@hivehub/rulebook 4.2.1 → 4.3.0

package/templates/core/RALPH.md

@@ -1,471 +1,471 @@
-<!-- RALPH:START -->
-# Ralph Autonomous Loop Integration
-
-Ralph is an autonomous AI agent loop that iteratively solves tasks with fresh context for each iteration. This document explains how to use Ralph with your Rulebook project.
-
-## Overview
-
-Ralph enables:
-- **Autonomous Multi-Iteration Development**: AI agents solve tasks across multiple iterations
-- **Quality Gates**: Automatic checks for type-safety, linting, testing, and coverage
-- **Progress Tracking**: Detailed history of each iteration with metrics and learnings
-- **Fresh Context per Iteration**: Each iteration starts with clean context, avoiding context window exhaustion
-- **Graceful Interruption**: Pause and resume capabilities for long-running tasks
-
-## Quick Start
-
-### Initialize Ralph
-```bash
-rulebook ralph init
-```
-This creates:
-- `.rulebook/ralph/prd.json` - Product Requirements Document from rulebook tasks
-- `.rulebook/ralph/state.json` - Current loop state and progress
-- `.rulebook/ralph/history/` - Per-iteration metadata and results
-
-### Run Ralph Loop
-```bash
-rulebook ralph run --max-iterations 10 --tool claude
-```
-Flags:
-- `--max-iterations N` - Maximum iterations before stopping (default: 10)
-- `--tool TOOL` - AI tool to use: `claude`, `amp`, or `gemini` (default: claude)
-
-### Check Status
-```bash
-rulebook ralph status
-```
-Shows:
-- Current iteration count
-- Completed vs. pending tasks
-- Success rate and quality metrics
-
-### View History
-```bash
-rulebook ralph history --limit 5
-```
-Displays:
-- Last 5 iterations with status and duration
-- Task completions and failures
-- Quality gate results
-
-### Pause/Resume
-```bash
-rulebook ralph pause    # Gracefully pause the loop
-rulebook ralph resume   # Continue from where it paused
-```
-
-## How Ralph Works
-
-### Iteration Loop
-```
-┌─────────────────────────────────────────┐
-│ 1. Load Next Pending Task from PRD      │
-├─────────────────────────────────────────┤
-│ 2. Execute Agent with Task Context      │
-│    - Fresh context per iteration        │
-│    - Task description and acceptance    │
-│      criteria from rulebook             │
-│    - Previous iteration learnings       │
-├─────────────────────────────────────────┤
-│ 3. Parse Agent Output                   │
-│    - Extract quality gate results       │
-│    - Extract learnings and errors       │
-│    - Extract git commit hash            │
-├─────────────────────────────────────────┤
-│ 4. Record Iteration Results             │
-│    - Save to history/.iteration-N.json  │
-│    - Update task status in PRD          │
-│    - Log progress to progress.txt       │
-├─────────────────────────────────────────┤
-│ 5. Check Loop Continuation              │
-│    - All tasks completed?    → Stop     │
-│    - Max iterations reached? → Stop     │
-│    - Otherwise              → Loop      │
-└─────────────────────────────────────────┘
-```
-
-### Quality Gates
-Ralph automatically checks these gates after each iteration:
-
-| Gate | Description | Pass Criteria |
-|------|-------------|---------------|
-| **type-check** | TypeScript compilation | No errors |
-| **lint** | ESLint code quality | No errors |
-| **tests** | Unit tests | All passing |
-| **coverage** | Code coverage | ≥95% |
-| **security** | Vulnerability scan | No findings above `failOn` severity |
-
-**Security Gate** scans for vulnerabilities using the first available tool:
-1. **trivy** — container and filesystem scanning (`trivy fs --format json`)
-2. **semgrep** — static analysis (`semgrep --config auto --json`)
-3. **npm audit** — Node.js dependency audit (always available in Node projects)
-
-Configure via `.rulebook` file:
-```json
-{
-  "ralph": {
-    "securityGate": {
-      "enabled": true,
-      "failOn": "high",
-      "tool": "auto"
-    }
-  }
-}
-```
-
-| Setting | Default | Options | Description |
-|---------|---------|---------|-------------|
-| `enabled` | `true` | `true`/`false` | Enable/disable security gate |
-| `failOn` | `"high"` | `critical`/`high`/`moderate`/`low` | Minimum severity to fail |
-| `tool` | `"auto"` | `auto`/`npm-audit`/`trivy`/`semgrep` | Scanner to use |
-
-**Iteration Status:**
-- ✅ **success** - All 5 gates pass
-- ⚠️ **partial** - 2-4 gates pass
-- ❌ **failed** - 0-1 gates pass
-
-### Fresh Context Strategy
-Each iteration provides context to the agent:
-1. **Task Description** - What to implement
-2. **Acceptance Criteria** - How to verify success
-3. **Quality Requirements** - Type-check, lint, tests, coverage
-4. **Previous Learnings** - Lessons from prior iterations
-5. **Project Context** - Recent git commits, file structure
-6. **Git Status** - Current diff and uncommitted changes
-
-This allows agents to make progress without exhausting context window.
-
-## Task Sizing for Ralph
-
-### Good Task Sizes
-✅ **Small Tasks** (1-2 iterations)
-- Add single feature (button, form field)
-- Fix localized bug
-- Refactor single function/component
-- Add utility function
-
-Example:
-```markdown
-## Task: Add Dark Mode Toggle
-
-What Changes: Add UI toggle for theme switching
-
-Acceptance Criteria:
-- [ ] Toggle button visible in settings
-- [ ] Click toggles theme immediately
-- [ ] Selection persists in localStorage
-- [ ] All tests passing
-```
-
-✅ **Medium Tasks** (3-5 iterations)
-- Implement new API endpoint
-- Refactor component with dependencies
-- Add authentication system
-- Build complex feature with multiple parts
-
-Example:
-```markdown
-## Task: Implement User Authentication
-
-What Changes: Add JWT-based user auth
-
-Acceptance Criteria:
-- [ ] Login endpoint working
-- [ ] Token generation and validation
-- [ ] Protected routes enforced
-- [ ] Refresh token rotation
-- [ ] Error handling comprehensive
-```
-
-### Poor Task Sizes
-❌ **Too Large** (>5 iterations)
-- Build entire new module
-- Major architectural redesign
-- Complete feature system
-
-→ Break into smaller, focused tasks
-
-❌ **Too Vague** (unclear acceptance criteria)
-- "Improve performance"
-- "Make code better"
-- "Add tests"
-
-→ Define specific, measurable criteria
-
-## Configuration
-
-### .rulebook File
-```json
-{
-  "ralph": {
-    "enabled": true,
-    "maxIterations": 10,
-    "tool": "claude",
-    "maxContextLoss": 3,
-    "securityGate": {
-      "enabled": true,
-      "failOn": "high",
-      "tool": "auto"
-    },
-    "contextCompression": {
-      "enabled": true,
-      "recentCount": 3,
-      "threshold": 5
-    }
-  }
-}
-```
-
-| Setting | Default | Description |
-|---------|---------|-------------|
-| `enabled` | `true` | Enable Ralph features |
-| `maxIterations` | `10` | Max iterations per run |
-| `tool` | `claude` | Default AI tool: `claude`, `amp`, `gemini` |
-| `maxContextLoss` | `3` | Allow 3 context loss events before stopping |
-| `securityGate.enabled` | `true` | Enable 5th quality gate (vulnerability scan) |
-| `securityGate.failOn` | `"high"` | Fail on: `critical`, `high`, `moderate`, `low` |
-| `securityGate.tool` | `"auto"` | Scanner: `auto`, `npm-audit`, `trivy`, `semgrep` |
-| `contextCompression.enabled` | `true` | Compress old iteration history in prompts |
-| `contextCompression.recentCount` | `3` | Iterations shown in full detail |
-| `contextCompression.threshold` | `5` | Min iterations before compression activates |
-
-## MCP Integration
-
-Ralph provides 4 MCP tools for programmatic access:
-
-### rulebook_ralph_init
-Initialize Ralph and generate PRD from tasks
-```json
-{
-  "type": "resource",
-  "name": "rulebook_ralph_init",
-  "description": "Initialize Ralph autonomous loop"
-}
-```
-
-### rulebook_ralph_run
-Execute autonomous iteration loop
-```json
-{
-  "type": "resource",
-  "name": "rulebook_ralph_run",
-  "description": "Run Ralph iteration loop"
-}
-```
-
-### rulebook_ralph_status
-Get current loop state and progress
-```json
-{
-  "type": "resource",
-  "name": "rulebook_ralph_status",
-  "description": "Get Ralph loop status"
-}
-```
-
-### rulebook_ralph_get_iteration_history
-Retrieve iteration metadata and statistics
-```json
-{
-  "type": "resource",
-  "name": "rulebook_ralph_get_iteration_history",
-  "description": "Get Ralph iteration history"
-}
-```
-
-## Directory Structure
-
-Ralph stores all data in `.rulebook/ralph/`:
-```
-.rulebook/ralph/
-├── prd.json              # Product Requirements Document
-├── state.json            # Current loop state
-├── progress.txt          # Append-only progress log
-└── history/
-    ├── iteration-1.json  # Metadata from iteration 1
-    ├── iteration-2.json  # Metadata from iteration 2
-    └── ...
-```
-
-### prd.json Structure
-```json
-{
-  "version": "1.0",
-  "generated_at": "2026-02-26T12:00:00Z",
-  "project_name": "my-project",
-  "total_tasks": 5,
-  "tasks": [
-    {
-      "id": "task-1",
-      "title": "Task Title",
-      "description": "Task description",
-      "status": "pending|in_iteration|completed|blocked",
-      "priority": 1,
-      "acceptance_criteria": ["Criterion 1", "Criterion 2"],
-      "estimated_iterations": 2,
-      "created_at": "2026-02-26T12:00:00Z",
-      "updated_at": "2026-02-26T12:00:00Z"
-    }
-  ]
-}
-```
-
-### iteration-N.json Structure
-```json
-{
-  "iteration": 1,
-  "started_at": "2026-02-26T12:00:00Z",
-  "completed_at": "2026-02-26T12:05:00Z",
-  "task_id": "task-1",
-  "task_title": "Task Title",
-  "duration_ms": 300000,
-  "status": "success|partial|failed",
-  "git_commit": "abc1234",
-  "quality_checks": {
-    "type_check": true,
-    "lint": true,
-    "tests": true,
-    "coverage_met": true
-  }
-}
-```
-
-## Best Practices
-
-### 1. Clear Acceptance Criteria
-❌ Bad: "Make it work"
-✅ Good: "User can click button and see modal dialog with form"
-
-### 2. Limit to 1 Task per Run
-Run Ralph on single focused task for best results
-```bash
-# Focus on one task at a time
-rulebook ralph run --max-iterations 5
-```
-
-### 3. Review Iteration Results
-Check quality gates and learnings between runs
-```bash
-rulebook ralph history --limit 1  # See latest iteration
-```
-
-### 4. Break Large Features
-Don't try to implement entire systems in one run
-- Task 1: API endpoint with basic validation
-- Task 2: Database models and migrations
-- Task 3: Error handling and edge cases
-- Task 4: Tests and documentation
-
-### 5. Inspect Failed Iterations
-If status is "failed", examine what went wrong:
-```bash
-cat .rulebook/ralph/history/iteration-N.json
-cat .rulebook/ralph/progress.txt
-```
-
-### 6. Use Learnings from History
-Ralph extracts and stores learnings from iterations. Review these to understand what strategies worked.
-
-## Troubleshooting
-
-### Iteration Loop Stops Early
-**Cause**: Hit max iterations or all tasks completed
-
-**Solution**:
-- Check status: `rulebook ralph status`
-- View history: `rulebook ralph history`
-- Increase max iterations: `rulebook ralph run --max-iterations 20`
-
-### Quality Gates Failing
-**Cause**: Type-check, lint, tests, or coverage not meeting standards
-
-**Solution**:
-- Review failing gate in iteration history
-- Check PRD acceptance criteria match quality gates
-- Ensure task is small enough for single iteration
-
-### Context Loss Events
-**Cause**: Agent ran out of context window during iteration
-
-**Solution**:
-- Break task into smaller pieces
-- Reduce task description complexity
-- Increase agent context window if possible
-
-### Ralph State Gets Stuck
-**Cause**: Invalid state.json or corrupted PRD
-
-**Solution**:
-```bash
-rm .rulebook/ralph/state.json      # Reset state
-rulebook ralph init                # Regenerate PRD
-rulebook ralph run                 # Restart loop
-```
-
-## Examples
-
-### Example 1: Simple Feature Implementation
-Task: Add dark mode toggle button
-
-```bash
-# Initialize Ralph
-rulebook ralph init
-
-# Run for up to 3 iterations
-rulebook ralph run --max-iterations 3
-
-# Check results
-rulebook ralph history
-
-# Review final output
-cat .rulebook/ralph/history/iteration-*.json
-```
-
-### Example 2: Multiple Task Execution
-Tasks: Several small features to implement
-
-```bash
-# Run multiple tasks with generous iteration limit
-rulebook ralph run --max-iterations 20
-
-# Monitor progress
-watch -n 5 'rulebook ralph status'
-
-# Pause if needed
-rulebook ralph pause
-
-# Review and resume
-rulebook ralph history
-rulebook ralph resume
-```
-
-### Example 3: Debugging Failed Iteration
-```bash
-# Run into failure
-rulebook ralph run --max-iterations 3
-
-# See what failed
-cat .rulebook/ralph/history/iteration-3.json
-
-# View full progress log
-tail -50 .rulebook/ralph/progress.txt
-
-# Manual fix and resume
-rulebook ralph resume
-```
-
-## Limitations
-
-- **Single Sequential Loop**: Ralph processes one task at a time sequentially
-- **No Parallel Execution**: Tasks don't run in parallel (by design)
-- **No Task Dependencies**: No automatic dependency resolution between tasks
-- **Manual Task Ordering**: You order tasks by priority in PRD
-
-## Related Documentation
-
-- See `/.rulebook/specs/RULEBOOK.md` for task structure
-- See `/.rulebook/specs/QUALITY_ENFORCEMENT.md` for quality gates
-- See `/rulebook/AGENTS.md` for AI agent integration
-
-<!-- RALPH:END -->
+<!-- RALPH:START -->
+# Ralph Autonomous Loop Integration
+
+Ralph is an autonomous AI agent loop that iteratively solves tasks with fresh context for each iteration. This document explains how to use Ralph with your Rulebook project.
+
+## Overview
+
+Ralph enables:
+- **Autonomous Multi-Iteration Development**: AI agents solve tasks across multiple iterations
+- **Quality Gates**: Automatic checks for type-safety, linting, testing, and coverage
+- **Progress Tracking**: Detailed history of each iteration with metrics and learnings
+- **Fresh Context per Iteration**: Each iteration starts with clean context, avoiding context window exhaustion
+- **Graceful Interruption**: Pause and resume capabilities for long-running tasks
+
+## Quick Start
+
+### Initialize Ralph
+```bash
+rulebook ralph init
+```
+This creates:
+- `.rulebook/ralph/prd.json` - Product Requirements Document from rulebook tasks
+- `.rulebook/ralph/state.json` - Current loop state and progress
+- `.rulebook/ralph/history/` - Per-iteration metadata and results
+
+### Run Ralph Loop
+```bash
+rulebook ralph run --max-iterations 10 --tool claude
+```
+Flags:
+- `--max-iterations N` - Maximum iterations before stopping (default: 10)
+- `--tool TOOL` - AI tool to use: `claude`, `amp`, or `gemini` (default: claude)
+
+### Check Status
+```bash
+rulebook ralph status
+```
+Shows:
+- Current iteration count
+- Completed vs. pending tasks
+- Success rate and quality metrics
+
+### View History
+```bash
+rulebook ralph history --limit 5
+```
+Displays:
+- Last 5 iterations with status and duration
+- Task completions and failures
+- Quality gate results
+
+### Pause/Resume
+```bash
+rulebook ralph pause    # Gracefully pause the loop
+rulebook ralph resume   # Continue from where it paused
+```
+
+## How Ralph Works
+
+### Iteration Loop
+```
+┌─────────────────────────────────────────┐
+│ 1. Load Next Pending Task from PRD      │
+├─────────────────────────────────────────┤
+│ 2. Execute Agent with Task Context      │
+│    - Fresh context per iteration        │
+│    - Task description and acceptance    │
+│      criteria from rulebook             │
+│    - Previous iteration learnings       │
+├─────────────────────────────────────────┤
+│ 3. Parse Agent Output                   │
+│    - Extract quality gate results       │
+│    - Extract learnings and errors       │
+│    - Extract git commit hash            │
+├─────────────────────────────────────────┤
+│ 4. Record Iteration Results             │
+│    - Save to history/.iteration-N.json  │
+│    - Update task status in PRD          │
+│    - Log progress to progress.txt       │
+├─────────────────────────────────────────┤
+│ 5. Check Loop Continuation              │
+│    - All tasks completed?    → Stop     │
+│    - Max iterations reached? → Stop     │
+│    - Otherwise              → Loop      │
+└─────────────────────────────────────────┘
+```
+
+### Quality Gates
+Ralph automatically checks these gates after each iteration:
+
+| Gate | Description | Pass Criteria |
+|------|-------------|---------------|
+| **type-check** | TypeScript compilation | No errors |
+| **lint** | ESLint code quality | No errors |
+| **tests** | Unit tests | All passing |
+| **coverage** | Code coverage | ≥95% |
+| **security** | Vulnerability scan | No findings above `failOn` severity |
+
+**Security Gate** scans for vulnerabilities using the first available tool:
+1. **trivy** — container and filesystem scanning (`trivy fs --format json`)
+2. **semgrep** — static analysis (`semgrep --config auto --json`)
+3. **npm audit** — Node.js dependency audit (always available in Node projects)
+
+Configure via `.rulebook` file:
+```json
+{
+  "ralph": {
+    "securityGate": {
+      "enabled": true,
+      "failOn": "high",
+      "tool": "auto"
+    }
+  }
+}
+```
+
+| Setting | Default | Options | Description |
+|---------|---------|---------|-------------|
+| `enabled` | `true` | `true`/`false` | Enable/disable security gate |
+| `failOn` | `"high"` | `critical`/`high`/`moderate`/`low` | Minimum severity to fail |
+| `tool` | `"auto"` | `auto`/`npm-audit`/`trivy`/`semgrep` | Scanner to use |
+
+**Iteration Status:**
+- ✅ **success** - All 5 gates pass
+- ⚠️ **partial** - 2-4 gates pass
+- ❌ **failed** - 0-1 gates pass
+
+### Fresh Context Strategy
+Each iteration provides context to the agent:
+1. **Task Description** - What to implement
+2. **Acceptance Criteria** - How to verify success
+3. **Quality Requirements** - Type-check, lint, tests, coverage
+4. **Previous Learnings** - Lessons from prior iterations
+5. **Project Context** - Recent git commits, file structure
+6. **Git Status** - Current diff and uncommitted changes
+
+This allows agents to make progress without exhausting context window.
+
+## Task Sizing for Ralph
+
+### Good Task Sizes
+✅ **Small Tasks** (1-2 iterations)
+- Add single feature (button, form field)
+- Fix localized bug
+- Refactor single function/component
+- Add utility function
+
+Example:
+```markdown
+## Task: Add Dark Mode Toggle
+
+What Changes: Add UI toggle for theme switching
+
+Acceptance Criteria:
+- [ ] Toggle button visible in settings
+- [ ] Click toggles theme immediately
+- [ ] Selection persists in localStorage
+- [ ] All tests passing
+```
+
+✅ **Medium Tasks** (3-5 iterations)
+- Implement new API endpoint
+- Refactor component with dependencies
+- Add authentication system
+- Build complex feature with multiple parts
+
+Example:
+```markdown
+## Task: Implement User Authentication
+
+What Changes: Add JWT-based user auth
+
+Acceptance Criteria:
+- [ ] Login endpoint working
+- [ ] Token generation and validation
+- [ ] Protected routes enforced
+- [ ] Refresh token rotation
+- [ ] Error handling comprehensive
+```
+
+### Poor Task Sizes
+❌ **Too Large** (>5 iterations)
+- Build entire new module
+- Major architectural redesign
+- Complete feature system
+
+→ Break into smaller, focused tasks
+
+❌ **Too Vague** (unclear acceptance criteria)
+- "Improve performance"
+- "Make code better"
+- "Add tests"
+
+→ Define specific, measurable criteria
+
+## Configuration
+
+### .rulebook File
+```json
+{
+  "ralph": {
+    "enabled": true,
+    "maxIterations": 10,
+    "tool": "claude",
+    "maxContextLoss": 3,
+    "securityGate": {
+      "enabled": true,
+      "failOn": "high",
+      "tool": "auto"
+    },
+    "contextCompression": {
+      "enabled": true,
+      "recentCount": 3,
+      "threshold": 5
+    }
+  }
+}
+```
+
+| Setting | Default | Description |
+|---------|---------|-------------|
+| `enabled` | `true` | Enable Ralph features |
+| `maxIterations` | `10` | Max iterations per run |
+| `tool` | `claude` | Default AI tool: `claude`, `amp`, `gemini` |
+| `maxContextLoss` | `3` | Allow 3 context loss events before stopping |
+| `securityGate.enabled` | `true` | Enable 5th quality gate (vulnerability scan) |
+| `securityGate.failOn` | `"high"` | Fail on: `critical`, `high`, `moderate`, `low` |
+| `securityGate.tool` | `"auto"` | Scanner: `auto`, `npm-audit`, `trivy`, `semgrep` |
+| `contextCompression.enabled` | `true` | Compress old iteration history in prompts |
+| `contextCompression.recentCount` | `3` | Iterations shown in full detail |
+| `contextCompression.threshold` | `5` | Min iterations before compression activates |
+
+## MCP Integration
+
+Ralph provides 4 MCP tools for programmatic access:
+
+### rulebook_ralph_init
+Initialize Ralph and generate PRD from tasks
+```json
+{
+  "type": "resource",
+  "name": "rulebook_ralph_init",
+  "description": "Initialize Ralph autonomous loop"
+}
+```
+
+### rulebook_ralph_run
+Execute autonomous iteration loop
+```json
+{
+  "type": "resource",
+  "name": "rulebook_ralph_run",
+  "description": "Run Ralph iteration loop"
+}
+```
+
+### rulebook_ralph_status
+Get current loop state and progress
+```json
+{
+  "type": "resource",
+  "name": "rulebook_ralph_status",
+  "description": "Get Ralph loop status"
+}
+```
+
+### rulebook_ralph_get_iteration_history
+Retrieve iteration metadata and statistics
+```json
+{
+  "type": "resource",
+  "name": "rulebook_ralph_get_iteration_history",
+  "description": "Get Ralph iteration history"
+}
+```
+
+## Directory Structure
+
+Ralph stores all data in `.rulebook/ralph/`:
+```
+.rulebook/ralph/
+├── prd.json              # Product Requirements Document
+├── state.json            # Current loop state
+├── progress.txt          # Append-only progress log
+└── history/
+    ├── iteration-1.json  # Metadata from iteration 1
+    ├── iteration-2.json  # Metadata from iteration 2
+    └── ...
+```
+
+### prd.json Structure
+```json
+{
+  "version": "1.0",
+  "generated_at": "2026-02-26T12:00:00Z",
+  "project_name": "my-project",
+  "total_tasks": 5,
+  "tasks": [
+    {
+      "id": "task-1",
+      "title": "Task Title",
+      "description": "Task description",
+      "status": "pending|in_iteration|completed|blocked",
+      "priority": 1,
+      "acceptance_criteria": ["Criterion 1", "Criterion 2"],
+      "estimated_iterations": 2,
+      "created_at": "2026-02-26T12:00:00Z",
+      "updated_at": "2026-02-26T12:00:00Z"
+    }
+  ]
+}
+```
+
+### iteration-N.json Structure
+```json
+{
+  "iteration": 1,
+  "started_at": "2026-02-26T12:00:00Z",
+  "completed_at": "2026-02-26T12:05:00Z",
+  "task_id": "task-1",
+  "task_title": "Task Title",
+  "duration_ms": 300000,
+  "status": "success|partial|failed",
+  "git_commit": "abc1234",
+  "quality_checks": {
+    "type_check": true,
+    "lint": true,
+    "tests": true,
+    "coverage_met": true
+  }
+}
+```
+
+## Best Practices
+
+### 1. Clear Acceptance Criteria
+❌ Bad: "Make it work"
+✅ Good: "User can click button and see modal dialog with form"
+
+### 2. Limit to 1 Task per Run
+Run Ralph on single focused task for best results
+```bash
+# Focus on one task at a time
+rulebook ralph run --max-iterations 5
+```
+
+### 3. Review Iteration Results
+Check quality gates and learnings between runs
+```bash
+rulebook ralph history --limit 1  # See latest iteration
+```
+
+### 4. Break Large Features
+Don't try to implement entire systems in one run
+- Task 1: API endpoint with basic validation
+- Task 2: Database models and migrations
+- Task 3: Error handling and edge cases
+- Task 4: Tests and documentation
+
+### 5. Inspect Failed Iterations
+If status is "failed", examine what went wrong:
+```bash
+cat .rulebook/ralph/history/iteration-N.json
+cat .rulebook/ralph/progress.txt
+```
+
+### 6. Use Learnings from History
+Ralph extracts and stores learnings from iterations. Review these to understand what strategies worked.
+
+## Troubleshooting
+
+### Iteration Loop Stops Early
+**Cause**: Hit max iterations or all tasks completed
+
+**Solution**:
+- Check status: `rulebook ralph status`
+- View history: `rulebook ralph history`
+- Increase max iterations: `rulebook ralph run --max-iterations 20`
+
+### Quality Gates Failing
+**Cause**: Type-check, lint, tests, or coverage not meeting standards
+
+**Solution**:
+- Review failing gate in iteration history
+- Check PRD acceptance criteria match quality gates
+- Ensure task is small enough for single iteration
+
+### Context Loss Events
+**Cause**: Agent ran out of context window during iteration
+
+**Solution**:
+- Break task into smaller pieces
+- Reduce task description complexity
+- Increase agent context window if possible
+
+### Ralph State Gets Stuck
+**Cause**: Invalid state.json or corrupted PRD
+
+**Solution**:
+```bash
+rm .rulebook/ralph/state.json      # Reset state
+rulebook ralph init                # Regenerate PRD
+rulebook ralph run                 # Restart loop
+```
+
+## Examples
+
+### Example 1: Simple Feature Implementation
+Task: Add dark mode toggle button
+
+```bash
+# Initialize Ralph
+rulebook ralph init
+
+# Run for up to 3 iterations
+rulebook ralph run --max-iterations 3
+
+# Check results
+rulebook ralph history
+
+# Review final output
+cat .rulebook/ralph/history/iteration-*.json
+```
+
+### Example 2: Multiple Task Execution
+Tasks: Several small features to implement
+
+```bash
+# Run multiple tasks with generous iteration limit
+rulebook ralph run --max-iterations 20
+
+# Monitor progress
+watch -n 5 'rulebook ralph status'
+
+# Pause if needed
+rulebook ralph pause
+
+# Review and resume
+rulebook ralph history
+rulebook ralph resume
+```
+
+### Example 3: Debugging Failed Iteration
+```bash
+# Run into failure
+rulebook ralph run --max-iterations 3
+
+# See what failed
+cat .rulebook/ralph/history/iteration-3.json
+
+# View full progress log
+tail -50 .rulebook/ralph/progress.txt
+
+# Manual fix and resume
+rulebook ralph resume
+```
+
+## Limitations
+
+- **Single Sequential Loop**: Ralph processes one task at a time sequentially
+- **No Parallel Execution**: Tasks don't run in parallel (by design)
+- **No Task Dependencies**: No automatic dependency resolution between tasks
+- **Manual Task Ordering**: You order tasks by priority in PRD
+
+## Related Documentation
+
+- See `/.rulebook/specs/RULEBOOK.md` for task structure
+- See `/.rulebook/specs/QUALITY_ENFORCEMENT.md` for quality gates
+- See `/rulebook/AGENTS.md` for AI agent integration
+
+<!-- RALPH:END -->