docs: update wf-stateless-design.md to reflect new/resume semantics

Refs #101

.changeset/bootstrap-onboarding.md

@@ -1,9 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: expand bootstrap prompt with full onboarding and upgrade guide
-
-Bootstrap now covers two scenarios:
-- Fresh install: CLI + adapter installation, `uwf setup` configuration, skill installation, end-to-end verification
-- Upgrade: package update, skill regeneration, breaking change migrations (e.g. $START new/resume)

.changeset/bootstrap-preflight.md

@@ -1,8 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: bootstrap adds Step 0 environment pre-flight check
-
-- Pre-flight checks for node, pnpm/npm, global bin PATH, hermes CLI with FIX instructions (#112)
-- Install commands changed from npm to pnpm (with npm fallback)

.changeset/bootstrap-schema-path.md

@@ -1,9 +0,0 @@
----
-"@united-workforce/cli": patch
-"@united-workforce/util": patch
----
-
-fix: workflow-authoring flat schema example uses enum, bootstrap adds PATH guidance
-
-- workflow-authoring: flat schema example uses `enum: [done]` instead of bare `const` (#110.3)
-- bootstrap: adds `which hermes` check and PATH guidance for venv installs (#110.4)

.changeset/bootstrap-ux-fixes.md

@@ -1,10 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: preset provider base-url auto-fill, bootstrap ACP docs, friendlier name mismatch error
-
-- `uwf setup --provider dashscope` now auto-fills `--base-url` from preset list (#106)
-- Bootstrap guide documents uwf-hermes ACP dependency (`pip install hermes-agent[acp]`) (#107)
-- Bootstrap verify step uses inline workflow instead of missing `examples/eval-simple.yaml` (#107)
-- Workflow filename mismatch error now suggests how to fix it (#108)

.changeset/setup-ux-improvements.md

@@ -1,12 +0,0 @@
----
-"@united-workforce/cli": patch
----
-
-fix: setup UX improvements (#114)
-
-- Setup validates adapter availability and prints install command if missing
-- Setup prints "Config saved to <path> ✓" on success
-- Spawn ENOENT gives actionable error ("not found in PATH" + which command)
-- SQLite ExperimentalWarning suppressed via NODE_OPTIONS in spawned processes
-- Bootstrap VERSION reads cli package version (was reading util version)
-- Bootstrap PATH guidance is shell-agnostic (no hardcoded .bashrc/.profile)

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@united-workforce/cli",
-  "version": "0.2.0",
+  "version": "0.1.1",
   "files": [
     "src",
     "dist",

packages/cli/src/__tests__/prompt.test.ts

@@ -71,22 +71,12 @@ describe("prompt commands", () => {
   test("prompt bootstrap returns framework-agnostic setup instructions", () => {
     const result = cmdPromptBootstrap();
     expect(typeof result).toBe("string");
-    // Skills installation
     expect(result).toContain("uwf prompt usage");
     expect(result).toContain("uwf prompt workflow-authoring");
     expect(result).toContain("uwf prompt adapter-developing");
     expect(result).toContain("uwf-usage");
     expect(result).toContain("uwf-workflow-authoring");
     expect(result).toContain("uwf-adapter-developing");
-    // Fresh install scenario
-    expect(result).toContain("Fresh Install");
-    expect(result).toContain("uwf setup");
-    expect(result).toContain("--provider");
-    expect(result).toContain("--api-key");
-    expect(result).toContain("agent adapter");
-    // Upgrade scenario
-    expect(result).toContain("Upgrade");
-    expect(result).toContain("Migrate");
     // Should NOT contain Hermes-specific paths
     expect(result).not.toContain("~/.hermes/skills/");
     expect(result).not.toContain("> ~/.hermes/");

packages/cli/src/cli.ts

@@ -11,7 +11,7 @@ import {
   cmdPromptUsage,
   cmdPromptWorkflowAuthoring,
 } from "./commands/prompt.js";
-import { cmdSetup, cmdSetupInteractive, resolvePresetBaseUrl } from "./commands/setup.js";
+import { cmdSetup, cmdSetupInteractive } from "./commands/setup.js";
 import { cmdStepFork, cmdStepList, cmdStepRead, cmdStepShow } from "./commands/step.js";
 import {
   cmdThreadCancel,
@@ -558,14 +558,10 @@ program
     }) => {
       const storageRoot = resolveStorageRoot();
       runAction(async () => {
-        // Resolve preset base-url when provider is known but --base-url is omitted
-        const resolvedBaseUrl =
-          opts.baseUrl ??
-          (opts.provider !== undefined ? resolvePresetBaseUrl(opts.provider) : null);
-        if (opts.provider && resolvedBaseUrl && opts.apiKey && opts.model) {
+        if (opts.provider && opts.baseUrl && opts.apiKey && opts.model) {
           const result = await cmdSetup({
             provider: opts.provider,
-            baseUrl: resolvedBaseUrl,
+            baseUrl: opts.baseUrl,
             apiKey: opts.apiKey,
             model: opts.model,
             agent: opts.agent ?? undefined,
@@ -576,7 +572,7 @@ program
           await cmdSetupInteractive(storageRoot);
         } else {
           throw new Error(
-            "Non-interactive setup requires: --provider, --api-key, --model (--base-url is optional for preset providers)",
+            "Non-interactive setup requires all of: --provider, --base-url, --api-key, --model",
           );
         }
       });

packages/cli/src/commands/prompt.ts

@@ -1,32 +1,10 @@
-import { readFileSync } from "node:fs";
-import { dirname, join } from "node:path";
-import { fileURLToPath } from "node:url";
 import {
   generateAdapterDevelopingReference,
   generateUsageReference,
   generateWorkflowAuthoringReference,
+  VERSION,
 } from "@united-workforce/util";
 
-// CLI package version (for bootstrap prompt — uwf --version prints this)
-// Walk up from __dirname to find the nearest package.json (works from both src/ and dist/)
-function _findCliVersion(): string {
-  let dir = dirname(fileURLToPath(import.meta.url));
-  for (let i = 0; i < 5; i++) {
-    const candidate = join(dir, "package.json");
-    try {
-      const pkg = JSON.parse(readFileSync(candidate, "utf-8")) as { name?: string; version?: string };
-      if (pkg.name === "@united-workforce/cli") {
-        return pkg.version ?? "0.0.0";
-      }
-    } catch {
-      // not found, keep walking
-    }
-    dir = dirname(dir);
-  }
-  return "0.0.0";
-}
-const CLI_VERSION = _findCliVersion();
-
 export {
   generateAdapterDevelopingReference as cmdPromptAdapterDeveloping,
   generateUsageReference as cmdPromptUsage,
@@ -44,189 +22,33 @@ export function cmdPromptList(): ReadonlyArray<string> {
 }
 
 export function cmdPromptBootstrap(): string {
-  return `# uwf Bootstrap
+  return `# uwf Skill Bootstrap
 
-Set up or upgrade uwf (United Workforce) — from zero to running your first workflow.
+Install or update the uwf (United Workforce) skills for your agent.
 
-## Scenario A: Fresh Install
+## Steps
 
-### Step 0 — Environment pre-flight check
+1. **Check if uwf CLI is installed:**
+   \`\`\`bash
+   uwf --help
+   \`\`\`
+   If not installed: \`npm install -g @united-workforce/cli\`
+   Current version: ${VERSION}
 
-Run ALL checks below BEFORE installing anything. If any check fails, fix it first — do NOT skip ahead.
+2. **Install the skills** by running each command below and creating a skill from its output:
+   - Run \`uwf prompt usage\` → create a skill named \`uwf-usage\`
+   - Run \`uwf prompt workflow-authoring\` → create a skill named \`uwf-workflow-authoring\`
+   - Run \`uwf prompt adapter-developing\` → create a skill named \`uwf-adapter-developing\`
 
-\`\`\`bash
-# 1. Node.js
-node --version   # need v20+
-# FIX: install via nvm (https://github.com/nvm-sh/nvm) or your OS package manager
+   Each command outputs a complete SKILL.md with YAML frontmatter — use your agent framework's skill creation API to save them.
 
-# 2. Package manager — pnpm (preferred) or npm
-pnpm --version || npm --version
-# FIX for pnpm: corepack enable pnpm   (Node 20+)
-#   or: npm install -g pnpm
-# If pnpm is not an option, npm works too — adjust install commands below
+3. **Verify** the skills are loadable by your agent framework.
 
-# 3. Global bin directory is in PATH
-# For pnpm:
-pnpm bin -g       # prints the global bin directory
-# For npm:
-npm prefix -g     # global prefix; bin is <prefix>/bin
-# The directory printed above MUST be in your PATH.
-# Test: echo \\$PATH | tr ':' '\\\\n'
-# FIX: add the missing directory to your shell's startup file
-#   (e.g. ~/.bashrc, ~/.zshrc, ~/.profile, or fish config):
-#   export PATH="<global-bin-dir>:$PATH"
+## Updating
 
-# 4. (uwf-hermes only) hermes CLI
-which hermes
-# FIX: if hermes is in a virtualenv, add it to PATH:
-#   export PATH="$HOME/.hermes/hermes-agent/.venv/bin:$PATH"
-#   or create a symlink: ln -s ~/.hermes/hermes-agent/.venv/bin/hermes ~/.local/bin/hermes
-\`\`\`
-
-**All checks must pass before continuing.** If you had to modify PATH, verify the change persists by opening a new shell or sourcing your shell config.
-
-### Step 1 — Install CLI and agent adapter
-
-\`\`\`bash
-pnpm add -g @united-workforce/cli    # or: npm install -g @united-workforce/cli
-uwf --version   # should print ${CLI_VERSION}
-\`\`\`
-
-Install an agent adapter (at least one is required):
-
-| Adapter | Install | When to use |
-|---------|---------|-------------|
-| uwf-hermes | \`pnpm add -g @united-workforce/agent-hermes\` | When your agent framework is Hermes Agent |
-| uwf-claude-code | \`pnpm add -g @united-workforce/agent-claude-code\` | When using Claude Code CLI directly |
-| uwf-builtin | \`pnpm add -g @united-workforce/agent-builtin\` | Lightweight built-in agent (no external dependency) |
-
-**uwf-hermes** also requires the Hermes ACP plugin. After installing \`hermes-agent\`, run:
-\`\`\`bash
-pip install hermes-agent[acp]   # or: pip install -e .[acp] if installed from source
-\`\`\`
-
-Verify the adapter is installed: \`uwf-hermes --version\` (or whichever you chose).
-
-### Step 2 — Configure provider and model
-
-uwf needs an LLM provider to run agents. **Ask the user** for their provider, API key, and model, then run:
-
-\`\`\`bash
-uwf setup --provider <name> --base-url <url> --api-key <key> --model <model> [--agent <adapter>]
-\`\`\`
-
-Preset providers (base-url is auto-filled when using a preset name):
-openai, xai, openrouter, venice, dashscope, deepseek, siliconflow, volcengine, kimi, glm, stepfun, minimax, ollama
-
-Example:
-\`\`\`bash
-uwf setup --provider openrouter --api-key sk-or-... --model anthropic/claude-sonnet-4 --agent uwf-hermes
-\`\`\`
-
-If the user doesn't know what to choose, suggest \`openrouter\` with \`anthropic/claude-sonnet-4\` as a sensible default.
-
-Config is saved to \`~/.uwf/config.yaml\`. Verify with \`cat ~/.uwf/config.yaml\`.
-
-### Step 3 — Install skills
-
-Run each command and create a skill from its output:
-
-\`\`\`bash
-uwf prompt usage               # → save as skill "uwf-usage"
-uwf prompt workflow-authoring  # → save as skill "uwf-workflow-authoring"
-uwf prompt adapter-developing  # → save as skill "uwf-adapter-developing"
-\`\`\`
-
-Each command outputs a complete SKILL.md with YAML frontmatter. Use your agent framework's skill creation API to save them (e.g. \`skill_manage(action='create', name='uwf-usage', content=<output>)\`).
-
-Verify skills are installed by listing them (e.g. \`skills_list()\`) and confirming all three appear.
-
-### Step 4 — Verify end-to-end
-
-Create a minimal workflow file to test your setup:
-
-\`\`\`bash
-cat > /tmp/hello.yaml << 'YAML'
-name: hello
-description: Minimal smoke test
-roles:
-  greeter:
-    description: "Greet the user"
-    goal: "Respond with a friendly greeting"
-    capabilities: []
-    procedure: "Write a short greeting based on the prompt."
-    output: "A greeting message."
-    frontmatter:
-      type: object
-      properties:
-        $status: { enum: [done] }
-        message: { type: string }
-      required: [$status, message]
-graph:
-  $START:
-    new: { role: greeter, prompt: "Say hello to the user." }
-    resume: { role: greeter, prompt: "Greet the user again." }
-  greeter:
-    done: { role: "$END", prompt: "Done." }
-YAML
-\`\`\`
-
-Then run:
-
-\`\`\`bash
-uwf thread start /tmp/hello.yaml -p "Hello, world!"
-uwf thread exec <thread-id>
-uwf thread show <thread-id>
-\`\`\`
-
-If the thread reaches \`$END\` with status \`completed\`, the setup is working.
-
-## Scenario B: Upgrade from Previous Version
-
-### Step 1 — Update packages
-
-\`\`\`bash
-pnpm add -g @united-workforce/cli@latest    # or: npm install -g @united-workforce/cli@latest
-uwf --version   # should print ${CLI_VERSION}
-
-# Also update your adapter(s)
-pnpm add -g @united-workforce/agent-hermes@latest
-\`\`\`
-
-### Step 2 — Regenerate skills
-
-Skill content is bundled with the CLI — always regenerate after upgrading:
-
-\`\`\`bash
-uwf prompt usage               # → update skill "uwf-usage"
-uwf prompt workflow-authoring  # → update skill "uwf-workflow-authoring"
-uwf prompt adapter-developing  # → update skill "uwf-adapter-developing"
-\`\`\`
-
-### Step 3 — Migrate workflow YAML files (if needed)
-
-Check the changelog for breaking changes. Known migrations:
-
-- **v0.2.0**: \`$START._\` → \`$START.new\` + \`$START.resume\`. All workflow YAML files must be updated:
-  \`\`\`yaml
-  # Before (v0.1.x)
-  $START:
-    _: { role: planner, prompt: "..." }
-
-  # After (v0.2.0+)
-  $START:
-    new: { role: planner, prompt: "..." }
-    resume: { role: planner, prompt: "Review previous run and continue." }
-  \`\`\`
-
-Update all \`.workflow/\` and \`.workflows/\` YAML files in your projects. \`uwf workflow add\` will reject files with the old \`_\` syntax.
-
-### Step 4 — Verify
-
-\`\`\`bash
-uwf thread start <your-workflow> -p "upgrade test"
-uwf thread exec <thread-id>
-\`\`\`
+When \`uwf\` is upgraded, re-run \`uwf prompt bootstrap\` and follow the steps again.
+The skill content is bundled with the CLI — always use \`uwf prompt <name>\` to get
+content matching your installed version.
 
 ## Available prompts
 
@@ -235,7 +57,6 @@ uwf prompt list                # list available prompt names
 uwf prompt usage               # CLI usage guide
 uwf prompt workflow-authoring  # workflow YAML design guide
 uwf prompt adapter-developing  # building agent adapters
-uwf prompt bootstrap           # this guide
 \`\`\`
 `;
 }

packages/cli/src/commands/setup.ts

@@ -1,4 +1,3 @@
-import { execFileSync } from "node:child_process";
 import { existsSync, mkdirSync, readdirSync, readFileSync, statSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 import { stdin as input, stdout as output } from "node:process";
@@ -73,12 +72,6 @@ const PRESET_PROVIDERS = [
   { name: "ollama", label: "Ollama (local)", baseUrl: "http://localhost:11434/v1" },
 ] as const;
 
-/** Look up the base URL for a preset provider name. Returns null if not a preset. */
-export function resolvePresetBaseUrl(providerName: string): string | null {
-  const preset = PRESET_PROVIDERS.find((p) => p.name === providerName);
-  return preset !== undefined ? preset.baseUrl : null;
-}
-
 type SetupArgs = {
   provider: string;
   baseUrl: string;
@@ -182,6 +175,7 @@ export async function _discoverAgents(): Promise<string[]> {
 
 async function _tryWhichDiscovery(): Promise<string[] | null> {
   try {
+    const { execFileSync } = await import("node:child_process");
     const text = execFileSync("which", ["-a", "uwf-hermes", "uwf-claude-code", "uwf-cursor"], {
       encoding: "utf-8",
       stdio: ["pipe", "pipe", "pipe"],
@@ -397,37 +391,6 @@ function mergeConfig(existing: Record<string, unknown>, args: SetupArgs): Record
   };
 }
 
-/**
- * Check if the configured adapter binary (and its dependencies) are in PATH.
- * Returns warnings array — empty means all good.
- */
-export function _checkAdapterAvailability(agentName: string): string[] {
-  const warnings: string[] = [];
-  const binary = `uwf-${agentName}`;
-
-  try {
-    execFileSync("which", [binary], { encoding: "utf8", stdio: ["pipe", "pipe", "pipe"] });
-  } catch {
-    warnings.push(
-      `${binary} not found in PATH. Install it: pnpm add -g @united-workforce/agent-${agentName}`,
-    );
-    return warnings; // skip dependency check if adapter itself is missing
-  }
-
-  // uwf-hermes depends on hermes CLI
-  if (agentName === "hermes") {
-    try {
-      execFileSync("which", ["hermes"], { encoding: "utf8", stdio: ["pipe", "pipe", "pipe"] });
-    } catch {
-      warnings.push(
-        'hermes CLI not found in PATH (required by uwf-hermes). Fix: export PATH="$HOME/.hermes/hermes-agent/.venv/bin:$PATH"',
-      );
-    }
-  }
-
-  return warnings;
-}
-
 /**
  * Non-interactive setup. All required args provided via CLI flags.
  */
@@ -442,28 +405,15 @@ export async function cmdSetup(args: SetupArgs): Promise<Record<string, unknown>
 
   writeFileSync(configPath, stringify(merged, { indent: 2 }), "utf8");
 
-  // Print config path to stderr (stdout is reserved for JSON output)
-  // biome-ignore lint/nursery/noConsole: CLI user-facing output
-  console.error(`Config saved to ${configPath} ✓`);
-
   // Validate model connectivity
   const validation = await validateModel(args.baseUrl, args.apiKey, args.model);
 
-  // Check adapter availability
-  const agentName = _agentNameFromBinary(args.agent ?? "hermes");
-  const adapterWarnings = _checkAdapterAvailability(agentName);
-  for (const w of adapterWarnings) {
-    // biome-ignore lint/nursery/noConsole: CLI user-facing output
-    console.error(`⚠ ${w}`);
-  }
-
   return {
     configPath,
     provider: args.provider,
     model: args.model,
     defaultAgent: merged.defaultAgent,
     validation,
-    adapterWarnings,
   };
 }
 

packages/cli/src/commands/thread.ts

@@ -1001,21 +1001,9 @@ function spawnAgent(
       stdio: ["ignore", "pipe", "pipe"],
       maxBuffer: 50 * 1024 * 1024, // 50 MB — stream-json output can be large
       cwd,
-      env: {
-        ...process.env,
-        NODE_OPTIONS: [process.env.NODE_OPTIONS, "--disable-warning=ExperimentalWarning"]
-          .filter(Boolean)
-          .join(" "),
-      },
     });
   } catch (e) {
     const err = e as NodeJS.ErrnoException & { stderr?: Buffer | string | null };
-    if (err.code === "ENOENT") {
-      failStep(
-        plog,
-        `"${agent.command}" not found in PATH. Install it or check your PATH config. Run: which ${agent.command}`,
-      );
-    }
     const stderr =
       err.stderr == null
         ? ""
@@ -1254,12 +1242,6 @@ async function cmdThreadStepBackground(
   const child = spawn(scriptPath, args, {
     detached: true,
     stdio: "ignore",
-    env: {
-      ...process.env,
-      NODE_OPTIONS: [process.env.NODE_OPTIONS, "--disable-warning=ExperimentalWarning"]
-        .filter(Boolean)
-        .join(" "),
-    },
   });
 
   child.unref();

packages/cli/src/validate.ts

@@ -99,7 +99,7 @@ export function checkWorkflowFilenameConsistency(
 ): string | null {
   const expected = workflowNameFromPath(filePath);
   if (payload.name !== expected) {
-    return `workflow name mismatch: file "${basename(filePath)}" implies name "${expected}" but YAML declares name "${payload.name}". Either rename the file to "${payload.name}.yaml" or change the YAML \`name\` field to "${expected}"`;
+    return `workflow name mismatch: file "${basename(filePath)}" implies name "${expected}" but YAML declares name "${payload.name}"`;
   }
   return null;
 }

packages/util/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@united-workforce/util",
-  "version": "0.1.2",
+  "version": "0.1.1",
   "files": [
     "src",
     "dist",

packages/util/src/index.ts

@@ -15,7 +15,7 @@ export {
   validateFrontmatter,
 } from "./frontmatter-markdown/index.js";
 export { createLogger } from "./logger.js";
-
+export { generateModeratorReference } from "./moderator-reference.js";
 export type {
   CreateProcessLoggerOptions,
   ProcessLogFn,
@@ -35,3 +35,4 @@ export { extractUlidTimestamp, generateUlid } from "./ulid.js";
 export { generateUsageReference } from "./usage-reference.js";
 export { VERSION } from "./version.js";
 export { generateWorkflowAuthoringReference } from "./workflow-authoring-reference.js";
+export { generateYamlReference } from "./yaml-reference.js";

packages/util/src/moderator-reference.ts

@@ -0,0 +1,57 @@
+export function generateModeratorReference(): string {
+  return `# Moderator Reference
+
+## Overview
+
+The moderator is the workflow engine's routing component. It evaluates the directed graph defined in the workflow YAML to determine the next role (or \`$END\`) after each step — with zero LLM cost.
+
+## Status-Based Routing
+
+The moderator uses **status-based routing**: it inspects the previous step's extracted output (specifically the \`$status\` field) and looks up the corresponding edge in the graph.
+
+### Graph Structure
+
+The graph is a nested map: \`Record<Role | "$START", Record<Status, Target>>\`. Each role maps its possible \`$status\` values to a target with a \`role\` and \`prompt\`:
+
+\`\`\`yaml
+graph:
+  $START:
+    new: { role: planner, prompt: "Analyze the issue." }
+    resume: { role: planner, prompt: "Review the previous run output and continue." }
+  planner:
+    ready: { role: developer, prompt: "Implement the plan (CAS hash: {{{plan}}})." }
+    insufficient_info: { role: $END, prompt: "Not enough info." }
+  developer:
+    done: { role: reviewer, prompt: "Review branch {{{branch}}} at {{{worktree}}}." }
+    failed: { role: $END, prompt: "Developer failed: {{{reason}}}." }
+  reviewer:
+    approved: { role: tester, prompt: "Run tests on {{{branch}}} at {{{worktree}}}." }
+    rejected: { role: developer, prompt: "Fix issues: {{{comments}}}." }
+\`\`\`
+
+### Routing Algorithm
+
+1. Look up \`graph[lastRole]\` to get the status map for the current role
+2. Look up \`statusMap[lastOutput.$status]\` to get the target
+3. If target role is \`$END\`, mark thread as completed
+4. Otherwise, render the edge prompt (Mustache templates with \`{{{field}}}\` from output) and spawn the next agent
+
+### Edge Prompts and Mustache Templates
+
+Edge prompts use triple-brace Mustache syntax (\`{{{field}}}\`) to interpolate values from the previous step's output into the next agent's task prompt. This passes structured data (branch names, file paths, CAS hashes) between roles without manual wiring.
+
+## Special Nodes
+
+- \`$START\` — entry point; uses status keys \`new\` (first start) and \`resume\` (resuming a completed thread)
+- \`$END\` — terminal node; thread completes when reached and is moved to history
+
+## Integration with Steps
+
+Each \`uwf thread exec\` cycle:
+1. Moderator reads the thread's head step output
+2. Looks up \`graph[lastRole][output.$status]\` to pick the next role
+3. If next is \`$END\`, marks thread as completed
+4. Otherwise, renders the edge prompt and spawns the agent for the selected role
+5. Extract pipeline parses agent output → new step node → append to CAS chain
+`;
+}

packages/util/src/workflow-authoring-reference.ts

@@ -92,21 +92,16 @@ Add any fields you need for data passing between roles. These are available in e
 
 ### Flat Schema (Single Status)
 
-When a role has only one outcome, use \`enum\` with a single value:
+When a role has only one outcome:
 
 \`\`\`yaml
 frontmatter:
-  type: object
   properties:
-    $status:
-      type: string
-      enum: [done]
+    $status: { const: "done" }
     summary: { type: string }
   required: [$status, summary]
 \`\`\`
 
-Note: \`$status: { const: "done" }\` is **not** valid in flat schemas — the validator requires \`enum\` or \`oneOf\` with \`const\`. Use \`const\` only inside \`oneOf\` variants.
-
 ## Graph Routing
 
 The graph maps each role's \`$status\` values to the next role:

packages/util/src/yaml-reference.ts

@@ -0,0 +1,83 @@
+export function generateYamlReference(): string {
+  return `# Workflow YAML Schema Reference
+
+## Top-Level Structure
+
+A workflow YAML file defines the complete workflow specification:
+
+\`\`\`yaml
+name: solve-issue          # verb-first kebab-case identifier
+description: "..."         # human-readable description
+
+roles:                     # named actors in the workflow
+  planner:
+    description: "Analyzes issue and outputs a plan"
+    goal: "You are a planning agent."
+    capabilities:
+      - issue-analysis
+      - planning
+    procedure: |
+      1. Read the issue
+      2. Produce a test spec
+    output: "Output the plan summary. Set $status to ready or insufficient_info."
+    frontmatter:           # JSON Schema for structured output (drives routing)
+      oneOf:
+        - properties:
+            $status: { const: ready }
+            plan: { type: string }
+          required: [$status, plan]
+        - properties:
+            $status: { const: insufficient_info }
+          required: [$status]
+
+graph:                     # status-based routing (nested map)
+  $START:
+    new: { role: planner, prompt: "Analyze the issue." }
+    resume: { role: planner, prompt: "Review the previous run output and continue." }
+  planner:
+    ready: { role: developer, prompt: "Implement plan {{{plan}}}." }
+    insufficient_info: { role: $END, prompt: "Not enough info." }
+\`\`\`
+
+## roles
+
+Each role defines an actor in the workflow:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| \`description\` | string | Short description of the role's purpose |
+| \`goal\` | string | System-level goal statement for the agent |
+| \`capabilities\` | string[] | Tags describing what the role can do |
+| \`procedure\` | string | Step-by-step instructions for the agent |
+| \`output\` | string | Description of expected output format |
+| \`frontmatter\` | JSON Schema | Defines the structured output the agent must produce |
+
+### frontmatter
+
+The \`frontmatter\` field is a standard JSON Schema object. The extract pipeline validates agent output against it. Key conventions:
+- \`$status\` field drives routing decisions in the graph
+- Use \`const\` or \`enum\` to constrain status values
+- Use \`oneOf\` to define multiple valid output shapes (one per status)
+- All \`required\` fields must appear in the agent's frontmatter output
+
+## graph
+
+The graph is a nested map defining status-based routing:
+
+\`\`\`
+Record<Role | "$START", Record<Status, { role: string, prompt: string }>>
+\`\`\`
+
+| Level | Key | Value |
+|-------|-----|-------|
+| Outer | Role name or \`$START\` | Status map for that role |
+| Inner | \`$status\` value | Target: \`{ role, prompt }\` |
+
+### Special Nodes
+- \`$START\` — entry point; uses status keys \`new\` (first start) and \`resume\` (resuming a completed thread)
+- \`$END\` — terminal node; thread completes when reached
+
+### Edge Prompts
+Prompts use triple-brace Mustache templates (\`{{{field}}}\`) to interpolate values from the previous step's output. Example: \`"Implement plan {{{plan}}} in repo {{{repoPath}}}."\`
+`;
+}