Ralph

Ralph - Autonomous Agent Loop

Ralph is an autonomous AI agent loop that implements PRDs by repeatedly spawning fresh AI instances to complete user stories one by one.

What This Skill Does

This skill handles two phases:

1. PRD Conversion: Converts a markdown PRD to prd.json format
2. Ralph Execution: Launches the autonomous loop that implements each story

Usage

/ralph [path-to-prd.md]

If no path is provided, looks for prd.json in the project root and runs directly.

Examples:

• /ralph tasks/prd-my-feature.md - Convert PRD and run Ralph
• /ralph prd.json - Run Ralph with existing prd.json
• /ralph - Run Ralph (looks for prd.json)

Phase 1: PRD Conversion (if markdown provided)

Step 1: Read the PRD

Read the provided markdown PRD file.

Step 2: Archive Previous Run (if needed)

Check if prd.json exists with a different branchName. If so:

1. Read current prd.json and extract branchName
2. Check if it differs from the new feature's branch
3. If different AND prd-progress.txt has content:
   • Create archive folder: .claude/archive/YYYY-MM-DD-[feature-name]/
   • Copy current prd.json and prd-progress.txt to archive
   • Reset prd-progress.txt with fresh header

Step 3: Convert to prd.json

Parse the PRD and generate prd.json in the project root:

{
  "project": "[Project Name from PRD or auto-detected]",
  "branchName": "ralph/[feature-name-kebab-case]",
  "description": "[Feature description from PRD]",
  "userStories": [
    {
      "id": "US-001",
      "title": "[Story title]",
      "description": "As a [user], I want [feature] so that [benefit]",
      "acceptanceCriteria": [
        "Criterion 1",
        "Criterion 2",
        "Typecheck passes"
      ],
      "priority": 1,
      "passes": false,
      "notes": ""
    }
  ]
}

Conversion Rules

1. Each user story becomes one JSON entry
2. IDs: Sequential (US-001, US-002, etc.)
3. Priority: Based on dependency order (schema → backend → UI)
4. All stories: passes: false and empty notes
5. branchName: Derive from feature name, kebab-case, prefixed with ralph/
6. Always add: "Typecheck passes" to every story
7. For UI stories: Add "Verify in browser using dev-browser skill"

Story Sizing is Critical

Each story MUST be completable in ONE Ralph iteration.

Right-sized stories:

• Add a database column and migration
• Add a UI component to an existing page
• Update a server action with new logic
• Add a filter dropdown to a list

Too big (must split):

• "Build the entire dashboard" → Split into: schema, queries, UI components, filters
• "Add authentication" → Split into: schema, middleware, login UI, session handling

Rule of thumb: If you cannot describe the change in 2-3 sentences, it is too big.

Story Ordering

Stories execute in priority order. Earlier stories must NOT depend on later ones.

Correct order:

1. Schema/database changes (migrations)
2. Server actions / backend logic
3. UI components that use the backend
4. Dashboard/summary views that aggregate data

Verification Before Proceeding

Before running Ralph, verify:

• Previous run archived (if applicable)
• Each story is completable in one iteration
• Stories are ordered by dependencies
• Every story has "Typecheck passes"
• UI stories have "Verify in browser using dev-browser skill"
• Acceptance criteria are verifiable (not vague)

Phase 2: Ralph Execution

Step 1: Pre-flight Checks

Verify the following before starting:

1. Amp CLI installed

   which amp
   

   If not found, instruct user to install from https://ampcode.com

2. jq installed

   which jq
   

   If not found, instruct user to install (brew install jq on macOS)

3. Git working directory clean

   git status --porcelain
   

   If not clean, ask user to commit or stash changes

4. prd.json exists and is valid

   cat prd.json | jq .
   

   If invalid, show error and exit

Step 2: Create or Checkout Feature Branch

Read branchName from prd.json:

jq -r '.branchName' prd.json

If branch doesn't exist, create it from main:

git checkout -b $(jq -r '.branchName' prd.json)

If branch exists, checkout it:

git checkout $(jq -r '.branchName' prd.json)

Step 3: Run Ralph Loop

Execute the Ralph script:

bash .claude/scripts/ralph.sh [max_iterations]

Default is 10 iterations. The script will:

1. Spawn a fresh Amp instance with .claude/scripts/prompt.md
2. Amp picks the highest priority story with passes: false
3. Amp implements that single story
4. Amp runs quality checks (typecheck, lint, test)
5. If checks pass, Amp commits with message: feat: [Story ID] - [Story Title]
6. Amp updates prd.json to set passes: true for the story
7. Amp appends progress to prd-progress.txt
8. Loop repeats until all stories pass or max iterations reached

Step 4: Monitor Progress

Show the user how to monitor:

# See which stories are done
cat prd.json | jq '.userStories[] | {id, title, passes}'

# See learnings from previous iterations
cat prd-progress.txt

# Check git history
git log --oneline -10

Step 5: Completion

When ALL stories have passes: true, Ralph will output:

<promise>COMPLETE</promise>

And the loop exits successfully.

Ralph's Inner Loop (What Happens During Execution)

Each iteration spawns a fresh Amp instance with instructions from .claude/scripts/prompt.md.

The Amp instance:

1. Reads prd.json to pick the next story
2. Reads prd-progress.txt for context
3. Implements that ONE story
4. Runs quality checks
5. Updates AGENTS.md files if patterns are discovered
6. Commits if checks pass
7. Updates prd.json to mark story complete
8. Appends learnings to prd-progress.txt

Memory Between Iterations

The only memory between iterations:

• Git history (commits from previous iterations)
• prd-progress.txt (learnings and context)
• prd.json (which stories are done)

Each iteration is a fresh Amp instance with clean context.

Key Files

Troubleshooting

Ralph stops mid-execution

Check prd-progress.txt for the last completed story, then resume:

# Just run ralph again, it will continue from where it left off
bash .claude/scripts/ralph.sh

Amp context fills up

Enable auto-handoff in ~/.config/amp/settings.json:

{
  "amp.experimental.autoHandoff": { "context": 90 }
}

Story fails quality checks

Ralph won't commit broken code. Check:

• prd-progress.txt for errors
• Git working directory for uncommitted changes
• Fix the issue and run Ralph again

Tips for Best Results

1. Keep stories small - Each must complete in one context window
2. Order by dependencies - Schema before backend before UI
3. Verifiable criteria - "Typecheck passes" not "Works correctly"
4. Clean git state - Start with a clean working directory
5. Monitor progress - Check prd-progress.txt and git log regularly

Example Session

User: /ralph tasks/prd-task-priority.md

Ralph Skill:
- Converting PRD to prd.json...
- Created prd.json with 4 user stories
- Checking prerequisites...
- Creating branch ralph/task-priority...
- Starting Ralph loop...

[Ralph executes iterations...]

Iteration 1: US-001 - Add priority field to database ✓
Iteration 2: US-002 - Display priority indicator on task cards ✓
Iteration 3: US-003 - Add priority selector to task edit ✓
Iteration 4: US-004 - Filter tasks by priority ✓

<promise>COMPLETE</promise>

All stories completed! Check prd-progress.txt for details.