melvynx/ralph-loop

<objective> Set up the Ralph autonomous coding loop in any project. Ralph runs AI agents in a loop, picking tasks from a PRD, implementing one at a time, committing after each, and accumulating learnings until all tasks are complete.

This skill ONLY sets up Ralph - you run the commands yourself. </objective>

<quick_start> Setup Ralph interactively (recommended):

/setup-ralph -i

Setup for specific feature:

/setup-ralph -f 01-add-authentication

What this does:

1. Creates .claude/ralph/ structure in your project
2. Runs setup script to create all Ralph files
3. (If -i): Brainstorms PRD with you interactively
4. Transforms PRD into user stories (prd.json)
5. Shows you the command to run Ralph (you run it yourself)

After setup, you run:

bun run .claude/ralph/ralph.sh -f <feature-name>

</quick_start>

<critical_rule> 🛑 NEVER run ralph.sh or any execution commands automatically 🛑 NEVER execute the loop - only set up files and show instructions ✅ ALWAYS let the user copy and run commands themselves ✅ ALWAYS end by showing the exact command to run </critical_rule>

<when_to_use> Use this skill when:

• Starting a new feature that can be broken into small stories
• Setting up Ralph in a new project
• Creating a new feature PRD interactively

Don't use for:

• Simple single-file changes
• Exploratory work without clear requirements
• Major refactors without acceptance criteria </when_to_use>

<parameters> | Flag | Description | |------|-------------| | `<project-path>` | Path to the project (defaults to current directory) | | `-i, --interactive` | Interactive mode: brainstorm PRD with AI assistance | | `-f, --feature <name>` | Feature folder name (e.g., `01-add-auth`) |

Examples:

/setup-ralph /path/to/project -i              # Interactive PRD creation
/setup-ralph . -f 01-add-auth                 # Setup for specific feature
/setup-ralph -i -f 02-user-dashboard          # Interactive with specific name

</parameters>

<state_variables> | Variable | Type | Description | |----------|------|-------------| | {project_path} | string | Absolute path to target project | | {ralph_dir} | string | Path to .claude/ralph in project | | {feature_name} | string | Feature folder name (e.g., 01-add-auth) | | {feature_dir} | string | Path to task folder | | {interactive_mode} | boolean | Whether to brainstorm PRD interactively | | {prd_content} | string | PRD markdown content | | {user_stories} | array | User stories extracted from PRD | | {branch_name} | string | Git branch for the feature | </state_variables>

<entry_point> Load steps/step-00-init.md </entry_point>

<step_files> | Step | File | Purpose | |------|------|---------| | 00 | step-00-init.md | Parse flags, run setup script, create structure | | 01 | step-01-interactive-prd.md | Interactive PRD brainstorming and creation | | 02 | step-02-create-stories.md | Transform PRD into user stories (prd.json) | | 03 | step-03-finish.md | Show run command (user runs it themselves) | </step_files>

<scripts> | Script | Purpose | |--------|---------| | `scripts/setup.sh` | Creates all Ralph files in the project | </scripts>

<execution_rules>

1. Progressive Loading: Load one step at a time
2. Script Execution: Use scripts/setup.sh to create files atomically
3. Interactive Mode: If -i flag, run brainstorming conversation
4. State Persistence: Track progress in feature_dir/progress.txt
5. Resume Support: Detect existing PRD.md and resume from there
6. NEVER RUN RALPH: Only setup and show commands - user runs them </execution_rules>

<success_criteria> ✅ Ralph structure created at {project_path}/.claude/ralph ✅ Feature folder created with PRD.md, prd.json, progress.txt ✅ User stories properly formatted in prd.json ✅ Clear run command provided to user (they run it themselves) </success_criteria>