LLM Wiki

Appearance

Built 26/04/17 09:31commit 4c9ce40

Ralph Step-By-Step Guide ​

中文 | English

Summary ​

This source turns the Ralph pattern into a concrete operating recipe: a small scripts/ralph/ folder, a fresh-context bash loop, a PRD that tracks story status, a progress log that accumulates learnings, and commit-after-pass discipline on every completed story.

Source ​

Key Contributions ​

  • Provides a minimal file layout for loop-driven coding work: ralph.sh, prompt.md, prd.json, and progress.txt.
  • Makes stop conditions and story state machine explicit instead of leaving completion to chat interpretation.
  • Persists cross-iteration memory through git commits, the PRD, and appended learnings, not through one growing thread.
  • Encodes a concrete operator habit: implement one story, run checks, commit only if passing, then record what future iterations should remember.

Practical Implications ​

  • Fresh-context automation works better when task state is serialized into a few small, stable files that each new run can reload quickly.
  • Progress logs should capture reusable patterns and gotchas, not just a changelog of what happened.
  • Branch choice, acceptance criteria, and done-state signaling all belong in durable artifacts if the loop is expected to run unattended.