createagent-loop-prompt

name

create:agent-loop-prompt

description

Generate a agent-loop-prompt.md file at the project root that defines an agent loop — the repeatable task cycle an autonomous agent follows: study the plan, pick the next task, implement, validate, update progress, and commit. Also known as "Ralph" — trigger when the user says "Ralph", "the Ralph process", "set up Ralph", "create an agent loop", "generate a agent-loop-prompt.md", "set up the agent loop", or "/create:agent-loop-prompt".

Create Agent Loop Prompt

Generate a agent-loop-prompt.md file at the project root that drives an agent loop through an implementation plan one task at a time. The generated agent-loop-prompt.md is consumed by agent-loop.sh, which feeds it to Claude Code on each iteration.

Process

1. Collect the plan path

Ask the user for the path to the implementation plan (e.g. plans/feature.md). Validate the file exists before continuing. If it does not exist, inform the user and ask again.

2. Check for existing agent-loop-prompt.md

If agent-loop-prompt.md already exists at the project root, warn the user and ask whether to overwrite it.

3. Initialize progress.txt

Create an empty progress.txt at the project root if one does not already exist.

4. Write agent-loop-prompt.md

Write the following to agent-loop-prompt.md at the project root, replacing <plan-path> with the user-provided path:

0. Study the `src/` directory to understand the codebase.
1. Study @<plan-path>
2. Review @progress.txt
3. Find the highest-priority incomplete task and implement it.
4. Run type checks, tests, linting, and formatting (if applicable). Do NOT commit if any fail. Fix issues first.
5. Update the plan — mark the task complete.
6. Append your progress to progress.txt:
   - Concise summary of what was implemented
   - Files changed
   - (Optional) learnings, patterns, gotchas
7. Commit your changes.

ONLY WORK ON A SINGLE TASK.

If, while implementing, you notice that all work is complete, output <promise>COMPLETE</promise>