Agent skill
marathon-ralph
Autonomous long-running development from specifications. Use when user wants to build an application from a spec file, run continuous development, or automate feature implementation. Triggers on phrases like "marathon this", "build from spec", "autonomous development", "keep coding until done".
Install this agent skill to your Project
npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/development/marathon-ralph
SKILL.md
Marathon Ralph - Autonomous Development
This skill enables autonomous, long-running development sessions that:
- Create Linear projects from specification files
- Work through issues systematically with verification
- Continue across multiple sessions via Stop hook
- Produce fully tested applications
Triggers
Activate this skill when the user:
- Wants to build an application from a specification
- Says "marathon this [spec]" or "build from [spec]"
- Asks for autonomous/continuous development
- Wants to "keep coding until done"
- Provides a spec file and wants it implemented
- Mentions "marathon development" or "marathon session"
- Wants to implement multiple features from a spec without stopping
Usage
Resume Existing Marathon
/marathon-ralph:run
Or naturally: "Continue the marathon" or "Keep going"
Start New Marathon
/marathon-ralph:run path/to/spec.md
Or naturally: "Marathon this spec.md until complete"
Check Progress
/marathon-ralph:status
Or naturally: "How's the marathon going?"
Cancel
/marathon-ralph:cancel
Or naturally: "Stop the marathon"
Prerequisites
User must have Linear MCP configured:
claude mcp add --transport http linear https://mcp.linear.app/mcp/mcp- Select Linear and authenticate via OAuth
How It Works
- Setup: Verifies Linear MCP connection is active
- Init: Creates Linear project and issues from the spec file
- Loop: For each issue in priority order:
- Verify (tests, lint, types) - ensures codebase is healthy
- Plan implementation - creates detailed implementation strategy
- Code the feature - implements according to plan
- Write tests - unit and integration tests
- Write E2E tests - for web projects only
- Commit and mark Done in Linear
- Continue: Stop hook automatically restarts for next issue until all complete
Example Workflows
Starting a New Project
User: "Marathon this todo-app-spec.md until it's done"
- Skill triggers on "marathon" + spec file reference
- Runs
/marathon-ralph:run --spec-file todo-app-spec.md - Creates Linear project with issues from spec
- Begins autonomous coding loop
Checking Progress
User: "How far along is the marathon?"
- Skill triggers on marathon progress query
- Runs
/marathon-ralph:status - Reports current phase, issue, and completion percentage
Stopping Early
User: "Stop the marathon, I need to change the spec"
- Skill triggers on stop/cancel request
- Runs
/marathon-ralph:cancel - Confirms with user and cleanly stops
- Linear project preserved for later
State Management
Marathon state is stored in .claude/marathon-ralph.json in the project root. This tracks:
- Current phase (setup, init, coding, complete)
- Linear project and issue information
- Progress statistics
- Timestamps for tracking
The Stop hook reads this state to determine whether to continue autonomous operation or allow exit.
Didn't find tool you were looking for?