow-runtime`
+- `@uncaged/workflow-agent-llm` → `@uncaged/workflow-runtime`
+- `@uncaged/workflow-agent-cursor` → `@uncaged/workflow-runtime`, `@uncaged/workflow-util-agent`, `zod`
+- `@uncaged/workflow-agent-hermes` → `@uncaged/workflow-runtime`, `@uncaged/workflow-util-agent`
+- `@uncaged/workflow-template-develop` → `@uncaged/workflow-register`, `@uncaged/workflow-runtime`, `zod`
+- `@uncaged/workflow-template-solve-issue` → `@uncaged/workflow-register`, `@uncaged/workflow-runtime`, `zod` (dev-only workspace deps: `@uncaged/workflow-cas`, `@uncaged/workflow-execute` for tests/tooling per `package.json`)
+
+## Package roles (detail)
+
+- **`workflow-protocol`** — Pure types (`WorkflowFn`, contexts, `CasStore` interface, descriptor shapes), `START` / `END`, `ok` / `err`. Depends only on peer `zod` for schema-related types in signatures.
+- **`workflow-runtime`** — Workflow author surface: `createWorkflow` from `src/create-workflow.js`, re-exports protocol types/constants used when authoring bundles.
+- **`workflow-util`** — Cross-cutting utilities: Crockford Base32, ULID, `createLogger`, `getDefaultWorkflowStorageRoot`, `getGlobalCasDir`, ref normalization; re-exports `ok`/`err` from protocol.
+- **`workflow-cas`** — Filesystem CAS (`createCasStore`), `hashString` / `hashWorkflowBundleBytes`, Merkle node serialization and helpers (`merkle.js`).
+- **`workflow-register`** — Bundle pipeline (`validateWorkflowBundle`, `extractBundleExports`, descriptor builders), registry YAML read/write, `resolveModel` / `splitProviderModelRef`.
+- **`workflow-execute`** — `executeThread`, supervisor/worker wiring (`engine/`), fork/GC/pause gate, `createExtract` + LLM extract helpers (`extract/`), `workflowAsAgent`. Imports `@uncaged/workflow-reactor` for LLM-backed extract/supervisor paths (`extract-fn.ts`, `supervisor.ts`).
+- **`workflow-reactor`** — `createLlmFn`, `createThreadReactor`, and thread tool-invocation types — consumed by `workflow-execute`.
+- **`cli-workflow`** — CLI commands and HTTP/dashboard-related wiring (`hono`, `yaml`); composes register + execute + CAS + util.
+- **`workflow-agent-*`** — Replaceable `AgentFn` implementations (Cursor / Hermes CLIs, or HTTP LLM).
+- **`workflow-util-agent`** — Shared prompt assembly and subprocess spawning for CLI agents.
+- **`workflow-template-*`** — Concrete `WorkflowDefinition` graphs + Zod role schemas + descriptor builders for publishing bundles.
+- **`workflow-dashboard`** — Standalone React UI; no published library entry matching `src/index.ts`.
+
+## Three-phase engine loop
+
+Each role round is implemented in `packages/workflow-runtime/src/create-workflow.ts` (`advanceOneRound`): moderator → agent → extractor, with progressive context types from `@uncaged/workflow-protocol`.

 ```
 ┌─→ Phase 1: MODERATOR
-│   Context: ModeratorContext { threadId, start, steps }
+│   Context: ModeratorContext { threadId, depth, start, steps }
 │   Action:  moderator(ctx) → role name | END
 │
 │   Phase 2: AGENT
@@ -82,90 +116,80 @@ Each role execution has three distinct phases with progressive context:
 │
 │   Phase 3: EXTRACTOR
 │   Context: ExtractContext = AgentCtx + { agentContent }
-│   Action:  extract(schema, extractPrompt, ctx) → typed meta
+│   Action:  runtime.extract(schema, extractPrompt, ctx) → typed meta
 │
-│   Merge: RoleStep { role, content, meta, timestamp }
+│   Merge: RoleStep { role, contentHash, meta, refs, timestamp }
 │   Append to steps
 └─────────────────────────────────────────────────────┘
 ```

-### Context Types (progressive)
+### Context types (progressive)
+
+Defined in `packages/workflow-protocol/src/types.ts`:

 ```typescript
-// Phase 1: Moderator sees accumulated state only
-type ModeratorContext<M> = {
-  threadId: string;
-  start: StartStep;
-  steps: RoleStep<M>[];
-};
-
-// Phase 2: Agent knows its identity
+type ModeratorContext<M> = ThreadContext<M>;
 type AgentContext<M> = ModeratorContext<M> & {
  currentRole: { name: string; systemPrompt: string };
 };
-
-// Phase 3: Extractor has agent output
-type ExtractContext<M> = AgentContext<M> & {
-  agentContent: string;
-};
-
-// ThreadContext is an alias for AgentContext (backward compat)
-type ThreadContext<M> = AgentContext<M>;
+type ExtractContext<M> = AgentContext<M> & { agentContent: string };
 ```

-### Key Properties
+### Key properties

- **Moderator is synchronous and pure** — no I/O, no state mutation
- **Agent gets context, not instructions** — reads `ctx.currentRole.systemPrompt`
- **Extractor is a general tool** — not limited to post-agent extraction; agents can use it too (e.g. Cursor agent extracts workspace path before execution)
- **extractPrompt is a call parameter**, not context state — different callers use different prompts
+- **Moderator is synchronous and pure** — no I/O, no state mutation inside `createWorkflow`’s moderator call path.
+- **Agent receives `AgentContext`** — reads `ctx.currentRole.systemPrompt`; raw output becomes `agentContent` for extract.
+- **Extractor is `WorkflowRuntime.extract`** — supplied by the engine from registry-resolved LLM config (`workflow-execute`); stores agent body in CAS and yields `contentHash` + `refs` on each step (`create-workflow.ts`).
+- **`extractPrompt` is a call parameter** on `RoleDefinition`, not implicit context state.

-## Agent Information Sources
+## Agent information sources

 An agent has exactly three information sources:

 1. **Prior knowledge** — LLM training, agent memory, agent skills
-2. **Thread context** — `AgentContext` (start, steps, currentRole)
+2. **Thread context** — `AgentContext` (`start`, `steps`, `currentRole`)
 3. **Derived information** — from 1 & 2 (e.g. tool calls, shell commands)

-No hidden environment parameters. If an agent needs something (like a workspace path), it extracts it from context using `ExtractFn`.
+No hidden environment parameters. If an agent needs something (like a workspace path), it obtains it via `ExtractFn` (e.g. Cursor agent).

-## Bundle Contract
+## Bundle contract

-A workflow bundle is a single `.esm.js` file with two named exports:
+A workflow bundle is a single `.esm.js` file with two named exports (see `WorkflowFn` / `WorkflowDescriptor` in `packages/workflow-protocol/src/types.ts`):

 ```typescript
-// Named exports (no default export)
 export const descriptor: WorkflowDescriptor;
 export const run: WorkflowFn;

 type WorkflowFn = (
-  input: { prompt: string; steps: RoleOutput[] },
-  options: { threadId: string; maxRounds: number },
-) => AsyncGenerator<RoleOutput, WorkflowResult>;
+  thread: ThreadContext,
+  runtime: WorkflowRuntime,
+) => AsyncGenerator<RoleOutput, WorkflowCompletion>;
 ```

+`RoleOutput` carries `contentHash`, `meta`, and `refs` (agent text lives in CAS, addressed by hash).
+
 ### Constraints

 - Single `.esm.js` file
- No dynamic `import()`
- All static imports must be Node built-in modules only
- XXH64 hash (Crockford Base32) = globally unique version ID
+- No dynamic `import()` in bundles (loader exempt in engine)
+- Portable bundle static imports are constrained by validation in `@uncaged/workflow-register` (`validateWorkflowBundle`)
+- XXH64 hash (Crockford Base32) = version ID

 ### Why AsyncGenerator?

- Each `yield` → engine writes to `.data.jsonl`, checks abort/pause
- `return` → engine marks thread complete
- Fork = pass historical steps as `input.steps` to a new generator
- Zero injection — bundle doesn't import from the engine
+- Each `yield` lets `workflow-execute` persist state, CAS rows, and enforce pause/abort
+- `return` supplies `WorkflowCompletion`
+- Fork replays historical steps into a new thread context
+- Bundle does not import the engine — only protocol/runtime types at build time

-## Storage Layout
+## Storage layout

 ```
 ~/.uncaged/workflow/
+├── cas/                           # Global content-addressed blobs (see getGlobalCasDir)
 ├── bundles/
-│   ├── C9NMV6V2TQT81.esm.js     # Crockford Base32 of XXH64
-│   └── C9NMV6V2TQT81.yaml       # Role descriptor
+│   ├── C9NMV6V2TQT81.esm.js       # Crockford Base32 of XXH64
+│   └── C9NMV6V2TQT81.yaml         # Role descriptor sidecar (when present)
 ├── logs/                          # One folder per bundle hash
 │   └── C9NMV6V2TQT81/
 │       ├── 01KQXKW…YG.data.jsonl  # Thread state
@@ -173,7 +197,7 @@ type WorkflowFn = (
 └── workflow.yaml                  # Registry
 ```

-### ID Encoding: Crockford Base32
+### ID encoding: Crockford Base32

 - Case-insensitive, filesystem-safe, no ambiguous chars (0/O, 1/I/L)
 - Bundle hash: XXH64 → 13-char
@@ -181,45 +205,36 @@ type WorkflowFn = (

 ### Registry (`workflow.yaml`)

-```yaml
-workflows:
-  solve-issue:
-    hash: "C9NMV6V2TQT81"
-    timestamp: 1714963200000
-    history:
-      - hash: "A7BKR3M1NPQ40"
-        timestamp: 1714876800000
-```
+Managed by `@uncaged/workflow-register` (`readWorkflowRegistry`, `writeWorkflowRegistry`, …). Shape includes workflow entries and a top-level `config` section used for extract/supervisor model resolution.

 ### Thread JSONL

-**`.data.jsonl`** — Line 1: start record, Line 2+: role outputs
+**`.data.jsonl`** — Line 1: start record; following lines: role steps with CAS-backed content.

 ```jsonc
 // Start record
 { "name": "solve-issue", "hash": "C9NMV6V2TQT81", "threadId": "01KQXKW…",
  "parameters": { "prompt": "Fix bug #3", "options": { "maxRounds": 5 } },
  "timestamp": 1714963200000 }
-// Role output
-{ "role": "planner", "content": "...", "meta": { "phases": [...] }, "timestamp": ... }
+// Role output (engine persists contentHash + refs; body in ~/.uncaged/workflow/cas/)
+{ "role": "planner", "contentHash": "…", "meta": { "phases": [...] }, "refs": ["…"], "timestamp": ... }
 ```

-**`.info.jsonl`** — Structured debug log
+**`.info.jsonl`** — Structured debug log via `@uncaged/workflow-util` `createLogger`:

 ```jsonc
 { "tag": "4KNMR2PX", "content": "Loading bundle...", "timestamp": ... }
 ```

-Tags are 8-char Crockford Base32 (40-bit random), one per call site. `grep "4KNMR2PX"` → instant code location.
+Tags are 8-char Crockford Base32 (40-bit random), one per call site. `grep "4KNMR2PX"` → code location.

-## Execution Model
+## Execution model

- **No daemon.** `uncaged-workflow run <name>` starts a worker process
- Same bundle's threads share one process (memory efficiency)
- Process exits when all threads complete
- Thread termination via IPC within the process
+- **No daemon.** `uncaged-workflow run <name>` starts a worker process (`workflow-execute` worker entry via `getWorkerHostScriptPath`)
+- Threads share bundle-scoped workers as implemented in CLI/engine
+- Pause/resume/abort via engine IPC and pause gate (`createThreadPauseGate`)

-## CLI Commands
+## CLI commands

 | Priority | Command | Description |
 |----------|---------|-------------|
@@ -239,18 +254,16 @@ Tags are 8-char Crockford Base32 (40-bit random), one per call site. `grep "4KNM
 | P2 | `resume <thread-id>` | Resume a paused thread |
 | P3 | `fork <thread-id> [--from-role <role>]` | Fork from historical state |

-All commands implemented and tested. ✅
-
-## Design Decisions
+## Design decisions

 | Decision | Rationale |
 |----------|-----------|
 | **Role = pure data** | Decouples definition from execution; same role with different agents |
-| **Agent bound at runtime** | WorkflowDefinition is reusable; agent choice is deployment concern |
-| **Three-phase context** | Each phase sees only what it needs; clean separation |
-| **ExtractFn as general tool** | Agents use it for pre-execution extraction; engine uses it for meta |
-| **Single-file ESM** | Hash = version, no dependency hell, self-contained |
-| **No daemon** | OS handles process lifecycle; unnecessary complexity |
+| **Agent bound at runtime** | `WorkflowDefinition` is reusable; agent choice is deployment concern |
+| **Three-phase context** | Each phase sees only what it needs; types live in `workflow-protocol` |
+| **`WorkflowRuntime.extract` + CAS `contentHash`** | Large agent bodies deduplicated globally; Merkle roots summarize threads |
+| **`workflow-reactor` split** | LLM tool-calling loop isolated from filesystem/registry concerns |
+| **Single-file ESM** | Hash = version, self-contained bundle |
+| **No daemon** | OS handles process lifecycle |
 | **Crockford Base32** | Filesystem-safe, readable, compact |
-| **No concurrency in registry** | Different workflows have different constraints; belongs at workflow/role level |
-| **No dryRun** | Tests use mock agents + mock fetch; simpler architecture |
+| **15-package split** | Clear boundaries: protocol ↔ runtime author API ↔ util/CAS/register ↔ execute ↔ CLI ↔ agents/templates/UI |
@@ -1,5 +1,7 @@
 # Workflow-as-Agent Implementation Plan

+> ⚠️ This plan references the pre-split package structure. File paths have changed.
+
 > **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.
@@ -0,0 +1,262 @@
+# RFC: CAS-Based Thread Storage
+
+> Status: Draft
+> Author: 小橘 🍊（NEKO Team）
+> Date: 2026-05-09
+
+## Summary
+
+Replace `.data.jsonl` with a fully CAS-based thread state chain. Threads become linked lists of immutable CAS nodes, indexed by a per-bundle `threads.json`.
+
+## Motivation
+
+`.data.jsonl` is a flat append-only file with three different row formats (start, role step, end). This makes forking expensive (copy file), deduplication impossible (forked threads repeat shared history), and GC complex (must parse every row to find CAS refs).
+
+Threads are inherently immutable append-only sequences — a natural fit for CAS hash chains, similar to git's commit DAG.
+
+## Design
+
+### Node Types
+
+Two CAS node types, using the existing `{ type, payload, refs }` CAS blob structure:
+
+#### StartNode
+
+Contains workflow-level parameters. **No threadId** (because the same StartNode can be shared across forks). Prompt is stored as a CAS blob and referenced via `refs[0]`.
+
+```
+CAS blob:
+{
+  type: "start",
+  payload: {
+    name: "solve-issue",
+    hash: "BUNDLE_HASH",
+    maxRounds: 10,
+    depth: 0
+  },
+  refs: [
+    <prompt_hash>    // refs[0]: initial task prompt (CAS blob)
+  ]
+}
+```
+
+- No `role`, `content`, `meta` — this is not a step, it's workflow metadata
+- Prompt is **not** inline — it lives in CAS and is referenced by hash
+
+#### StateNode
+
+One per role step (including `__end__`).
+
+```
+CAS blob:
+{
+  type: "state",
+  payload: {
+    role: "coder",
+    meta: { ... },
+    start: "<start_hash>",
+    content: "<content_merkle_hash>",
+    ancestors: ["<parent_hash>", "<grandparent_hash>", ...],
+    compact: null,
+    timestamp: 1234567890
+  },
+  refs: [<start_hash>, <content_hash>, <parent_hash>, ...]
+}
+```
+
+**Payload is the source of truth.** Application code reads named fields from payload. `refs[]` is a **GC index** — automatically derived from payload by collecting all CAS hashes. GC only scans `refs[]` without understanding payload structure.
+
+**Payload fields:**
+
+| Field | Type | Meaning |
+|-------|------|---------|
+| `role` | `string` | Role name, or `"__end__"` for completion |
+| `meta` | `object` | Structured metadata extracted from agent output |
+| `start` | `string` | StartNode hash |
+| `content` | `string` | Content Merkle node hash (carries role artifact refs) |
+| `ancestors` | `string[]` | `[parent, grandparent, ...]` — up to 11 entries (1 parent + 10 skip-list). Empty for first step after start. `ancestors[0]` is the direct parent. |
+| `compact` | `string \| null` | CAS hash of a compacted summary of all nodes before this one. When present, LLM context assembly can use this instead of walking the full chain. |
+| `timestamp` | `number` | Unix timestamp in ms |
+
+### Content Merkle Node
+
+The content at `refs[2]` of each StateNode is itself a CAS Merkle node. This is where **role artifact references** live:
+
+```
+CAS blob:
+{
+  type: "content",
+  payload: "<role output text>",
+  refs: [
+    <artifact_hash_1>,   // e.g. a commit, a file, a sub-result
+    <artifact_hash_2>,
+    ...
+  ]
+}
+```
+
+The Extractor is responsible for producing both `meta` and `refs` from raw agent output:
+
+```
+Agent raw output
+    ↓
+Extractor → { meta, contentPayload, refs[] }
+    ↓
+CAS put content Merkle: { type: "content", payload: contentPayload, refs }
+    ↓ contentHash
+StateNode: { ..., refs: [start, parent, contentHash, ...ancestors] }
+```
+
+This keeps StateNode refs fixed and simple. All role-specific artifact references are encapsulated in the content Merkle node. GC follows: `thread head → StateNode.refs → content Merkle.refs → artifacts`, full chain recursive.
+
+### End Node
+
+An end is just a StateNode with `role: "__end__"`:
+
+```
+{
+  type: "state",
+  payload: {
+    role: "__end__",
+    meta: { returnCode: 0, summary: "completed successfully" },
+    start: "<start_hash>",
+    content: "<content_hash>",
+    ancestors: ["<parent_hash>", ...],
+    compact: null,
+    timestamp: 1234567891
+  },
+  refs: [<start_hash>, <content_hash>, <parent_hash>, ...]
+}
+```
+
+### Thread Index: `threads.json`
+
+Per-bundle directory, one `threads.json` file. **Only active (in-progress) threads** live here:
+
+```
+~/.uncaged/workflow/bundles/<hash>/threads.json
+```
+
+```json
+{
+  "01JTHREAD1AAAAAAAAAAAAAAA": {
+    "head": "<latest_state_node_hash>",
+    "start": "<start_node_hash>",
+    "updatedAt": 1234567891
+  }
+}
+```
+
+When a thread completes (`__end__`), it is **removed from `threads.json`** and appended to a date-partitioned history file:
+
+```
+~/.uncaged/workflow/bundles/<hash>/history/{YYYY-MM-DD}.jsonl
+```
+
+Each line:
+
+```json
+{"threadId":"01JTHREAD1AAAAAAAAAAAAAAA","head":"<end_node_hash>","start":"<start_node_hash>","completedAt":1234567891}
+```
+
+Benefits:
+- `threads.json` stays small — only in-flight threads
+- Dashboard watches `threads.json` for real-time updates; completed threads don't trigger watches
+- History is queryable by date but not actively monitored
+- GC roots = all heads from `threads.json` + all heads from `history/*.jsonl`
+
+### Ancestor Skip-List
+
+Each StateNode carries up to 11 entries in `payload.ancestors` (1 parent + 10 skip-list, newest first):
+
+```
+Node 15: ancestors = [node14, node13, node12, node11, node10, node9, node8, node7, node6, node5, node4]
+                      ^parent  ^--- skip-list (10 most recent) ---^
+```
+
+This enables:
+- **Paginated fetch**: jump to any recent ancestor without walking the full chain
+- **Partial replay**: fetch last N steps without loading the entire history
+- The list is capped at 10 to keep node size bounded
+
+### Fork
+
+Forking a thread at step N:
+
+1. Create new threadId
+2. Create a new StateNode whose `parent` (refs[1]) points to the fork point's StateNode
+3. Register the new threadId in `threads.json` with its own head
+4. **Zero data duplication** — the forked thread shares all ancestor nodes via CAS
+
+### Compact
+
+When a StateNode has `payload.compact` set:
+
+```json
+{
+  "type": "state",
+  "payload": {
+    "role": "coder",
+    "meta": { ... },
+    "compact": "<cas_hash_of_summary>",
+    "timestamp": 1234
+  },
+  "refs": [...]
+}
+```
+
+This means: "everything before this node has been summarized into the blob at `compact`". When building LLM context:
+
+1. Walk back from head
+2. If a node has `compact`, stop walking — use the compact summary + all nodes after it
+3. If no compact found, use full chain
+
+This enables long-running threads without unbounded context growth.
+
+### GC
+
+Simple mark-and-sweep:
+
+1. **Roots**: all `head` and `start` hashes from `threads.json` + all `history/*.jsonl` files
+2. **Mark**: from each root, recursively mark all reachable hashes via `refs[]` (including content Merkle → artifact refs)
+3. **Sweep**: delete unmarked CAS blobs
+
+No per-row format parsing needed. GC only needs to understand `refs[]`.
+
+### refs[] Derivation
+
+`refs[]` is auto-derived from payload at write time via a `collectRefs(payload)` function that extracts all CAS hash strings from named fields (`start`, `content`, `ancestors`, `compact`). Application code never reads `refs[]` — it reads named payload fields. This makes `refs[]` a pure GC optimization with zero semantic coupling.
+
+### Extract Phase
+
+The Extractor is expanded from the current design. Currently it only extracts `meta` from agent output. In the new design it extracts:
+
+| Output | Purpose |
+|--------|---------|
+| `meta` | Structured metadata (same as before) |
+| `contentPayload` | The text payload for the content Merkle node |
+| `refs[]` | CAS hashes of artifacts produced by this role step |
+
+The `refs[]` become the content Merkle node's refs, enabling GC to trace all role-produced artifacts.
+
+## What Stays Unchanged
+
+- `.info.jsonl` — debug logging stays as-is (high-frequency append, not suitable for CAS)
+- CAS blob storage format (`~/.uncaged/workflow/cas/`)
+- Bundle registry (`workflow.yaml`)
+
+## Migration
+
+Breaking change. Old `.data.jsonl` files become incompatible. No backward compat fallback (per project convention).
+
+## Changes by Package
+
+| Package | Changes |
+|---------|---------|
+| `workflow-protocol` | Replace `StartStep`, `RoleStep` types with `StartNode`, `StateNode`. Add `ContentMerkleNode` type. Expand `ExtractResult` to include `refs[]`. |
+| `workflow-cas` | Add `findReachableHashes(roots)` for GC mark phase |
+| `workflow-execute` | Rewrite engine to write CAS nodes + update `threads.json` instead of appending JSONL. Move completed threads to `history/`. Simplify `gc.ts`. Simplify `fork-thread.ts`. Expand extract phase to produce refs. |
+| `workflow-runtime` | `ThreadContext` built by walking chain from head. `start.prompt` resolved from CAS via StartNode.refs[0]. |
+| `cli-workflow` | `thread list/show/rm` read from `threads.json` + `history/`. SSE watches `threads.json`. |
+| `workflow-dashboard` | Watch `threads.json` instead of `.data.jsonl` |
+| Templates & Agents | Update extract definitions to produce `refs[]`. Update `ctx.start.content` → CAS resolved. |
@@ -3,13 +3,9 @@ import { mkdir, mkdtemp, readFile, rm, unlink, writeFile } from "node:fs/promise
 import { tmpdir } from "node:os";
 import { join } from "node:path";

-import {
-  createContentMerkleNode,
-  getGlobalCasDir,
-  getRegisteredWorkflow,
-  readWorkflowRegistry,
-  serializeMerkleNode,
-} from "@uncaged/workflow";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createContentMerkleNode, serializeMerkleNode } from "@uncaged/workflow-cas";
+import { getRegisteredWorkflow, readWorkflowRegistry } from "@uncaged/workflow-register";
 import { cmdCasGet, cmdCasList, cmdCasPut, cmdCasRm } from "../src/commands/cas/index.js";
 import {
  cmdAdd,
@@ -25,7 +21,7 @@ import { addCliArgs } from "./bundle-fixture.js";
 const fixtureDescriptor = `export const descriptor = { description: "fixture", roles: {} };
 `;

-const wfPutImport = `import { putContentMerkleNode } from "@uncaged/workflow";
+const wfPutImport = `import { putContentMerkleNode } from "@uncaged/workflow-cas";
 `;

 function casStoredForm(raw: string): string {
@@ -2,7 +2,8 @@ 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 { createCasStore, getContentMerklePayload } from "@uncaged/workflow-cas";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
 import { cmdFork, cmdRun } from "../src/commands/thread/index.js";
 import { cmdAdd } from "../src/commands/workflow/index.js";
 import { pathExists } from "../src/fs-utils.js";
@@ -10,7 +11,7 @@ import { addCliArgs } from "./bundle-fixture.js";
 import { ensureTestWorkflowRegistryConfig } from "./workflow-registry-fixture.js";

 /** Three-role workflow that respects `input.steps` for fork/resume. */
-const threeRoleBundleSource = `import { putContentMerkleNode } from "@uncaged/workflow";
+const threeRoleBundleSource = `import { putContentMerkleNode } from "@uncaged/workflow-cas";

 export const descriptor = {
  description: "fork-cli",
@@ -4,12 +4,9 @@ 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 { createCasStore, putContentMerkleNode } from "@uncaged/workflow-cas";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { garbageCollectCas } from "@uncaged/workflow-execute";
 import { cmdThreadRemove } from "../src/commands/thread/index.js";
 import { pathExists } from "../src/fs-utils.js";

@@ -50,7 +50,6 @@ describe("init template", () => {
      dependencies: Record<string, string>;
    };
    expect(pkg.type).toBe("module");
-    expect(pkg.dependencies["@uncaged/workflow"]).toBeDefined();
    expect(pkg.dependencies["@uncaged/workflow-runtime"]).toBeDefined();
    expect(pkg.dependencies.zod).toBeDefined();
    expect(pkg.name).toContain("review-pr");
@@ -46,7 +46,7 @@ describe("init workspace", () => {
      dependencies: Record<string, string>;
    };
    expect(wfPkg.type).toBe("module");
-    expect(wfPkg.dependencies["@uncaged/workflow"]).toBeDefined();
+    expect(wfPkg.dependencies["@uncaged/workflow-runtime"]).toBeDefined();
    expect(wfPkg.dependencies.zod).toBeDefined();

    const tsconfig = JSON.parse(await readFile(join(root, "tsconfig.json"), "utf8")) as {
@@ -5,7 +5,8 @@ import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { fileURLToPath } from "node:url";

-import { createCasStore, getGlobalCasDir, putContentMerkleNode } from "@uncaged/workflow";
+import { createCasStore, putContentMerkleNode } from "@uncaged/workflow-cas";
+import { getGlobalCasDir } from "@uncaged/workflow-util";

 import {
  formatLiveDebugLine,
@@ -1,6 +1,6 @@
 import { describe, expect, test } from "bun:test";

-import { createContentMerkleNode, serializeMerkleNode } from "@uncaged/workflow";
+import { createContentMerkleNode, serializeMerkleNode } from "@uncaged/workflow-cas";

 import { createApp } from "../src/commands/serve/app.js";

@@ -77,6 +77,83 @@ describe("serve /api/cas", () => {
  });
 });

+describe("serve error handling", () => {
+  test("POST /api/threads with invalid JSON body → 400", async () => {
+    const { fetch } = buildApp("/tmp/uncaged-serve-test-nonexistent");
+    const res = await fetch("/api/threads", {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: "not json",
+    });
+    expect(res.status).toBe(400);
+    const body = (await res.json()) as { error: string };
+    expect(body.error).toBe("invalid JSON body");
+  });
+
+  test("POST /api/cas with invalid JSON body → 400", async () => {
+    const { fetch } = buildApp("/tmp/uncaged-serve-test-nonexistent");
+    const res = await fetch("/api/cas", {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: "not json",
+    });
+    expect(res.status).toBe(400);
+    const body = (await res.json()) as { error: string };
+    expect(body.error).toBe("invalid JSON body");
+  });
+
+  test("POST /api/threads with missing required fields → 400", async () => {
+    const { fetch } = buildApp("/tmp/uncaged-serve-test-nonexistent");
+    const res = await fetch("/api/threads", {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ foo: "bar" }),
+    });
+    expect(res.status).toBe(400);
+    const body = (await res.json()) as { error: string };
+    expect(body.error).toContain("required");
+  });
+
+  test("global error handler returns 500 with JSON", async () => {
+    const app = createApp("/tmp/uncaged-serve-test-nonexistent");
+    app.get("/test-error", () => {
+      throw new Error("boom");
+    });
+    const res = await app.fetch(new Request("http://localhost/test-error"));
+    expect(res.status).toBe(500);
+    const body = (await res.json()) as { error: string };
+    expect(body.error).toBe("Internal server error");
+  });
+});
+
+describe("serve security", () => {
+  test("CORS headers present on responses", async () => {
+    const app = createApp("/tmp/uncaged-serve-test-nonexistent");
+    const res2 = await app.fetch(
+      new Request("http://localhost/healthz", {
+        headers: { Origin: "http://localhost:5173" },
+      }),
+    );
+    expect(res2.headers.get("Access-Control-Allow-Origin")).toBe("http://localhost:5173");
+  });
+
+  test("POST with body > 1MB → 413", async () => {
+    const { fetch } = buildApp("/tmp/uncaged-serve-test-nonexistent");
+    const largeBody = "x".repeat(1_048_577);
+    const res = await fetch("/api/cas", {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        "Content-Length": String(largeBody.length),
+      },
+      body: largeBody,
+    });
+    expect(res.status).toBe(413);
+    const body = (await res.json()) as { error: string };
+    expect(body.error).toBe("Payload too large");
+  });
+});
+
 describe("serve CAS round-trip", () => {
  const tmpDir = `/tmp/uncaged-serve-cas-test-${Date.now()}`;

@@ -1,5 +1,5 @@
 import { afterEach, beforeEach, describe, expect, test } from "bun:test";
-import { getDefaultWorkflowStorageRoot } from "@uncaged/workflow";
+import { getDefaultWorkflowStorageRoot } from "@uncaged/workflow-util";
 import { resolveWorkflowStorageRoot } from "../src/storage-env.js";

 describe("resolveWorkflowStorageRoot", () => {
@@ -4,7 +4,7 @@ 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 { getGlobalCasDir } from "@uncaged/workflow-util";
 import { cmdCasPut } from "../src/commands/cas/index.js";
 import {
  cmdKill,
@@ -21,7 +21,7 @@ import { pathExists, readTextFileIfExists } from "../src/fs-utils.js";
 import { addCliArgs } from "./bundle-fixture.js";
 import { ensureTestWorkflowRegistryConfig } from "./workflow-registry-fixture.js";

-const wfPutImport = `import { putContentMerkleNode } from "@uncaged/workflow";
+const wfPutImport = `import { putContentMerkleNode } from "@uncaged/workflow-cas";
 `;

 const threadFixtureDescriptor = `export const descriptor = {
@@ -6,8 +6,12 @@
    "uncaged-workflow": "src/cli.ts"
  },
  "dependencies": {
+    "@uncaged/workflow-protocol": "workspace:*",
+    "@uncaged/workflow-util": "workspace:*",
+    "@uncaged/workflow-cas": "workspace:*",
+    "@uncaged/workflow-execute": "workspace:*",
+    "@uncaged/workflow-register": "workspace:*",
    "@uncaged/workflow-runtime": "workspace:*",
-    "@uncaged/workflow": "workspace:*",
    "hono": "^4.12.18",
    "yaml": "^2.8.4"
  },
@@ -1,7 +1,7 @@
 import { mkdir, readFile, writeFile } from "node:fs/promises";
 import { join } from "node:path";

-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import { pathExists } from "./fs-utils.js";

@@ -9,9 +9,6 @@ import { createThreadDispatcher, dispatchLive, dispatchRun } from "./commands/th
 import { createWorkflowDispatcher } from "./commands/workflow/index.js";
 import { formatSkillIndex, formatSkillTopic, getSkillTopics } from "./skill.js";

-export type { CommandEntry, CommandGroup, DispatchFn } from "./cli-command-types.js";
-export { getCommandRegistry } from "./cli-registry.js";
-
 function dispatchGroup(
  tableName: string,
  table: Record<string, CommandEntry>,
@@ -1,4 +1,5 @@
-import { type GcResult, garbageCollectCas, type Result } from "@uncaged/workflow";
+import type { Result } from "@uncaged/workflow-protocol";
+import { type GcResult, garbageCollectCas } from "@uncaged/workflow-execute";

 export async function cmdGc(storageRoot: string): Promise<Result<GcResult, string>> {
  return garbageCollectCas(storageRoot);
@@ -1,4 +1,6 @@
-import { createCasStore, err, getGlobalCasDir, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createCasStore } from "@uncaged/workflow-cas";

 export async function cmdCasGet(
  storageRoot: string,
@@ -1,4 +1,6 @@
-import { createCasStore, getGlobalCasDir, ok, type Result } from "@uncaged/workflow";
+import { ok, type Result } from "@uncaged/workflow-protocol";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createCasStore } from "@uncaged/workflow-cas";

 export async function cmdCasList(storageRoot: string): Promise<Result<string[], string>> {
  const cas = createCasStore(getGlobalCasDir(storageRoot));
@@ -1,4 +1,6 @@
-import { createCasStore, getGlobalCasDir, ok, type Result } from "@uncaged/workflow";
+import { ok, type Result } from "@uncaged/workflow-protocol";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createCasStore } from "@uncaged/workflow-cas";

 export async function cmdCasPut(
  storageRoot: string,
@@ -1,4 +1,6 @@
-import { createCasStore, getGlobalCasDir, ok, type Result } from "@uncaged/workflow";
+import { ok, type Result } from "@uncaged/workflow-protocol";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createCasStore } from "@uncaged/workflow-cas";

 export async function cmdCasRm(storageRoot: string, hash: string): Promise<Result<void, string>> {
  const cas = createCasStore(getGlobalCasDir(storageRoot));
@@ -1,7 +1,7 @@
 import { mkdir, readFile, writeFile } from "node:fs/promises";
 import { dirname, join, resolve } from "node:path";

-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import { pathExists } from "../../fs-utils.js";

@@ -6,7 +6,6 @@ export function templatePackageJson(templateName: string): string {
      private: true,
      type: "module",
      dependencies: {
-        "@uncaged/workflow": "^0.1.0",
        "@uncaged/workflow-runtime": "^0.1.0",
        zod: "^4.0.0",
      },
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 /** Validates a single path segment for workspace / template names (no separators, not `.` / `..`). */
 export function validateWorkspaceSegment(name: string): Result<void, string> {
@@ -1,7 +1,7 @@
 import { mkdir, writeFile } from "node:fs/promises";
 import { join } from "node:path";

-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import { pathExists } from "../../fs-utils.js";
 import type { CmdInitWorkspaceSuccess } from "./types.js";
@@ -28,7 +28,7 @@ function workflowsPackageJson(): string {
      private: true,
      type: "module",
      dependencies: {
-        "@uncaged/workflow": "^0.1.0",
+        "@uncaged/workflow-runtime": "^0.1.0",
        zod: "^4.0.0",
      },
    },
@@ -107,7 +107,7 @@ Init 生成的骨架：\`templates/\` 下放可复用定义，\`workflows/\` 下
 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)\`（或项目约定的封装）绑定 **AgentFn**；**ExtractFn** 由引擎从 **workflow.yaml** 注入 \`WorkflowFnOptions\`。
+5. **实例化**：在 workflow 包中使用 \`createWorkflow(def, binding)\`（或项目约定的封装）绑定 **AgentFn**；**ExtractFn** 由引擎从 **workflow.yaml** 注入 \`WorkflowRuntime\`。
 6. **构建**：打包为单个 **.esm.js** bundle，使用 **uncaged-workflow add** 注册。

 ## 4. 编码规范
@@ -2,18 +2,46 @@ import { Hono } from "hono";
 import { cors } from "hono/cors";

 import { createCasRoutes } from "./routes-cas.js";
+import { createLiveRoutes } from "./routes-live.js";
 import { createThreadRoutes } from "./routes-thread.js";
 import { createWorkflowRoutes } from "./routes-workflow.js";

+const MAX_BODY_SIZE = 1_048_576; // 1 MB
+
 export function createApp(storageRoot: string): Hono {
  const app = new Hono();

-  app.use("*", cors());
+  app.onError((_err, c) => {
+    return c.json({ error: "Internal server error" }, 500);
+  });
+
+  app.use(
+    "*",
+    cors({
+      origin: [
+        "http://localhost:5173",
+        "http://127.0.0.1:5173",
+        "http://localhost:7860",
+        "http://127.0.0.1:7860",
+      ],
+    }),
+  );
+
+  app.use("*", async (c, next) => {
+    if (c.req.method === "POST") {
+      const contentLength = c.req.header("content-length");
+      if (contentLength !== undefined && Number(contentLength) > MAX_BODY_SIZE) {
+        return c.json({ error: "Payload too large" }, 413);
+      }
+    }
+    await next();
+  });

  app.get("/healthz", (c) => c.json({ ok: true }));

  app.route("/api/workflows", createWorkflowRoutes(storageRoot));
  app.route("/api/threads", createThreadRoutes(storageRoot));
+  app.route("/api/threads", createLiveRoutes(storageRoot));
  app.route("/api/cas", createCasRoutes(storageRoot));

  return app;
@@ -1,19 +1,19 @@
-import { createCasStore, garbageCollectCas, getGlobalCasDir } from "@uncaged/workflow";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createCasStore } from "@uncaged/workflow-cas";
+import { garbageCollectCas } from "@uncaged/workflow-execute";
 import { Hono } from "hono";

 export function createCasRoutes(storageRoot: string): Hono {
  const app = new Hono();
+  const casDir = getGlobalCasDir(storageRoot);
+  const cas = createCasStore(casDir);

  app.get("/", async (c) => {
-    const casDir = getGlobalCasDir(storageRoot);
-    const cas = createCasStore(casDir);
    const hashes = await cas.list();
    return c.json({ hashes });
  });

  app.get("/:hash", async (c) => {
-    const casDir = getGlobalCasDir(storageRoot);
-    const cas = createCasStore(casDir);
    const content = await cas.get(c.req.param("hash"));
    if (content === null) {
      return c.json({ error: "not found" }, 404);
@@ -22,19 +22,20 @@ export function createCasRoutes(storageRoot: string): Hono {
  });

  app.post("/", async (c) => {
-    const body = await c.req.json<{ content: string }>();
+    let body: { content: string };
+    try {
+      body = (await c.req.json()) as { content: string };
+    } catch {
+      return c.json({ error: "invalid JSON body" }, 400);
+    }
    if (typeof body.content !== "string") {
      return c.json({ error: "content field required" }, 400);
    }
-    const casDir = getGlobalCasDir(storageRoot);
-    const cas = createCasStore(casDir);
    const hash = await cas.put(body.content);
    return c.json({ hash }, 201);
  });

  app.delete("/:hash", async (c) => {
-    const casDir = getGlobalCasDir(storageRoot);
-    const cas = createCasStore(casDir);
    const hash = c.req.param("hash");
    const content = await cas.get(hash);
    if (content === null) {
@@ -0,0 +1,198 @@
+import { statSync, watch } from "node:fs";
+import { dirname, join } from "node:path";
+import { Hono } from "hono";
+import { streamSSE } from "hono/streaming";
+
+import { resolveThreadDataPath } from "../../thread-scan.js";
+
+type PumpState = {
+  contentOffset: number;
+  carry: string;
+};
+
+function fileSize(path: string): number {
+  try {
+    return statSync(path).size;
+  } catch {
+    return 0;
+  }
+}
+
+async function readNewBytes(path: string, state: PumpState): Promise<string | null> {
+  const size = fileSize(path);
+  if (size < state.contentOffset) {
+    // File was truncated — reset
+    state.contentOffset = 0;
+    state.carry = "";
+  }
+  if (size <= state.contentOffset) {
+    return null;
+  }
+  const blob = Bun.file(path).slice(state.contentOffset, size);
+  const chunk = await blob.text();
+  state.contentOffset = size;
+  return chunk;
+}
+
+function parseJsonLine(line: string): unknown {
+  try {
+    return JSON.parse(line) as unknown;
+  } catch {
+    return { raw: line };
+  }
+}
+
+function isWorkflowResult(record: unknown): boolean {
+  return (
+    record !== null &&
+    typeof record === "object" &&
+    "type" in (record as Record<string, unknown>) &&
+    (record as Record<string, unknown>).type === "workflow-result"
+  );
+}
+
+function parseNewLines(chunk: string, state: PumpState): string[] {
+  state.carry += chunk;
+
+  const parts = state.carry.split("\n");
+  state.carry = parts.pop() ?? "";
+
+  const lines: string[] = [];
+  for (const line of parts) {
+    const trimmed = line.trim();
+    if (trimmed !== "") {
+      lines.push(trimmed);
+    }
+  }
+  return lines;
+}
+
+export function createLiveRoutes(storageRoot: string): Hono {
+  const app = new Hono();
+
+  app.get("/:threadId/live", async (c) => {
+    const threadId = c.req.param("threadId");
+    const dataPath = await resolveThreadDataPath(storageRoot, threadId);
+    if (dataPath === null) {
+      return c.json({ error: `thread not found: ${threadId}` }, 404);
+    }
+    const resolvedDataPath = dataPath;
+
+    const infoPath = join(dirname(resolvedDataPath), `${threadId}.info.jsonl`);
+
+    return streamSSE(c, async (stream) => {
+      const dataState: PumpState = { contentOffset: 0, carry: "" };
+      const infoState: PumpState = { contentOffset: 0, carry: "" };
+      let eventId = 0;
+
+      async function pumpData(): Promise<boolean> {
+        let chunk: string | null;
+        try {
+          chunk = await readNewBytes(resolvedDataPath, dataState);
+        } catch {
+          return false;
+        }
+        if (chunk === null) {
+          return false;
+        }
+
+        const lines = parseNewLines(chunk, dataState);
+        for (const line of lines) {
+          const record = parseJsonLine(line);
+          eventId++;
+          await stream.writeSSE({
+            event: "record",
+            data: JSON.stringify(record),
+            id: String(eventId),
+          });
+
+          if (isWorkflowResult(record)) {
+            return true;
+          }
+        }
+        return false;
+      }
+
+      async function pumpInfo(): Promise<void> {
+        let chunk: string | null;
+        try {
+          chunk = await readNewBytes(infoPath, infoState);
+        } catch {
+          return;
+        }
+        if (chunk === null) {
+          return;
+        }
+
+        const lines = parseNewLines(chunk, infoState);
+        for (const line of lines) {
+          const record = parseJsonLine(line);
+          if (
+            typeof record === "object" &&
+            record !== null &&
+            "raw" in (record as Record<string, unknown>)
+          ) {
+            continue;
+          }
+          eventId++;
+          await stream.writeSSE({
+            event: "info",
+            data: JSON.stringify(record),
+            id: String(eventId),
+          });
+        }
+      }
+
+      // Initial pump
+      const done = await pumpData();
+      await pumpInfo();
+      if (done) {
+        return;
+      }
+
+      // Watch for changes
+      const controller = new AbortController();
+      let completed = false;
+
+      const dataWatcher = watch(resolvedDataPath, async () => {
+        if (completed) return;
+        const finished = await pumpData();
+        if (finished) {
+          completed = true;
+          controller.abort();
+        }
+      });
+
+      let infoWatcher: ReturnType<typeof watch> | null = null;
+      try {
+        infoWatcher = watch(infoPath, async () => {
+          if (completed) return;
+          await pumpInfo();
+        });
+      } catch {
+        // info file may not exist
+      }
+
+      stream.onAbort(() => {
+        completed = true;
+        dataWatcher.close();
+        infoWatcher?.close();
+      });
+
+      // Keep stream alive until completion or client disconnect
+      await new Promise<void>((resolve) => {
+        if (completed) {
+          resolve();
+          return;
+        }
+        controller.signal.addEventListener("abort", () => resolve(), { once: true });
+        stream.onAbort(() => resolve());
+      });
+
+      dataWatcher.close();
+      infoWatcher?.close();
+    });
+  });
+
+  return app;
+}
@@ -6,6 +6,8 @@ import {
  listRunningThreads,
  resolveThreadDataPath,
 } from "../../thread-scan.js";
+import { cmdKill, cmdPause, cmdResume } from "../thread/control.js";
+import { cmdRun } from "../thread/run.js";

 export function createThreadRoutes(storageRoot: string): Hono {
  const app = new Hono();
@@ -42,5 +44,55 @@ export function createThreadRoutes(storageRoot: string): Hono {
    return c.json({ threadId, records });
  });

+  app.post("/", async (c) => {
+    let body: Record<string, unknown>;
+    try {
+      body = (await c.req.json()) as Record<string, unknown>;
+    } catch {
+      return c.json({ error: "invalid JSON body" }, 400);
+    }
+
+    const name = body.workflow;
+    const prompt = body.prompt;
+    const maxRounds = typeof body.maxRounds === "number" ? body.maxRounds : 10;
+
+    if (typeof name !== "string" || typeof prompt !== "string") {
+      return c.json({ error: "workflow (string) and prompt (string) are required" }, 400);
+    }
+
+    const result = await cmdRun(storageRoot, name, prompt, maxRounds);
+    if (!result.ok) {
+      return c.json({ error: result.error }, 400);
+    }
+    return c.json({ threadId: result.value.threadId }, 201);
+  });
+
+  app.post("/:threadId/kill", async (c) => {
+    const threadId = c.req.param("threadId");
+    const result = await cmdKill(storageRoot, threadId);
+    if (!result.ok) {
+      return c.json({ error: result.error }, 400);
+    }
+    return c.json({ ok: true });
+  });
+
+  app.post("/:threadId/pause", async (c) => {
+    const threadId = c.req.param("threadId");
+    const result = await cmdPause(storageRoot, threadId);
+    if (!result.ok) {
+      return c.json({ error: result.error }, 400);
+    }
+    return c.json({ ok: true });
+  });
+
+  app.post("/:threadId/resume", async (c) => {
+    const threadId = c.req.param("threadId");
+    const result = await cmdResume(storageRoot, threadId);
+    if (!result.ok) {
+      return c.json({ error: result.error }, 400);
+    }
+    return c.json({ ok: true });
+  });
+
  return app;
 }
@@ -2,7 +2,7 @@ import {
  getRegisteredWorkflow,
  listRegisteredWorkflowNames,
  readWorkflowRegistry,
-} from "@uncaged/workflow";
+} from "@uncaged/workflow-register";
 import { Hono } from "hono";

 export function createWorkflowRoutes(storageRoot: string): Hono {
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
 import { serve } from "bun";

 import { printCliLine } from "../../cli-output.js";
@@ -1,4 +1,4 @@
-import type { Result } from "@uncaged/workflow";
+import type { Result } from "@uncaged/workflow-protocol";

 import {
  readWorkerCtl,
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import type { ParsedForkArgv } from "./types.js";

@@ -1,6 +1,8 @@
 import { join } from "node:path";

-import { buildForkPlan, err, generateUlid, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { generateUlid } from "@uncaged/workflow-util";
+import { buildForkPlan } from "@uncaged/workflow-execute";

 import { pathExists, readTextFileIfExists } from "../../fs-utils.js";
 import { resolveThreadDataPath } from "../../thread-scan.js";
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import { listHistoricalThreads } from "../../thread-scan.js";
 import { validateCliWorkflowName } from "../../workflow-name.js";
@@ -2,15 +2,10 @@ 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,
-} from "@uncaged/workflow";
-import type { WorkflowCompletion } from "@uncaged/workflow-runtime";
+import type { CasStore, WorkflowCompletion } from "@uncaged/workflow-protocol";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+import { createCasStore, getContentMerklePayload } from "@uncaged/workflow-cas";
+import { tryParseRoleStepRecord, tryParseWorkflowResultRecord } from "@uncaged/workflow-execute";

 import { dimGreyLine, highlightLiveRole } from "../../cli-color.js";
 import { printCliError, printCliLine } from "../../cli-output.js";
@@ -1,7 +1,8 @@
 import { unlink } from "node:fs/promises";
 import { dirname, join } from "node:path";

-import { err, garbageCollectCas, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { garbageCollectCas } from "@uncaged/workflow-execute";

 import { resolveThreadDataPath } from "../../thread-scan.js";

@@ -1,13 +1,8 @@
 import { join } from "node:path";

-import {
-  err,
-  generateUlid,
-  getRegisteredWorkflow,
-  ok,
-  type Result,
-  readWorkflowRegistry,
-} from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { generateUlid } from "@uncaged/workflow-util";
+import { getRegisteredWorkflow, readWorkflowRegistry } from "@uncaged/workflow-register";
 import { ensureWorkerForHash, sendWorkerTcpCommand } from "../../worker-spawn.js";
 import { validateCliWorkflowName } from "../../workflow-name.js";

@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import { readTextFileIfExists } from "../../fs-utils.js";
 import { resolveThreadDataPath } from "../../thread-scan.js";
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 import type { ParsedAddArgv } from "./types.js";

@@ -1,18 +1,16 @@
 import { readFile, stat } from "node:fs/promises";
 import { basename, resolve } from "node:path";

+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { hashWorkflowBundleBytes } from "@uncaged/workflow-cas";
 import {
-  err,
  extractBundleExports,
-  hashWorkflowBundleBytes,
-  ok,
-  type Result,
  readWorkflowRegistry,
  registerWorkflowVersion,
  stringifyWorkflowDescriptor,
  validateWorkflowBundle,
  writeWorkflowRegistry,
-} from "@uncaged/workflow";
+} from "@uncaged/workflow-register";

 import { storeWorkflowBundleArtifacts } from "../../bundle-store.js";
 import { validateCliWorkflowName } from "../../workflow-name.js";
@@ -1,10 +1,5 @@
-import {
-  err,
-  getRegisteredWorkflow,
-  ok,
-  type Result,
-  readWorkflowRegistry,
-} from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { getRegisteredWorkflow, readWorkflowRegistry } from "@uncaged/workflow-register";

 import { validateCliWorkflowName } from "../../workflow-name.js";

@@ -1,11 +1,9 @@
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
 import {
-  err,
  listRegisteredWorkflowNames,
-  ok,
-  type Result,
  readWorkflowRegistry,
  type WorkflowRegistryFile,
-} from "@uncaged/workflow";
+} from "@uncaged/workflow-register";

 export async function cmdList(storageRoot: string): Promise<Result<WorkflowRegistryFile, string>> {
  const reg = await readWorkflowRegistry(storageRoot);
@@ -1,11 +1,9 @@
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
 import {
-  err,
-  ok,
-  type Result,
  readWorkflowRegistry,
  unregisterWorkflow,
  writeWorkflowRegistry,
-} from "@uncaged/workflow";
+} from "@uncaged/workflow-register";

 import { validateCliWorkflowName } from "../../workflow-name.js";

@@ -1,14 +1,12 @@
 import { join } from "node:path";

+import { err, ok, type Result } from "@uncaged/workflow-protocol";
 import {
-  err,
  getRegisteredWorkflow,
-  ok,
-  type Result,
  readWorkflowRegistry,
  rollbackWorkflowToHistoryHash,
  writeWorkflowRegistry,
-} from "@uncaged/workflow";
+} from "@uncaged/workflow-register";

 import { pathExists } from "../../fs-utils.js";
 import { validateCliWorkflowName } from "../../workflow-name.js";
@@ -1,11 +1,9 @@
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
 import {
-  err,
  getRegisteredWorkflow,
-  ok,
-  type Result,
  readWorkflowRegistry,
  type WorkflowRegistryEntry,
-} from "@uncaged/workflow";
+} from "@uncaged/workflow-register";
 import { stringify } from "yaml";

 import { validateCliWorkflowName } from "../../workflow-name.js";
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 export type ParsedLiveArgv = {
  threadId: string | null;
@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 export type ParsedRunArgv = {
  name: string;
@@ -203,7 +203,6 @@ Each role has:
 | \`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

@@ -1,4 +1,4 @@
-import { getDefaultWorkflowStorageRoot } from "@uncaged/workflow";
+import { getDefaultWorkflowStorageRoot } from "@uncaged/workflow-util";

 /**
 * Resolve storage root with env var override support.
@@ -3,7 +3,8 @@ import { mkdir, readdir, unlink, writeFile } from "node:fs/promises";
 import { createConnection } from "node:net";
 import { join } from "node:path";

-import { err, getWorkerHostScriptPath, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";
+import { getWorkerHostScriptPath } from "@uncaged/workflow-execute";

 import { pathExists, readTextFileIfExists } from "./fs-utils.js";

@@ -1,4 +1,4 @@
-import { err, ok, type Result } from "@uncaged/workflow";
+import { err, ok, type Result } from "@uncaged/workflow-protocol";

 const WORKFLOW_NAME_RE = /^[a-z][a-z0-9]*(-[a-z0-9]+)*$/;

@@ -17,6 +17,13 @@
    "rootDir": "src",
    "types": ["bun-types"]
  },
-  "references": [{ "path": "../workflow-runtime" }, { "path": "../workflow" }],
+  "references": [
+    { "path": "../workflow-runtime" },
+    { "path": "../workflow-protocol" },
+    { "path": "../workflow-util" },
+    { "path": "../workflow-cas" },
+    { "path": "../workflow-execute" },
+    { "path": "../workflow-register" }
+  ],
  "include": ["src/**/*.ts"]
 }
@@ -1,31 +0,0 @@
-import { useState } from "react";
-import { Sidebar } from "./components/sidebar.tsx";
-import { ThreadList } from "./components/thread-list.tsx";
-import { ThreadDetail } from "./components/thread-detail.tsx";
-import { WorkflowList } from "./components/workflow-list.tsx";
-import { StatusBar } from "./components/status-bar.tsx";
-
-type View = "threads" | "workflows";
-
-export function App() {
-  const [view, setView] = useState<View>("threads");
-  const [selectedThread, setSelectedThread] = useState<string | null>(null);
-
-  return (
-    <div className="flex h-screen">
-      <Sidebar view={view} onViewChange={setView} />
-      <main className="flex-1 overflow-hidden flex flex-col">
-        <StatusBar />
-        <div className="flex-1 overflow-auto p-6">
-          {view === "threads" && !selectedThread && (
-            <ThreadList onSelect={setSelectedThread} />
-          )}
-          {view === "threads" && selectedThread && (
-            <ThreadDetail threadId={selectedThread} onBack={() => setSelectedThread(null)} />
-          )}
-          {view === "workflows" && <WorkflowList />}
-        </div>
-      </main>
-    </div>
-  );
-}
@@ -1,24 +0,0 @@
-import { getHealth } from "../api.ts";
-import { useFetch } from "../hooks.ts";
-
-export function StatusBar() {
-  const health = useFetch(() => getHealth(), []);
-
-  return (
-    <div
-      className="flex items-center justify-between px-6 py-2 text-xs border-b"
-      style={{ borderColor: "var(--color-border)", background: "var(--color-surface)" }}
-    >
-      <span style={{ color: "var(--color-text-muted)" }}>Local API: 127.0.0.1:7860</span>
-      <span>
-        {health.status === "loading" && "⏳ Connecting..."}
-        {health.status === "ok" && (
-          <span style={{ color: "var(--color-success)" }}>● Connected</span>
-        )}
-        {health.status === "error" && (
-          <span style={{ color: "var(--color-error)" }}>● Offline</span>
-        )}
-      </span>
-    </div>
-  );
-}
@@ -1,62 +0,0 @@
-import { getThread } from "../api.ts";
-import { useFetch } from "../hooks.ts";
-
-type Props = {
-  threadId: string;
-  onBack: () => void;
-};
-
-export function ThreadDetail({ threadId, onBack }: Props) {
-  const { status, data, error } = useFetch(() => getThread(threadId), [threadId]);
-
-  return (
-    <div>
-      <button
-        onClick={onBack}
-        className="text-sm mb-4 hover:underline"
-        style={{ color: "var(--color-accent)" }}
-      >
-        ← Back to threads
-      </button>
-      <h2 className="text-xl font-semibold mb-4 font-mono">{threadId}</h2>
-
-      {status === "loading" && <p style={{ color: "var(--color-text-muted)" }}>Loading...</p>}
-      {status === "error" && <p style={{ color: "var(--color-error)" }}>Error: {error}</p>}
-      {status === "ok" && (
-        <div className="space-y-3">
-          {data.records.map((r, i) => (
-            <div
-              key={i}
-              className="p-3 rounded border text-sm"
-              style={{ background: "var(--color-surface)", borderColor: "var(--color-border)" }}
-            >
-              <div className="flex items-center gap-2 mb-1">
-                <span
-                  className="text-xs px-1.5 py-0.5 rounded font-mono"
-                  style={{ background: "var(--color-border)", color: "var(--color-accent)" }}
-                >
-                  {r.type}
-                </span>
-                {r.role && (
-                  <span className="text-xs" style={{ color: "var(--color-text-muted)" }}>
-                    {r.role}
-                  </span>
-                )}
-                {r.timestamp && (
-                  <span className="text-xs ml-auto" style={{ color: "var(--color-text-muted)" }}>
-                    {new Date(r.timestamp).toLocaleTimeString()}
-                  </span>
-                )}
-              </div>
-              {r.content && (
-                <pre className="whitespace-pre-wrap text-xs mt-1" style={{ color: "var(--color-text)" }}>
-                  {typeof r.content === "string" ? r.content : JSON.stringify(r.content, null, 2)}
-                </pre>
-              )}
-            </div>
-          ))}
-        </div>
-      )}
-    </div>
-  );
-}
@@ -7,10 +7,10 @@ The agent builds a full prompt (system + task + step history via `@uncaged/workf
 ## Install

 ```bash
-bun add @uncaged/workflow-agent-cursor @uncaged/workflow @uncaged/workflow-util-agent zod
+bun add @uncaged/workflow-agent-cursor @uncaged/workflow-runtime @uncaged/workflow-util-agent zod
 ```

-In this monorepo: `"@uncaged/workflow-agent-cursor": "workspace:*"` plus `workspace:*` for `@uncaged/workflow` and `@uncaged/workflow-util-agent`.
+In this monorepo: `"@uncaged/workflow-agent-cursor": "workspace:*"` plus `workspace:*` for `@uncaged/workflow-runtime` and `@uncaged/workflow-util-agent`, and `zod` ^4.

 ## Usage

@@ -28,9 +28,8 @@ const agent = createCursorAgent({

 | Export | Description |
 |--------|-------------|
-| `createCursorAgent(config)` | Returns `AgentFn` that runs `cursor-agent` with `buildAgentPrompt(ctx)` |
+| `createCursorAgent(config)` | Returns `AgentFn` that runs `cursor-agent` with `buildAgentPrompt(ctx)` from `@uncaged/workflow-util-agent` |
 | `CursorAgentConfig` | `model`, `timeout`, `extract` (must supply workspace path) |
 | `validateCursorAgentConfig` | Config validation result |
-| `buildAgentPrompt` | Re-exported from `@uncaged/workflow-util-agent` |

 Requires `cursor-agent` on `PATH` at runtime.
@@ -5,7 +5,6 @@ import * as z from "zod/v4";
 import type { CursorAgentConfig } from "./types.js";
 import { validateCursorAgentConfig } from "./validate-config.js";

-export { buildAgentPrompt } from "@uncaged/workflow-util-agent";
 export type { CursorAgentConfig } from "./types.js";
 export { validateCursorAgentConfig } from "./validate-config.js";

@@ -6,5 +6,5 @@
    "composite": true
  },
  "include": ["src/**/*.ts"],
-  "references": [{ "path": "../workflow" }, { "path": "../workflow-util-agent" }]
+  "references": [{ "path": "../workflow-runtime" }, { "path": "../workflow-util-agent" }]
 }
@@ -7,10 +7,10 @@ The agent composes the same thread-aware prompt as other CLI-backed agents via `
 ## Install

 ```bash
-bun add @uncaged/workflow-agent-hermes @uncaged/workflow @uncaged/workflow-util-agent
+bun add @uncaged/workflow-agent-hermes @uncaged/workflow-runtime @uncaged/workflow-util-agent
 ```

-In this monorepo: use `workspace:*` for all three `@uncaged/*` packages.
+In this monorepo: use `workspace:*` for `@uncaged/workflow-agent-hermes`, `@uncaged/workflow-runtime`, and `@uncaged/workflow-util-agent`.

 ## Usage

@@ -30,6 +30,5 @@ const agent = createHermesAgent({
 | `createHermesAgent(config)` | Returns `AgentFn` wrapping `hermes chat -q ...` |
 | `HermesAgentConfig` | `model`, `timeout` |
 | `validateHermesAgentConfig` | Config validation result |
-| `buildAgentPrompt` | Re-exported from `@uncaged/workflow-util-agent` |

 Requires `hermes` on `PATH` at runtime.
@@ -6,7 +6,6 @@ import { validateHermesAgentConfig } from "./validate-config.js";

 const HERMES_DEFAULT_MAX_TURNS = 90;

-export { buildAgentPrompt } from "@uncaged/workflow-util-agent";
 export type { HermesAgentConfig } from "./types.js";
 export { validateHermesAgentConfig } from "./validate-config.js";

@@ -6,5 +6,5 @@
    "composite": true
  },
  "include": ["src/**/*.ts"],
-  "references": [{ "path": "../workflow" }, { "path": "../workflow-util-agent" }]
+  "references": [{ "path": "../workflow-runtime" }, { "path": "../workflow-util-agent" }]
 }
@@ -1,16 +1,16 @@
 # @uncaged/workflow-agent-llm

-`AgentFn` adapter that calls an OpenAI-compatible `POST /chat/completions` endpoint using `@uncaged/workflow`’s `LlmProvider` (base URL, API key, model).
+`AgentFn` adapter that calls an OpenAI-compatible `POST /chat/completions` endpoint using `LlmProvider` from `@uncaged/workflow-runtime`.

 Single-turn: system text is the current role’s `systemPrompt`, user text is the thread’s initial prompt (`ctx.start.content`). Errors from HTTP, JSON, or empty choices are thrown as `Error` with a JSON payload string.

 ## Install

 ```bash
-bun add @uncaged/workflow-agent-llm @uncaged/workflow
+bun add @uncaged/workflow-agent-llm @uncaged/workflow-runtime zod
 ```

-In this monorepo: `"@uncaged/workflow-agent-llm": "workspace:*"`, `"@uncaged/workflow": "workspace:*"`.
+In this monorepo: `"@uncaged/workflow-agent-llm": "workspace:*"`, `"@uncaged/workflow-runtime": "workspace:*"` (and satisfy `zod` ^4 as required by `@uncaged/workflow-runtime`).

 ## Usage

@@ -1,16 +1,9 @@
 import { describe, expect, test } from "bun:test";
-import { mkdtempSync } from "node:fs";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import { createCasStore } from "@uncaged/workflow";
-import { START, type ThreadContext } from "@uncaged/workflow-runtime";
+import { type AgentContext, START } from "@uncaged/workflow-runtime";

 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 {
+function makeCtx(userContent: string): AgentContext {
  return {
    start: {
      role: START,
@@ -22,7 +15,6 @@ function makeCtx(userContent: string): ThreadContext {
    steps: [],
    threadId: "01TEST000000000000000000TR",
    currentRole: { name: "planner", systemPrompt: "system instructions" },
-    cas: testCas,
  };
 }

@@ -8,7 +8,6 @@
    "test": "bun test"
  },
  "dependencies": {
-    "@uncaged/workflow": "workspace:*",
    "@uncaged/workflow-runtime": "workspace:*"
  }
 }
@@ -6,5 +6,5 @@
    "composite": true
  },
  "include": ["src/**/*.ts"],
-  "references": [{ "path": "../workflow" }]
+  "references": [{ "path": "../workflow-runtime" }]
 }
@@ -0,0 +1,31 @@
+# @uncaged/workflow-cas
+
+Content-addressable storage implementation, bundle hashing, and Merkle helpers.
+
+## What This Package Does
+
+It implements `CasStore` from `@uncaged/workflow-protocol`, hashes workflow bundle bytes and strings with XXH64, and builds serializable Merkle nodes for thread/step/content payloads used when persisting execution artifacts.
+
+## Key Exports
+
+From `src/index.ts`:
+
+- **CAS:** `createCasStore`
+- **Hash:** `hashString`, `hashWorkflowBundleBytes`
+- **Merkle:** `createContentMerkleNode`, `getContentMerklePayload`, `parseMerkleNode`, `putContentMerkleNode`, `putStepMerkleNode`, `putThreadMerkleNode`, `serializeMerkleNode`
+- **Types:** `CasStore`, `MerkleNode`, `MerkleNodeType`, `StepMerklePayload`, `ThreadMerklePayload`
+
+## Dependencies
+
+- **Workspace:** `@uncaged/workflow-protocol` (`CasStore` contract), `@uncaged/workflow-util`
+- **npm:** `xxhashjs`, `yaml`
+
+## Usage
+
+```typescript
+import { createCasStore, hashWorkflowBundleBytes } from "@uncaged/workflow-cas";
+import { getGlobalCasDir } from "@uncaged/workflow-util";
+
+const store = createCasStore(getGlobalCasDir());
+const hash = await hashWorkflowBundleBytes(esmJsBytes);
+```
@@ -0,0 +1,65 @@
+import { describe, expect, test } from "bun:test";
+import type { StateNode } from "@uncaged/workflow-protocol";
+
+import { collectRefs } from "../src/collect-refs.js";
+
+function payload(
+  partial: Partial<StateNode["payload"]> & Pick<StateNode["payload"], "role">,
+): StateNode["payload"] {
+  return {
+    role: partial.role,
+    meta: partial.meta ?? {},
+    start: partial.start ?? "STARTHASH000000000000001",
+    content: partial.content ?? "CONTENTHASH00000000000001",
+    ancestors: partial.ancestors ?? [],
+    compact: partial.compact ?? null,
+    timestamp: partial.timestamp ?? 0,
+  };
+}
+
+describe("collectRefs", () => {
+  test("collects start, content, ancestors, and compact hashes in order", () => {
+    const refs = collectRefs(
+      payload({
+        role: "coder",
+        start: "01START00000000000000001",
+        content: "01CONTENT0000000000000001",
+        ancestors: ["01PARENT0000000000000001", "01GRAND000000000000000001"],
+        compact: "01COMPACT0000000000000001",
+      }),
+    );
+    expect(refs).toEqual([
+      "01START00000000000000001",
+      "01CONTENT0000000000000001",
+      "01PARENT0000000000000001",
+      "01GRAND000000000000000001",
+      "01COMPACT0000000000000001",
+    ]);
+  });
+
+  test("does not collect compact when compact is null", () => {
+    const refs = collectRefs(
+      payload({
+        role: "coder",
+        start: "S1",
+        content: "C1",
+        ancestors: ["A1"],
+        compact: null,
+      }),
+    );
+    expect(refs).toEqual(["S1", "C1", "A1"]);
+  });
+
+  test("returns only start and content when ancestors is empty", () => {
+    const refs = collectRefs(
+      payload({
+        role: "coder",
+        start: "S2",
+        content: "C2",
+        ancestors: [],
+        compact: null,
+      }),
+    );
+    expect(refs).toEqual(["S2", "C2"]);
+  });
+});
@@ -0,0 +1,69 @@
+import { describe, expect, test } from "bun:test";
+import type { CasStore } from "@uncaged/workflow-protocol";
+import { stringify } from "yaml";
+
+import { findReachableHashes } from "../src/reachable.js";
+
+function yamlBlob(refs: readonly string[]): string {
+  return stringify({ type: "node", payload: {}, refs: [...refs] }, { indent: 2 });
+}
+
+function memoryCas(entries: Record<string, string>): CasStore {
+  const map = { ...entries };
+  return {
+    async put(): Promise<string> {
+      throw new Error("memoryCas.put not used in tests");
+    },
+    async get(hash: string): Promise<string | null> {
+      return map[hash] ?? null;
+    },
+    async delete(): Promise<void> {},
+    async list(): Promise<string[]> {
+      return Object.keys(map);
+    },
+  };
+}
+
+describe("findReachableHashes", () => {
+  test("walks refs recursively from a single root", async () => {
+    const cas = memoryCas({
+      R1: yamlBlob(["R2"]),
+      R2: yamlBlob(["R3"]),
+      R3: yamlBlob([]),
+    });
+    const reachable = await findReachableHashes(["R1"], cas);
+    expect([...reachable].sort()).toEqual(["R1", "R2", "R3"]);
+  });
+
+  test("union of reachability from multiple roots", async () => {
+    const cas = memoryCas({
+      A: yamlBlob(["X"]),
+      B: yamlBlob(["Y"]),
+      X: yamlBlob([]),
+      Y: yamlBlob(["Z"]),
+      Z: yamlBlob([]),
+    });
+    const reachable = await findReachableHashes(["A", "B"], cas);
+    expect([...reachable].sort()).toEqual(["A", "B", "X", "Y", "Z"]);
+  });
+
+  test("handles cycles via visited set", async () => {
+    const cas = memoryCas({
+      C1: yamlBlob(["C2"]),
+      C2: yamlBlob(["C1"]),
+    });
+    const reachable = await findReachableHashes(["C1"], cas);
+    expect(reachable.size).toBe(2);
+    expect(reachable.has("C1")).toBe(true);
+    expect(reachable.has("C2")).toBe(true);
+  });
+
+  test("does not throw when a ref points to a missing blob", async () => {
+    const cas = memoryCas({
+      H1: yamlBlob(["MISSINGHASH0000000000001"]),
+    });
+    const reachable = await findReachableHashes(["H1"], cas);
+    expect(reachable.has("H1")).toBe(true);
+    expect(reachable.has("MISSINGHASH0000000000001")).toBe(false);
+  });
+});
@@ -0,0 +1,23 @@
+{
+  "name": "@uncaged/workflow-cas",
+  "version": "0.1.0",
+  "type": "module",
+  "scripts": {
+    "test": "bun test"
+  },
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./src/index.ts"
+    }
+  },
+  "dependencies": {
+    "@uncaged/workflow-protocol": "workspace:*",
+    "@uncaged/workflow-util": "workspace:*",
+    "xxhashjs": "^0.2.2",
+    "yaml": "^2.7.1"
+  },
+  "devDependencies": {
+    "@types/bun": "latest"
+  }
+}
@@ -3,10 +3,14 @@ import { join } from "node:path";

 import { hashString } from "./hash.js";
 import { createContentMerkleNode, parseMerkleNode, serializeMerkleNode } from "./merkle.js";
+import { isCasNodeYaml } from "./nodes.js";
 import type { CasStore } from "./types.js";

 /** Raw strings become content merkle YAML; already-valid merkle documents pass through. */
 function normalizeCasPutContent(content: string): string {
+  if (isCasNodeYaml(content)) {
+    return content;
+  }
  try {
    parseMerkleNode(content);
    return content;
@@ -0,0 +1,13 @@
+import type { StateNode } from "@uncaged/workflow-protocol";
+
+/** Collects CAS hashes from {@link StateNode} payload fields for GC `refs[]` derivation. */
+export function collectRefs(payload: StateNode["payload"]): string[] {
+  const out: string[] = [payload.start, payload.content];
+  for (const h of payload.ancestors) {
+    out.push(h);
+  }
+  if (payload.compact !== null) {
+    out.push(payload.compact);
+  }
+  return out;
+}
@@ -2,7 +2,7 @@ import { Buffer } from "node:buffer";

 import XXH from "xxhashjs";

-import { encodeUint64AsCrockford } from "../util/index.js";
+import { encodeUint64AsCrockford } from "@uncaged/workflow-util";

 function digestToUint64(digest: { toString(radix?: number): string }): bigint {
  const hex = digest.toString(16).padStart(16, "0");
@@ -1,4 +1,5 @@
 export { createCasStore } from "./cas.js";
+export { collectRefs } from "./collect-refs.js";
 export { hashString, hashWorkflowBundleBytes } from "./hash.js";
 export {
  createContentMerkleNode,
@@ -9,6 +10,14 @@ export {
  putThreadMerkleNode,
  serializeMerkleNode,
 } from "./merkle.js";
+export {
+  isCasNodeYaml,
+  putContentNodeWithRefs,
+  putStartNode,
+  putStateNode,
+  serializeCasNode,
+} from "./nodes.js";
+export { findReachableHashes } from "./reachable.js";
 export type {
  CasStore,
  MerkleNode,
@@ -82,7 +82,12 @@ export async function putContentMerkleNode(store: CasStore, content: string): Pr
  return store.put(content);
 }

-/** Loads a CAS blob and returns the payload string for a `content` Merkle node. */
+/**
+ * Loads a CAS blob and returns the payload string for a `content` node.
+ *
+ * Accepts both the legacy `{type:content, payload, children}` Merkle layout
+ * and the RFC v3 `{type:content, payload, refs}` content node layout.
+ */
 export async function getContentMerklePayload(
  store: CasStore,
  hash: string,
@@ -91,9 +96,13 @@ export async function getContentMerklePayload(
  if (yamlText === null) {
    return null;
  }
-  const node = parseMerkleNode(yamlText);
-  if (node.type !== "content" || typeof node.payload !== "string") {
+  const raw = parse(yamlText) as unknown;
+  if (raw === null || typeof raw !== "object") {
    return null;
  }
-  return node.payload;
+  const rec = raw as Record<string, unknown>;
+  if (rec.type !== "content" || typeof rec.payload !== "string") {
+    return null;
+  }
+  return rec.payload;
 }
@@ -0,0 +1,79 @@
+import type { ContentMerkleNode, StartNode, StateNode } from "@uncaged/workflow-protocol";
+import { parse, stringify } from "yaml";
+
+import { collectRefs } from "./collect-refs.js";
+import type { CasStore } from "./types.js";
+
+/** YAML-serialize a CAS node carrying `{type, payload, refs}` (RFC v3 thread storage format). */
+export function serializeCasNode(node: StartNode | StateNode | ContentMerkleNode): string {
+  return stringify({ type: node.type, payload: node.payload, refs: node.refs }, { indent: 2 });
+}
+
+/**
+ * Recognizes a YAML CAS blob with the `{type, payload, refs[]}` shape used by
+ * `start` / `state` / `content` thread nodes. Used by {@link createCasStore}
+ * to skip the legacy auto-wrap step when the caller already supplied a
+ * pre-serialized RFC v3 node.
+ */
+export function isCasNodeYaml(content: string): boolean {
+  let raw: unknown;
+  try {
+    raw = parse(content) as unknown;
+  } catch {
+    return false;
+  }
+  if (raw === null || typeof raw !== "object" || Array.isArray(raw)) {
+    return false;
+  }
+  const rec = raw as Record<string, unknown>;
+  if (typeof rec.type !== "string") {
+    return false;
+  }
+  if (!Array.isArray(rec.refs)) {
+    return false;
+  }
+  for (const r of rec.refs) {
+    if (typeof r !== "string") {
+      return false;
+    }
+  }
+  return true;
+}
+
+export async function putStartNode(
+  store: CasStore,
+  payload: StartNode["payload"],
+  promptHash: string,
+): Promise<string> {
+  const node: StartNode = {
+    type: "start",
+    payload,
+    refs: [promptHash],
+  };
+  return store.put(serializeCasNode(node));
+}
+
+export async function putStateNode(
+  store: CasStore,
+  payload: StateNode["payload"],
+): Promise<string> {
+  const node: StateNode = {
+    type: "state",
+    payload,
+    refs: collectRefs(payload),
+  };
+  return store.put(serializeCasNode(node));
+}
+
+export async function putContentNodeWithRefs(
+  store: CasStore,
+  payload: string,
+  refs: readonly string[],
+): Promise<string> {
+  const node: ContentMerkleNode = {
+    type: "content",
+    payload,
+    refs: [...refs],
+  };
+  return store.put(serializeCasNode(node));
+}
@@ -0,0 +1,55 @@
+import { parse } from "yaml";
+
+import type { CasStore } from "./types.js";
+
+function refsFromBlob(content: string): string[] {
+  try {
+    const raw = parse(content) as unknown;
+    if (raw === null || typeof raw !== "object") {
+      return [];
+    }
+    const rec = raw as Record<string, unknown>;
+    const refs = rec.refs;
+    if (!Array.isArray(refs)) {
+      return [];
+    }
+    const out: string[] = [];
+    for (const r of refs) {
+      if (typeof r === "string") {
+        out.push(r);
+      }
+    }
+    return out;
+  } catch {
+    return [];
+  }
+}
+
+/** Recursively collects all CAS hashes reachable from `roots` via each blob's `refs[]`. */
+export async function findReachableHashes(
+  roots: readonly string[],
+  cas: CasStore,
+): Promise<ReadonlySet<string>> {
+  const visited = new Set<string>();
+  const stack = [...roots];
+  while (stack.length > 0) {
+    const hash = stack.pop();
+    if (hash === undefined) {
+      break;
+    }
+    if (visited.has(hash)) {
+      continue;
+    }
+    const blob = await cas.get(hash);
+    if (blob === null) {
+      continue;
+    }
+    visited.add(hash);
+    for (const ref of refsFromBlob(blob)) {
+      if (!visited.has(ref)) {
+        stack.push(ref);
+      }
+    }
+  }
+  return visited;
+}
@@ -1,4 +1,4 @@
-export type { CasStore } from "@uncaged/workflow-runtime";
+export type { CasStore } from "@uncaged/workflow-protocol";

 export type MerkleNodeType = "content" | "step" | "thread";

@@ -0,0 +1,9 @@
+{
+  "extends": "../../tsconfig.json",
+  "compilerOptions": {
+    "rootDir": "src",
+    "outDir": "dist"
+  },
+  "include": ["src"],
+  "references": [{ "path": "../workflow-protocol" }, { "path": "../workflow-util" }]
+}
@@ -1,5 +1,18 @@
 const BASE = "/api";

+async function postJson<T>(path: string, body: unknown): Promise<T> {
+  const res = await fetch(`${BASE}${path}`, {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify(body),
+  });
+  if (!res.ok) {
+    const err = (await res.json().catch(() => ({ error: res.statusText }))) as { error: string };
+    throw new Error(err.error || `API ${res.status}`);
+  }
+  return res.json() as Promise<T>;
+}
+
 async function fetchJson<T>(path: string): Promise<T> {
  const res = await fetch(`${BASE}${path}`);
  if (!res.ok) {
@@ -46,6 +59,26 @@ export function getThread(id: string): Promise<{ records: ThreadRecord[] }> {
  return fetchJson(`/threads/${id}`);
 }

+export function runThread(
+  workflow: string,
+  prompt: string,
+  maxRounds: number = 10,
+): Promise<{ threadId: string }> {
+  return postJson("/threads", { workflow, prompt, maxRounds });
+}
+
+export function killThread(threadId: string): Promise<{ ok: boolean }> {
+  return postJson(`/threads/${threadId}/kill`, {});
+}
+
+export function pauseThread(threadId: string): Promise<{ ok: boolean }> {
+  return postJson(`/threads/${threadId}/pause`, {});
+}
+
+export function resumeThread(threadId: string): Promise<{ ok: boolean }> {
+  return postJson(`/threads/${threadId}/resume`, {});
+}
+
 export function getHealth(): Promise<{ ok: boolean }> {
  return fetchJson("/healthz");
 }
@@ -0,0 +1,38 @@
+import { useState } from "react";
+import { RunDialog } from "./components/run-dialog.tsx";
+import { Sidebar } from "./components/sidebar.tsx";
+import { StatusBar } from "./components/status-bar.tsx";
+import { ThreadDetail } from "./components/thread-detail.tsx";
+import { ThreadList } from "./components/thread-list.tsx";
+import { WorkflowList } from "./components/workflow-list.tsx";
+import { useHashRoute } from "./use-hash-route.ts";
+
+export function App() {
+  const { view, threadId, setView, setThreadId } = useHashRoute();
+  const [showRun, setShowRun] = useState(false);
+
+  return (
+    <div className="flex h-screen">
+      <Sidebar view={view} onViewChange={setView} />
+      <main className="flex-1 overflow-hidden flex flex-col">
+        <StatusBar onRun={() => setShowRun(true)} />
+        <div className="flex-1 overflow-auto p-6">
+          {view === "threads" && threadId === null && <ThreadList onSelect={setThreadId} />}
+          {view === "threads" && threadId !== null && (
+            <ThreadDetail threadId={threadId} onBack={() => setThreadId(null)} />
+          )}
+          {view === "workflows" && <WorkflowList />}
+        </div>
+      </main>
+      {showRun && (
+        <RunDialog
+          onClose={() => setShowRun(false)}
+          onCreated={(id) => {
+            setShowRun(false);
+            setThreadId(id);
+          }}
+        />
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,147 @@
+import { useState } from "react";
+import { listWorkflows, runThread } from "../api.ts";
+import { useFetch } from "../hooks.ts";
+
+type Props = {
+  onClose: () => void;
+  onCreated: (threadId: string) => void;
+};
+
+export function RunDialog({ onClose, onCreated }: Props) {
+  const workflows = useFetch(() => listWorkflows(), []);
+  const [workflow, setWorkflow] = useState("");
+  const [prompt, setPrompt] = useState("");
+  const [maxRounds, setMaxRounds] = useState(10);
+  const [submitting, setSubmitting] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+
+  async function handleSubmit(e: React.FormEvent) {
+    e.preventDefault();
+    if (!workflow || !prompt) return;
+    setSubmitting(true);
+    setError(null);
+    try {
+      const result = await runThread(workflow, prompt, maxRounds);
+      onCreated(result.threadId);
+    } catch (err) {
+      setError(err instanceof Error ? err.message : String(err));
+      setSubmitting(false);
+    }
+  }
+
+  return (
+    <div
+      className="fixed inset-0 flex items-center justify-center z-50"
+      style={{ background: "rgba(0,0,0,0.6)" }}
+    >
+      <div
+        className="w-full max-w-lg p-6 rounded-lg border"
+        style={{ background: "var(--color-surface)", borderColor: "var(--color-border)" }}
+      >
+        <h3 className="text-lg font-semibold mb-4">Run Thread</h3>
+        <form onSubmit={handleSubmit} className="space-y-4">
+          <div>
+            <label
+              htmlFor="run-workflow"
+              className="text-sm block mb-1"
+              style={{ color: "var(--color-text-muted)" }}
+            >
+              Workflow
+            </label>
+            <select
+              id="run-workflow"
+              value={workflow}
+              onChange={(e) => setWorkflow(e.target.value)}
+              className="w-full px-3 py-2 rounded border text-sm"
+              style={{
+                background: "var(--color-bg)",
+                borderColor: "var(--color-border)",
+                color: "var(--color-text)",
+              }}
+            >
+              <option value="">Select a workflow...</option>
+              {workflows.status === "ok" &&
+                workflows.data.workflows.map((w) => (
+                  <option key={w.name} value={w.name}>
+                    {w.name}
+                  </option>
+                ))}
+            </select>
+          </div>
+          <div>
+            <label
+              htmlFor="run-prompt"
+              className="text-sm block mb-1"
+              style={{ color: "var(--color-text-muted)" }}
+            >
+              Prompt
+            </label>
+            <textarea
+              id="run-prompt"
+              value={prompt}
+              onChange={(e) => setPrompt(e.target.value)}
+              rows={4}
+              className="w-full px-3 py-2 rounded border text-sm"
+              style={{
+                background: "var(--color-bg)",
+                borderColor: "var(--color-border)",
+                color: "var(--color-text)",
+              }}
+              placeholder="Enter the task prompt..."
+            />
+          </div>
+          <div>
+            <label
+              htmlFor="run-max-rounds"
+              className="text-sm block mb-1"
+              style={{ color: "var(--color-text-muted)" }}
+            >
+              Max Rounds
+            </label>
+            <input
+              id="run-max-rounds"
+              type="number"
+              value={maxRounds}
+              onChange={(e) => setMaxRounds(Number(e.target.value))}
+              min={1}
+              max={100}
+              className="w-24 px-3 py-2 rounded border text-sm"
+              style={{
+                background: "var(--color-bg)",
+                borderColor: "var(--color-border)",
+                color: "var(--color-text)",
+              }}
+            />
+          </div>
+          {error && (
+            <p className="text-sm" style={{ color: "var(--color-error)" }}>
+              {error}
+            </p>
+          )}
+          <div className="flex gap-2 justify-end">
+            <button
+              type="button"
+              onClick={onClose}
+              className="px-4 py-2 text-sm rounded border"
+              style={{ borderColor: "var(--color-border)", color: "var(--color-text-muted)" }}
+            >
+              Cancel
+            </button>
+            <button
+              type="submit"
+              disabled={submitting || !workflow || !prompt}
+              className="px-4 py-2 text-sm rounded"
+              style={{
+                background: submitting ? "var(--color-accent-dim)" : "var(--color-accent)",
+                color: "#fff",
+                opacity: !workflow || !prompt ? 0.5 : 1,
+              }}
+            >
+              {submitting ? "Starting..." : "Run"}
+            </button>
+          </div>
+        </form>
+      </div>
+    </div>
+  );
+}
@@ -10,16 +10,22 @@ export function Sidebar({ view, onViewChange }: Props) {
  ];

  return (
-    <aside className="w-56 border-r flex flex-col" style={{ borderColor: "var(--color-border)", background: "var(--color-surface)" }}>
+    <aside
+      className="w-56 border-r flex flex-col"
+      style={{ borderColor: "var(--color-border)", background: "var(--color-surface)" }}
+    >
      <div className="p-4 border-b" style={{ borderColor: "var(--color-border)" }}>
        <h1 className="text-lg font-semibold" style={{ color: "var(--color-accent)" }}>
          ⚙ Workflow
        </h1>
-        <p className="text-xs mt-1" style={{ color: "var(--color-text-muted)" }}>Dashboard</p>
+        <p className="text-xs mt-1" style={{ color: "var(--color-text-muted)" }}>
+          Dashboard
+        </p>
      </div>
      <nav className="flex-1 p-2 space-y-1">
        {items.map((item) => (
          <button
+            type="button"
            key={item.key}
            onClick={() => onViewChange(item.key)}
            className="w-full text-left px-3 py-2 rounded text-sm transition-colors"
@@ -0,0 +1,65 @@
+import { useCallback, useEffect, useRef, useState } from "react";
+import { getHealth } from "../api.ts";
+
+type HealthStatus = "connected" | "disconnected" | "reconnecting";
+
+type Props = {
+  onRun: () => void;
+};
+
+function statusLabel(status: HealthStatus): { text: string; color: string } {
+  if (status === "connected") {
+    return { text: "● Connected", color: "var(--color-success)" };
+  }
+  if (status === "reconnecting") {
+    return { text: "● Reconnecting...", color: "var(--color-warning, #f59e0b)" };
+  }
+  return { text: "● Offline", color: "var(--color-error)" };
+}
+
+export function StatusBar({ onRun }: Props) {
+  const [status, setStatus] = useState<HealthStatus>("disconnected");
+  const wasConnectedRef = useRef(false);
+
+  const checkHealth = useCallback(async () => {
+    try {
+      await getHealth();
+      wasConnectedRef.current = true;
+      setStatus("connected");
+    } catch {
+      if (wasConnectedRef.current) {
+        setStatus("reconnecting");
+      } else {
+        setStatus("disconnected");
+      }
+    }
+  }, []);
+
+  useEffect(() => {
+    checkHealth();
+    const interval = setInterval(checkHealth, 10_000);
+    return () => clearInterval(interval);
+  }, [checkHealth]);
+
+  const label = statusLabel(status);
+
+  return (
+    <div
+      className="flex items-center justify-between px-6 py-2 text-xs border-b"
+      style={{ borderColor: "var(--color-border)", background: "var(--color-surface)" }}
+    >
+      <div className="flex items-center gap-4">
+        <span style={{ color: "var(--color-text-muted)" }}>Local API: 127.0.0.1:7860</span>
+        <button
+          type="button"
+          onClick={onRun}
+          className="px-3 py-1 rounded text-xs font-medium"
+          style={{ background: "var(--color-accent)", color: "#fff" }}
+        >
+          ▶ Run Thread
+        </button>
+      </div>
+      <span style={{ color: label.color }}>{label.text}</span>
+    </div>
+  );
+}
@@ -0,0 +1,143 @@
+import { useEffect, useRef, useState } from "react";
+import { getThread, killThread, pauseThread, resumeThread } from "../api.ts";
+import { useFetch } from "../hooks.ts";
+import { useSSE } from "../use-sse.ts";
+
+type Props = {
+  threadId: string;
+  onBack: () => void;
+};
+
+export function ThreadDetail({ threadId, onBack }: Props) {
+  const sse = useSSE(threadId);
+  const { status, data, error } = useFetch(() => getThread(threadId), [threadId]);
+  const [actionStatus, setActionStatus] = useState<string | null>(null);
+  const recordsEndRef = useRef<HTMLDivElement>(null);
+
+  const liveActive = sse.connected && !sse.completed;
+  const records = liveActive
+    ? sse.records
+    : status === "ok"
+      ? data.records
+      : ([] as typeof sse.records);
+
+  // biome-ignore lint/correctness/useExhaustiveDependencies: scroll when the rendered record list grows
+  useEffect(() => {
+    recordsEndRef.current?.scrollIntoView({ behavior: "smooth" });
+  }, [records.length]);
+
+  async function handleAction(action: "kill" | "pause" | "resume") {
+    setActionStatus(`${action}ing...`);
+    try {
+      const fn = action === "kill" ? killThread : action === "pause" ? pauseThread : resumeThread;
+      await fn(threadId);
+      setActionStatus(`${action} sent ✓`);
+    } catch (e) {
+      setActionStatus(`${action} failed: ${e instanceof Error ? e.message : String(e)}`);
+    }
+  }
+
+  return (
+    <div>
+      <div className="flex items-center justify-between mb-4">
+        <button
+          type="button"
+          onClick={onBack}
+          className="text-sm hover:underline"
+          style={{ color: "var(--color-accent)" }}
+        >
+          ← Back to threads
+        </button>
+        <div className="flex gap-2">
+          <button
+            type="button"
+            onClick={() => handleAction("pause")}
+            className="px-3 py-1 text-xs rounded border"
+            style={{ borderColor: "var(--color-warning)", color: "var(--color-warning)" }}
+          >
+            ⏸ Pause
+          </button>
+          <button
+            type="button"
+            onClick={() => handleAction("resume")}
+            className="px-3 py-1 text-xs rounded border"
+            style={{ borderColor: "var(--color-success)", color: "var(--color-success)" }}
+          >
+            ▶ Resume
+          </button>
+          <button
+            type="button"
+            onClick={() => handleAction("kill")}
+            className="px-3 py-1 text-xs rounded border"
+            style={{ borderColor: "var(--color-error)", color: "var(--color-error)" }}
+          >
+            ✕ Kill
+          </button>
+        </div>
+      </div>
+
+      <h2 className="text-xl font-semibold mb-2 font-mono flex items-center gap-2 flex-wrap">
+        <span>{threadId}</span>
+        {sse.connected && (
+          <span
+            className="text-xs font-medium px-2 py-0.5 rounded"
+            style={{ background: "var(--color-success)", color: "var(--color-bg)" }}
+          >
+            Live
+          </span>
+        )}
+      </h2>
+      {actionStatus && (
+        <p className="text-xs mb-4" style={{ color: "var(--color-text-muted)" }}>
+          {actionStatus}
+        </p>
+      )}
+
+      {status === "loading" && !liveActive && records.length === 0 && (
+        <p style={{ color: "var(--color-text-muted)" }}>Loading...</p>
+      )}
+      {status === "error" && !liveActive && (
+        <p style={{ color: "var(--color-error)" }}>Error: {error}</p>
+      )}
+      {(status === "ok" || liveActive || records.length > 0) && (
+        <div className="space-y-3">
+          {records.map((r) => (
+            <div
+              key={`${threadId}-${r.type}-${String(r.timestamp)}-${r.role ?? ""}-${r.content ?? ""}`}
+              className="p-3 rounded border text-sm"
+              style={{ background: "var(--color-surface)", borderColor: "var(--color-border)" }}
+            >
+              <div className="flex items-center gap-2 mb-1">
+                <span
+                  className="text-xs px-1.5 py-0.5 rounded font-mono"
+                  style={{ background: "var(--color-border)", color: "var(--color-accent)" }}
+                >
+                  {r.type}
+                </span>
+                {r.role && (
+                  <span className="text-xs" style={{ color: "var(--color-text-muted)" }}>
+                    {r.role}
+                  </span>
+                )}
+                {r.timestamp !== null && (
+                  <span className="text-xs ml-auto" style={{ color: "var(--color-text-muted)" }}>
+                    {new Date(r.timestamp).toLocaleTimeString()}
+                  </span>
+                )}
+              </div>
+              {r.content && (
+                <pre
+                  className="whitespace-pre-wrap text-xs mt-1"
+                  style={{ color: "var(--color-text)" }}
+                >
+                  {typeof r.content === "string" ? r.content : JSON.stringify(r.content, null, 2)}
+                </pre>
+              )}
+            </div>
+          ))}
+          <div ref={recordsEndRef} aria-hidden />
+        </div>
+      )}
+    </div>
+  );
+}
@@ -8,7 +8,8 @@ type Props = {
 export function ThreadList({ onSelect }: Props) {
  const { status, data, error } = useFetch(() => listThreads(), []);

-  if (status === "loading") return <p style={{ color: "var(--color-text-muted)" }}>Loading threads...</p>;
+  if (status === "loading")
+    return <p style={{ color: "var(--color-text-muted)" }}>Loading threads...</p>;
  if (status === "error") return <p style={{ color: "var(--color-error)" }}>Error: {error}</p>;

  const threads = data.threads;
@@ -22,6 +23,7 @@ export function ThreadList({ onSelect }: Props) {
        <div className="space-y-2">
          {threads.map((t) => (
            <button
+              type="button"
              key={t.threadId}
              onClick={() => onSelect(t.threadId)}
              className="w-full text-left p-4 rounded-lg border transition-colors hover:border-[var(--color-accent-dim)]"
@@ -4,7 +4,8 @@ import { useFetch } from "../hooks.ts";
 export function WorkflowList() {
  const { status, data, error } = useFetch(() => listWorkflows(), []);

-  if (status === "loading") return <p style={{ color: "var(--color-text-muted)" }}>Loading workflows...</p>;
+  if (status === "loading")
+    return <p style={{ color: "var(--color-text-muted)" }}>Loading workflows...</p>;
  if (status === "error") return <p style={{ color: "var(--color-error)" }}>Error: {error}</p>;

  const workflows = data.workflows;
@@ -28,7 +29,10 @@ export function WorkflowList() {
                  {w.versions} version{w.versions !== 1 ? "s" : ""}
                </span>
              </div>
-              <code className="text-xs mt-1 block font-mono" style={{ color: "var(--color-accent)" }}>
+              <code
+                className="text-xs mt-1 block font-mono"
+                style={{ color: "var(--color-accent)" }}
+              >
                {w.currentHash}
              </code>
            </div>
@@ -30,6 +30,7 @@ export function useFetch<T>(fetcher: () => Promise<T>, deps: unknown[] = []): Fe
    return () => {
      cancelled = true;
    };
+    // biome-ignore lint/correctness/useExhaustiveDependencies: this helper intentionally accepts caller-provided dependency arrays
  }, deps);

  return state;
@@ -0,0 +1,64 @@
+import { useCallback, useEffect, useState } from "react";
+
+type View = "threads" | "workflows";
+
+type HashRoute = {
+  view: View;
+  threadId: string | null;
+};
+
+function parseHash(hash: string): HashRoute {
+  const raw = hash.replace(/^#\/?/, "");
+  if (raw.startsWith("threads/")) {
+    const id = raw.slice("threads/".length);
+    if (id.length > 0) {
+      return { view: "threads", threadId: id };
+    }
+  }
+  if (raw === "workflows") {
+    return { view: "workflows", threadId: null };
+  }
+  return { view: "threads", threadId: null };
+}
+
+function buildHash(route: HashRoute): string {
+  if (route.view === "workflows") {
+    return "#workflows";
+  }
+  if (route.threadId !== null) {
+    return `#threads/${route.threadId}`;
+  }
+  return "#threads";
+}
+
+export function useHashRoute(): {
+  view: View;
+  threadId: string | null;
+  setView: (v: View) => void;
+  setThreadId: (id: string | null) => void;
+} {
+  const [route, setRoute] = useState<HashRoute>(() => parseHash(window.location.hash));
+
+  useEffect(() => {
+    function onHashChange(): void {
+      setRoute(parseHash(window.location.hash));
+    }
+    window.addEventListener("hashchange", onHashChange);
+    return () => window.removeEventListener("hashchange", onHashChange);
+  }, []);
+
+  const navigate = useCallback((next: HashRoute) => {
+    const hash = buildHash(next);
+    window.location.hash = hash;
+    setRoute(next);
+  }, []);
+
+  const setView = useCallback((v: View) => navigate({ view: v, threadId: null }), [navigate]);
+
+  const setThreadId = useCallback(
+    (id: string | null) => navigate({ view: "threads", threadId: id }),
+    [navigate],
+  );
+
+  return { view: route.view, threadId: route.threadId, setView, setThreadId };
+}
@@ -0,0 +1,161 @@
+import {
+  type Dispatch,
+  type MutableRefObject,
+  type SetStateAction,
+  useEffect,
+  useRef,
+  useState,
+} from "react";
+
+import type { ThreadRecord } from "./api.ts";
+
+export type UseSSEReturn = {
+  records: ThreadRecord[];
+  connected: boolean;
+  completed: boolean;
+};
+
+function isWorkflowResult(record: ThreadRecord): boolean {
+  return record.type === "workflow-result";
+}
+
+function parseRecord(data: string): ThreadRecord | null {
+  try {
+    return JSON.parse(data) as ThreadRecord;
+  } catch {
+    return null;
+  }
+}
+
+type RecordEventContext = {
+  cancelled: boolean;
+  completedRef: MutableRefObject<boolean>;
+  setRecords: Dispatch<SetStateAction<ThreadRecord[]>>;
+  setCompleted: (value: boolean) => void;
+  setConnected: (value: boolean) => void;
+  cleanupEs: () => void;
+};
+
+function handleRecordEvent(ev: Event, ctx: RecordEventContext): void {
+  if (ctx.cancelled) {
+    return;
+  }
+  const msg = ev as MessageEvent;
+  const raw = typeof msg.data === "string" ? msg.data : "";
+  const parsed = parseRecord(raw);
+  if (parsed === null) {
+    return;
+  }
+  ctx.setRecords((prev) => [...prev, parsed]);
+  if (!isWorkflowResult(parsed)) {
+    return;
+  }
+  ctx.completedRef.current = true;
+  ctx.setCompleted(true);
+  ctx.setConnected(false);
+  ctx.cleanupEs();
+}
+
+export function useSSE(threadId: string | null): UseSSEReturn {
+  const [records, setRecords] = useState<ThreadRecord[]>([]);
+  const [connected, setConnected] = useState(false);
+  const [completed, setCompleted] = useState(false);
+
+  const completedRef = useRef(false);
+  const reconnectAttemptsRef = useRef(0);
+
+  useEffect(() => {
+    if (threadId === null) {
+      completedRef.current = false;
+      reconnectAttemptsRef.current = 0;
+      setRecords([]);
+      setConnected(false);
+      setCompleted(false);
+      return;
+    }
+
+    const tid = threadId;
+
+    completedRef.current = false;
+    reconnectAttemptsRef.current = 0;
+    setRecords([]);
+    setConnected(false);
+    setCompleted(false);
+
+    let es: EventSource | null = null;
+    let reconnectTimer: ReturnType<typeof setTimeout> | null = null;
+    let cancelled = false;
+
+    function cleanupEs(): void {
+      if (es !== null) {
+        es.close();
+        es = null;
+      }
+    }
+
+    function scheduleReconnect(): void {
+      if (cancelled || completedRef.current) {
+        return;
+      }
+      const delayMs = Math.min(1000 * 2 ** reconnectAttemptsRef.current, 8000);
+      reconnectAttemptsRef.current += 1;
+      reconnectTimer = setTimeout(() => {
+        reconnectTimer = null;
+        if (!cancelled && !completedRef.current) {
+          connect();
+        }
+      }, delayMs);
+    }
+
+    function connect(): void {
+      if (cancelled || completedRef.current) {
+        return;
+      }
+
+      cleanupEs();
+      const url = `/api/threads/${encodeURIComponent(tid)}/live`;
+      es = new EventSource(url);
+
+      es.onopen = () => {
+        if (cancelled) {
+          return;
+        }
+        reconnectAttemptsRef.current = 0;
+        setConnected(true);
+        setRecords([]);
+      };
+
+      es.addEventListener("record", (ev: Event) =>
+        handleRecordEvent(ev, {
+          cancelled,
+          completedRef,
+          setRecords,
+          setCompleted,
+          setConnected,
+          cleanupEs,
+        }),
+      );
+
+      es.onerror = () => {
+        if (cancelled || completedRef.current) {
+          return;
+        }
+        setConnected(false);
+        cleanupEs();
+        scheduleReconnect();
+      };
+    }
+
+    connect();
+
+    return () => {
+      cancelled = true;
+      if (reconnectTimer !== null) {
+        clearTimeout(reconnectTimer);
+      }
+      cleanupEs();
+    };
+  }, [threadId]);
+
+  return { records, connected, completed };
+}
@@ -2,6 +2,7 @@ import tailwindcss from "@tailwindcss/vite";
 import react from "@vitejs/plugin-react";
 import { defineConfig } from "vite";

+// biome-ignore lint/style/noDefaultExport: Vite loads config from default export.
 export default defineConfig({
  plugins: [react(), tailwindcss()],
  server: {
@@ -0,0 +1,33 @@
+# @uncaged/workflow-execute
+
+Thread engine: execution, fork/GC, extract pipeline, supervisor/worker wiring, and workflow-as-agent.
+
+## What This Package Does
+
+It runs `WorkflowFn` generators against disk-backed threads, integrates CAS and registry-backed extract (`createExtract`), coordinates LLM tool usage via `@uncaged/workflow-reactor`, handles fork plans and garbage collection, and exposes `workflowAsAgent` for nesting workflows.
+
+## Key Exports
+
+From `src/index.ts`:
+
+- **Engine:** `createWorkflow` (engine-local re-export), `executeThread`, `getWorkerHostScriptPath`
+- **Fork / parse:** `buildForkPlan`, `parseThreadDataJsonl`, `selectForkHistoricalSteps`, `tryParseRoleStepRecord`, `tryParseWorkflowResultRecord`
+- **GC / pause:** `garbageCollectCas`, `createThreadPauseGate`
+- **Engine types:** `ExecuteThreadIo`, `ExecuteThreadOptions`, `ForkHistoricalStep`, `ForkPlan`, `GcResult`, `ParsedThreadStartRecord`, `PrefilledDiskStep`, `SupervisorDecision`, `ThreadPauseGate`
+- **Extract:** `buildExtractUserContent`, `createExtract`, `extractFunctionToolFromZodSchema`, `llmErrorToCause`, `llmExtract`, types `ExtractFn`, `ExtractThreadContext`, `LlmError`, `LlmExtractArgs`
+- **Agent composition:** `workflowAsAgent`, `WorkflowAsAgentOptions`
+
+## Dependencies
+
+- **Workspace:** `@uncaged/workflow-protocol`, `@uncaged/workflow-runtime`, `@uncaged/workflow-util`, `@uncaged/workflow-cas`, `@uncaged/workflow-reactor`, `@uncaged/workflow-register`
+- **npm:** `yaml`
+- **Peer:** `zod` ^4
+
+`@uncaged/workflow-reactor` is used for LLM-backed extract and supervisor flows (`extract-fn.ts`, `supervisor.ts`).
+
+## Usage
+
+```typescript
+import { executeThread } from "@uncaged/workflow-execute";
+// Typical callers are CLI/tests that supply ExecuteThreadIo (paths, CAS, abort, logger, …).
+```
@@ -0,0 +1,317 @@
+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 } from "@uncaged/workflow-cas";
+
+import type {
+  RoleOutput,
+  ThreadContext,
+  WorkflowCompletion,
+  WorkflowFn,
+  WorkflowRuntime,
+} from "@uncaged/workflow-runtime";
+import { parse as parseYaml } from "yaml";
+
+import { executeThread } from "../src/engine/engine.js";
+import type { ExecuteThreadIo, ExecuteThreadOptions } from "../src/engine/types.js";
+
+const TEST_REGISTRY_YAML = `config:
+  maxDepth: 3
+  supervisorInterval: 0
+  providers:
+    stub:
+      baseUrl: http://127.0.0.1:9
+      apiKey: test
+  models:
+    default: stub/m
+workflows: {}
+`;
+
+function noLogger(): (tag: string, content: string) => void {
+  return () => {};
+}
+
+function makeOptions(overrides: Partial<ExecuteThreadOptions>): ExecuteThreadOptions {
+  return {
+    maxRounds: 5,
+    depth: 0,
+    signal: new AbortController().signal,
+    awaitAfterEachYield: async () => {},
+    forkSourceThreadId: null,
+    prefilledDiskSteps: null,
+    storageRoot: "/tmp/never",
+    ...overrides,
+  };
+}
+
+async function setupStorage(): Promise<{
+  storageRoot: string;
+  casDir: string;
+  bundleHash: string;
+  bundleDir: string;
+}> {
+  const storageRoot = await mkdtemp(join(tmpdir(), "uncaged-wf-engine-"));
+  await writeFile(join(storageRoot, "workflow.yaml"), TEST_REGISTRY_YAML, "utf8");
+  const casDir = join(storageRoot, "cas");
+  await mkdir(casDir, { recursive: true });
+  const bundleHash = "TESTHASH00001";
+  const bundleDir = join(storageRoot, "bundles", bundleHash);
+  return { storageRoot, casDir, bundleHash, bundleDir };
+}
+
+function readCasNode(casDir: string, hash: string): Record<string, unknown> {
+  const text = require("node:fs").readFileSync(join(casDir, `${hash}.txt`), "utf8") as string;
+  return parseYaml(text) as Record<string, unknown>;
+}
+
+describe("executeThread (Phase 2 — CAS thread storage)", () => {
+  let storageRoot: string;
+  let casDir: string;
+  let bundleHash: string;
+  let bundleDir: string;
+
+  beforeEach(async () => {
+    const setup = await setupStorage();
+    storageRoot = setup.storageRoot;
+    casDir = setup.casDir;
+    bundleHash = setup.bundleHash;
+    bundleDir = setup.bundleDir;
+  });
+
+  afterEach(async () => {
+    await rm(storageRoot, { recursive: true, force: true });
+  });
+
+  test("writes a StartNode whose refs[0] is the prompt CAS hash", async () => {
+    const cas = createCasStore(casDir);
+
+    // biome-ignore lint/correctness/useYield: deliberately empty generator — exercises the start/end path with no role steps
+    const wf: WorkflowFn = async function* (
+      _thread: ThreadContext,
+      _runtime: WorkflowRuntime,
+    ): AsyncGenerator<RoleOutput, WorkflowCompletion> {
+      return { returnCode: 0, summary: "no-op" };
+    };
+
+    const io: ExecuteThreadIo = {
+      threadId: "T01",
+      hash: bundleHash,
+      infoJsonlPath: join(storageRoot, "logs", bundleHash, "T01.info.jsonl"),
+      cas,
+    };
+
+    const result = await executeThread(
+      wf,
+      "demo",
+      { prompt: "hello", steps: [] },
+      makeOptions({ storageRoot, maxRounds: 5 }),
+      io,
+      noLogger(),
+    );
+
+    expect(result.returnCode).toBe(0);
+
+    const historyText = await readFile(
+      (await import("node:fs/promises")).readdir ? await firstHistoryFile(bundleDir) : "",
+      "utf8",
+    );
+    const histLine = historyText.trim().split("\n")[0] ?? "";
+    const histEntry = JSON.parse(histLine) as Record<string, unknown>;
+    expect(histEntry.threadId).toBe("T01");
+
+    const startHash = histEntry.start as string;
+    const startNode = readCasNode(casDir, startHash);
+    expect(startNode.type).toBe("start");
+    expect((startNode.payload as Record<string, unknown>).name).toBe("demo");
+    expect((startNode.payload as Record<string, unknown>).hash).toBe(bundleHash);
+    expect((startNode.payload as Record<string, unknown>).maxRounds).toBe(5);
+
+    const refs = startNode.refs as string[];
+    expect(refs.length).toBe(1);
+
+    const promptBlob = await cas.get(refs[0] ?? "");
+    expect(promptBlob).not.toBeNull();
+    const promptParsed = parseYaml(promptBlob ?? "") as Record<string, unknown>;
+    expect(promptParsed.payload).toBe("hello");
+  });
+
+  test("each role yield produces a chained StateNode and updates threads.json head", async () => {
+    const cas = createCasStore(casDir);
+
+    const wf: WorkflowFn = async function* (
+      _thread: ThreadContext,
+      runtime: WorkflowRuntime,
+    ): AsyncGenerator<RoleOutput, WorkflowCompletion> {
+      const h1 = await runtime.cas.put("plan-text");
+      yield { role: "planner", contentHash: h1, meta: { plan: 1 }, refs: [h1] };
+      const h2 = await runtime.cas.put("code-text");
+      yield { role: "coder", contentHash: h2, meta: { diff: "y" }, refs: [h2] };
+      return { returnCode: 0, summary: "done" };
+    };
+
+    const io: ExecuteThreadIo = {
+      threadId: "T02",
+      hash: bundleHash,
+      infoJsonlPath: join(storageRoot, "logs", bundleHash, "T02.info.jsonl"),
+      cas,
+    };
+
+    let observedHead: string | null = null;
+    let observedHeadAtSecondYield: string | null = null;
+
+    const opts = makeOptions({
+      storageRoot,
+      maxRounds: 5,
+      awaitAfterEachYield: async () => {
+        const text = await readFile(join(bundleDir, "threads.json"), "utf8");
+        const parsed = JSON.parse(text) as Record<string, { head: string }>;
+        const head = parsed.T02?.head ?? null;
+        if (observedHead === null) {
+          observedHead = head;
+        } else if (observedHeadAtSecondYield === null) {
+          observedHeadAtSecondYield = head;
+        }
+      },
+    });
+
+    const result = await executeThread(
+      wf,
+      "demo",
+      { prompt: "p", steps: [] },
+      opts,
+      io,
+      noLogger(),
+    );
+    expect(result.returnCode).toBe(0);
+
+    expect(observedHead).not.toBeNull();
+    expect(observedHeadAtSecondYield).not.toBeNull();
+    expect(observedHead).not.toBe(observedHeadAtSecondYield);
+
+    const firstState = readCasNode(casDir, observedHead ?? "");
+    expect(firstState.type).toBe("state");
+    expect((firstState.payload as Record<string, unknown>).role).toBe("planner");
+    expect((firstState.payload as Record<string, unknown>).ancestors).toEqual([]);
+
+    const secondState = readCasNode(casDir, observedHeadAtSecondYield ?? "");
+    expect(secondState.type).toBe("state");
+    expect((secondState.payload as Record<string, unknown>).role).toBe("coder");
+    expect((secondState.payload as Record<string, unknown>).ancestors).toEqual([observedHead]);
+    expect((secondState.payload as Record<string, unknown>).start).toBe(
+      (firstState.payload as Record<string, unknown>).start,
+    );
+  });
+
+  test("on completion: removes threads.json entry, appends history with __end__ head", async () => {
+    const cas = createCasStore(casDir);
+
+    const wf: WorkflowFn = async function* (
+      _thread: ThreadContext,
+      runtime: WorkflowRuntime,
+    ): AsyncGenerator<RoleOutput, WorkflowCompletion> {
+      const h = await runtime.cas.put("only-step");
+      yield { role: "only", contentHash: h, meta: {}, refs: [h] };
+      return { returnCode: 0, summary: "completed" };
+    };
+
+    const io: ExecuteThreadIo = {
+      threadId: "T03",
+      hash: bundleHash,
+      infoJsonlPath: join(storageRoot, "logs", bundleHash, "T03.info.jsonl"),
+      cas,
+    };
+
+    const result = await executeThread(
+      wf,
+      "demo",
+      { prompt: "p", steps: [] },
+      makeOptions({ storageRoot, maxRounds: 5 }),
+      io,
+      noLogger(),
+    );
+
+    expect(result.returnCode).toBe(0);
+
+    const indexText = await readFile(join(bundleDir, "threads.json"), "utf8");
+    const indexParsed = JSON.parse(indexText) as Record<string, unknown>;
+    expect(indexParsed).toEqual({});
+
+    const historyPath = await firstHistoryFile(bundleDir);
+    const historyText = await readFile(historyPath, "utf8");
+    const lines = historyText.trim().split("\n");
+    expect(lines.length).toBe(1);
+    const entry = JSON.parse(lines[0] ?? "") as Record<string, unknown>;
+    expect(entry.threadId).toBe("T03");
+    expect(entry.head).toBe(result.rootHash);
+
+    const endNode = readCasNode(casDir, String(entry.head));
+    expect(endNode.type).toBe("state");
+    expect((endNode.payload as Record<string, unknown>).role).toBe("__end__");
+    expect((endNode.payload as Record<string, unknown>).meta).toEqual({
+      returnCode: 0,
+      summary: "completed",
+    });
+  });
+
+  test("does not write any .data.jsonl file under storageRoot", async () => {
+    const cas = createCasStore(casDir);
+
+    const wf: WorkflowFn = async function* (
+      _thread: ThreadContext,
+      runtime: WorkflowRuntime,
+    ): AsyncGenerator<RoleOutput, WorkflowCompletion> {
+      const h = await runtime.cas.put("step");
+      yield { role: "only", contentHash: h, meta: {}, refs: [h] };
+      return { returnCode: 0, summary: "done" };
+    };
+
+    const io: ExecuteThreadIo = {
+      threadId: "T04",
+      hash: bundleHash,
+      infoJsonlPath: join(storageRoot, "logs", bundleHash, "T04.info.jsonl"),
+      cas,
+    };
+
+    await executeThread(
+      wf,
+      "demo",
+      { prompt: "p", steps: [] },
+      makeOptions({ storageRoot, maxRounds: 5 }),
+      io,
+      noLogger(),
+    );
+
+    const fsp = await import("node:fs/promises");
+    const found: string[] = [];
+    async function walk(dir: string): Promise<void> {
+      let entries: { name: string; isDirectory: () => boolean; isFile: () => boolean }[];
+      try {
+        entries = await fsp.readdir(dir, { withFileTypes: true });
+      } catch {
+        return;
+      }
+      for (const ent of entries) {
+        const p = join(dir, ent.name);
+        if (ent.isDirectory()) {
+          await walk(p);
+        } else if (ent.isFile() && ent.name.endsWith(".data.jsonl")) {
+          found.push(p);
+        }
+      }
+    }
+    await walk(storageRoot);
+    expect(found).toEqual([]);
+  });
+});
+
+async function firstHistoryFile(bundleDir: string): Promise<string> {
+  const fsp = await import("node:fs/promises");
+  const dir = join(bundleDir, "history");
+  const entries = await fsp.readdir(dir);
+  const file = entries.find((n) => n.endsWith(".jsonl"));
+  if (file === undefined) {
+    throw new Error(`no history file under ${dir}`);
+  }
+  return join(dir, file);
+}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
xiaoju	81c582ae0e	feat: Phase 2 — engine write path (CAS nodes + threads.json) - Engine writes StartNode, StateNode, ContentMerkleNode as CAS blobs - threads.json tracks active threads, completed → history/{date}.jsonl - No more .data.jsonl writes - ancestors skip-list: [parent, ...parentAncestors] capped at 11 - Tests: 4 pass (engine write path) Refs #155, closes #157 小橘 <xiaoju@shazhou.work>	2026-05-09 07:53:44 +00:00
xiaoju	6f000512d2	feat: Phase 1 — CAS thread storage types + helpers - Add StartNode, StateNode, ContentMerkleNode types to workflow-protocol - Add collectRefs() to workflow-cas — extracts CAS hashes from StateNode payload - Add findReachableHashes() to workflow-cas — recursive mark traversal via refs[] - Tests: 7 pass (collect-refs + reachable) Refs #155, closes #156 小橘 <xiaoju@shazhou.work>	2026-05-09 07:30:47 +00:00
xiaoju	8f78a00063	docs: RFC v3 — named payload fields, refs as GC index, merge parent+ancestors - payload is source of truth with named fields (start, content, ancestors, compact) - refs[] auto-derived by collectRefs(), pure GC index - parent merged into ancestors[0] 小橘 <xiaoju@shazhou.work>	2026-05-09 07:12:29 +00:00
xiaoju	6c2a137aef	docs: update CAS thread storage RFC - StartNode prompt via refs[0] instead of inline - threads.json active-only, completed → history/{date}.jsonl - Content Merkle node carries role artifact refs - Extract phase expanded to produce refs[] 小橘 <xiaoju@shazhou.work>	2026-05-09 07:08:10 +00:00
xiaoju	6cd856ca99	docs: add RFC for CAS-based thread storage 小橘 <xiaoju@shazhou.work>	2026-05-09 06:34:24 +00:00
xiaoju	064696c558	docs: update architecture docs and package READMEs for post-split structure - Rewrite docs/architecture.md with 15-package map, dependency graph, updated engine paths - Update CLAUDE.md monorepo structure section - Add READMEs for: workflow-protocol, workflow-runtime, workflow-util, workflow-cas, workflow-register, workflow-execute, workflow-reactor - Fix agent READMEs: update deps from @uncaged/workflow to actual packages - Mark workflow-as-agent plan as outdated Fixes #153 小橘 <xiaoju@shazhou.work>	2026-05-09 04:39:57 +00:00
xiaoju	0f28e9b61a	refactor: remove non-index re-exports Remove re-exports from non-index files to enforce folder module discipline: - workflow-agent-hermes: remove buildAgentPrompt re-export - workflow-agent-cursor: remove buildAgentPrompt re-export - cli-workflow/cli-dispatch: remove CommandEntry/CommandGroup/DispatchFn/getCommandRegistry re-exports 小橘 <xiaoju@shazhou.work>	2026-05-09 04:19:49 +00:00
xiaoju	1ea56009a2	Merge pull request 'chore: rename dashboard folder' (#152 ) from chore/rename-dashboard-folder into main	2026-05-09 03:59:32 +00:00
xiaoju	6cc2481a16	chore: remove accidental pnpm-lock.yaml	2026-05-09 03:58:30 +00:00
xiaoju	44018bd17d	chore: rename packages/dashboard → packages/workflow-dashboard Align folder name with package name @uncaged/workflow-dashboard, consistent with all other packages in the monorepo.	2026-05-09 03:57:49 +00:00
xingyue	28c35bb3e0	Merge pull request 'refactor: 七包拆分 — protocol / runtime / util / cas / reactor / register / execute' (#151 ) from refactor/143-split-packages into main	2026-05-09 03:53:54 +00:00
xiaoju	b8b557baf6	fix: migrate template test imports from @uncaged/workflow to new packages 小橘 🍊（NEKO Team）	2026-05-09 03:51:48 +00:00
xingyue	727b4bb3ed	refactor(workflow): fix tsconfig references, template imports, delete old packages/workflow - Update root tsconfig.json references: replace packages/workflow with 6 new packages - Update cli-workflow tsconfig references to new packages - Add tsconfig references to workflow-util, workflow-runtime, workflow-execute - Fix workflow-agent-llm, workflow-agent-cursor, workflow-agent-hermes, workflow-util-agent tsconfig references (../workflow -> ../workflow-runtime) - Remove stale @uncaged/workflow deps from agent package.json files - Change template packages to import buildDescriptor from @uncaged/workflow-register - Normalize package.json exports field across all new packages - Delete old packages/workflow/ directory	2026-05-09 11:46:57 +08:00
xingyue	9bbdfc41bd	feat(execute): create @uncaged/workflow-execute + CLI migration Phase 7: Engine + extract + workflow-as-agent merged into execute package. All CLI imports migrated from @uncaged/workflow to specific packages. 105 CLI tests pass, 0 failures. Changes: - New @uncaged/workflow-execute package (engine/, extract/, workflow-as-agent) - CLI src/ and __tests__/ rewritten to import from split packages - bundle-validator updated to allow @uncaged/workflow-cas imports - ensure-uncaged-workflow-symlink creates symlinks for all new packages Ref: #143, closes #150	2026-05-09 11:35:03 +08:00
xingyue	b07f8cf166	feat(register): create @uncaged/workflow-register package Merges bundle/ + registry/ + config/ modules. The config↔registry circular dependency is resolved: ProviderConfig and WorkflowConfig now come from @uncaged/workflow-protocol. Ref: #143, closes #149	2026-05-09 11:16:27 +08:00
xingyue	1a1e8b3398	feat(cas,reactor): create @uncaged/workflow-cas and @uncaged/workflow-reactor Phase 4: CAS module extracted with Merkle types, hash functions, and fs-backed store. Imports CasStore type from protocol. Phase 5: Reactor (ReAct loop) extracted as independent package. Only depends on protocol — no cas or engine dependency. Ref: #143, closes #147, closes #148	2026-05-09 11:11:33 +08:00
xingyue	39d2a61686	refactor(runtime): types down to @uncaged/workflow-protocol All type definitions now originate from @uncaged/workflow-protocol. Runtime re-exports them for backward compatibility. Local AdvanceOutcome duplicate in create-workflow.ts removed (now imported from protocol). Ref: #143, closes #146	2026-05-09 11:09:38 +08:00
xingyue	bf0bc47a3f	feat(util): create @uncaged/workflow-util package Extract pure utility functions from workflow/src/util/ into standalone package. Types (Result, ok, err) now come from @uncaged/workflow-protocol. Contains: base32 encoding, ULID generation, structured logger, storage-root helpers, refs-field normalization. Ref: #143, closes #145	2026-05-09 11:08:04 +08:00
xingyue	2cffaad127	feat(protocol): create @uncaged/workflow-protocol package Extract all cross-package type definitions and constructor functions into a dedicated protocol layer. This is the foundation for the seven-package split (RFC #143). Contains: - Result<T,E>, ok(), err() - START, END constants - CasStore, WorkflowFn, RoleOutput, WorkflowCompletion - WorkflowDescriptor, WorkflowRoleDescriptor - ProviderConfig, WorkflowConfig, ResolvedModel (fixes config↔registry cycle) - RoleDefinition, Moderator, WorkflowDefinition - AgentFn, ExtractFn, and all thread context types Ref: #143, closes #144	2026-05-09 11:06:10 +08:00
xiaomo	9a3daac657	Merge pull request 'feat(workflow): ThreadReactor — generic ReAct loop + extract/supervisor migration' (#142 ) from feat/139-thread-reactor into main	2026-05-09 02:28:09 +00:00
xiaoju	b8f9ffcb59	feat(workflow): migrate supervisor to ThreadReactor (Phase 2) - Rewrite supervisor to use createThreadReactor + createLlmFn - No direct fetch/HTTP calls in supervisor - All 266 tests passing Refs #139, relates #141	2026-05-09 02:26:39 +00:00
xiaoju	a7171f05f6	feat(workflow): add ThreadReactor generic ReAct loop + migrate extract (Phase 1) - New src/reactor/ module: createThreadReactor, createLlmFn, types - Two-stage API: config (llm, systemPrompt, tools, toolHandler) + per-call (thread, input, schema) - All tool failures are recoverable (returned to LLM as error message) - Rewrite createExtract to use createThreadReactor - Delete reactExtract old implementation - Fix template test imports (START/END from runtime, validateWorkflowDescriptor from engine) 268 tests passing. Refs #139, relates #140	2026-05-09 02:15:38 +00:00
xiaoju	b53667a2aa	Merge pull request 'refactor(workflow): move descriptor validation out of runtime' (#135 ) from refactor/runtime-descriptor-boundary into main	2026-05-08 15:05:24 +00:00
Scott Wei	5b60fa6454	refactor(workflow-runtime): flatten package layout and centralize types Collapse bundle/cas/extract/util stubs into types.ts; move createWorkflow and Result helpers to src root. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-05-08 23:03:53 +08:00
xiaomo	2c0e744ebf	Merge pull request 'perf(serve): SSE live pump reads incrementally' (#137 ) from fix/130-sse-incremental into main	2026-05-08 15:00:44 +00:00
xiaomo	ae16f09688	Merge pull request 'feat(dashboard): hash routing + health check polling' (#138 ) from fix/128-dashboard-enhancements into main	2026-05-08 15:00:34 +00:00
xiaomo	73a3638ad9	Merge pull request 'fix(serve): error handling, CORS, body limit, CAS store reuse' (#136 ) from fix/120-serve-hardening into main	2026-05-08 15:00:32 +00:00
xingyue	7b0260cedd	feat(dashboard): hash routing + health check polling - Hash-based URL routing (#threads, #threads/{id}, #workflows) for bookmarkable/shareable thread links - Health check polls every 10s with reconnecting state - useHashRoute hook for clean route management Closes #128	2026-05-08 18:16:09 +08:00
xingyue	61fc1cfe1b	perf(serve): SSE live pump reads incrementally instead of full file Use Bun.file().slice() to read only new bytes from the last known offset instead of re-reading the entire JSONL file on every fs watch. - readNewBytes() helper with byte-offset tracking - Handles file truncation (resets offset) - Early return when no new data Closes #130	2026-05-08 18:14:04 +08:00
xingyue	6b1e728700	fix(serve): error handling, CORS, body limit, CAS store reuse - Global error handler (app.onError → 500 JSON) - JSON parse validation on POST routes (400) - CORS restricted to localhost origins - 1MB body size limit on POST (413) - CAS store created once per route group, not per-request - 6 new tests covering all changes Closes #120	2026-05-08 18:11:59 +08:00
xiaomo	dedab62c49	Merge pull request 'feat(dashboard): connect thread detail to SSE live stream' (#134 ) from feat/131-dashboard-sse into main	2026-05-08 09:53:10 +00:00
xingyue	a44f1f34a8	feat(dashboard): connect thread detail to SSE live stream Add useSSE hook that connects to /api/threads/:id/live via EventSource. Thread detail page now shows records in real-time with auto-scroll. - useSSE hook: EventSource connection, record accumulation, auto-reconnect with exponential backoff, cleanup on unmount - Thread detail: Live badge, SSE-first with fetch fallback, smooth scroll - Records clear on reconnect (server replays full file) Closes #131, testing verified per #133 Refs: #118	2026-05-08 17:52:10 +08:00
Scott Wei	8ff6f7e778	refactor(workflow): move descriptor validation out of runtime Keep @uncaged/workflow-runtime focused on bundle runtime capabilities by relocating descriptor validation implementation to @uncaged/workflow. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-05-08 17:45:15 +08:00
xiaoju	e04e75bdee	chore: remove stale self-referencing symlink 小橘 🍊（NEKO Team）	2026-05-08 09:35:32 +00:00
xiaoju	c65c29c1b5	Merge pull request 'refactor(workflow): simplify extraction + thread runtime contract' (#132 ) from refactor/thread-context-runtime into main	2026-05-08 09:34:26 +00:00
Scott Wei	cc3f2b576c	refactor(workflow): decouple agent context from CAS and fix monorepo checks Move CAS access into extract dependencies so AgentContext stays state-only, and clean up type/lint/check regressions across CLI/dashboard to keep full check green. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-05-08 17:30:07 +08:00
Scott Wei	884ff85205	refactor(workflow): remove dead extract retry export Drop unused llmExtractWithRetry implementation and public exports. Add solve-issue template coverage for tool_calls extraction path. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-05-08 17:10:31 +08:00
Scott Wei	a11cc62a81	refactor(workflow-runtime): use full ThreadContext in WorkflowFn Redefine WorkflowFn to accept a complete ThreadContext plus WorkflowRuntime dependencies, removing ThreadInput and WorkflowFnOptions. Move thread context construction into engine executeThread, update runtime loop/agent paths, and align templates/docs/tests with template-only definition exports. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-05-08 17:08:01 +08:00
Scott Wei	34f5e655d1	refactor(workflow): unify extraction behind ExtractFn Route createExtract through reactExtract with plain-JSON correction retry. Remove WorkflowFnOptions.llmProvider, ExtractMode, RoleDefinition.extractMode, ResolveRoleMetaFn. Runtime createWorkflow calls options.extract directly; engine passes extract only. Update templates, CLI skill docs, and tests. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-05-08 17:08:01 +08:00
xiaomo	44fb0694aa	Merge pull request 'feat(serve+dashboard): write endpoints, SSE live, run dialog' (#129 ) from feat/118-serve-write-sse into main	2026-05-08 09:01:09 +00:00
xingyue	cdcaff15ab	feat(serve+dashboard): write endpoints, SSE live, run dialog Serve API: - POST /api/threads — run a new thread - POST /api/threads/:id/kill — kill thread - POST /api/threads/:id/pause — pause thread - POST /api/threads/:id/resume — resume thread - GET /api/threads/:id/live — SSE stream of thread records Dashboard: - Run Thread dialog (select workflow, enter prompt, set maxRounds) - Thread detail controls (pause/resume/kill buttons) - postJson API helper for write operations 262 tests pass. Refs: #118	2026-05-08 16:07:02 +08:00
xiaomo	402479ddef	Merge pull request 'feat(dashboard): workflow dashboard' (#127 ) from feat/118-dashboard into main	2026-05-08 07:22:00 +00:00