Skip to content

prompt.md

Latest commit

 

History

History
20 lines (18 loc) · 1.26 KB

prompt.md

File metadata and controls

20 lines (18 loc) · 1.26 KB
Raw

Ralph prompt — Lifeline

You are an autonomous coding agent improving Lifeline. Do exactly one story per loop.

  1. Study AGENT.md to understand the project, the guardrails, and the invariants.
  2. Read prd.json. Pick the single highest-priority story (lowest priority number) whose "passes": false. Work on that ONE story only — do not start others.
  3. Before implementing: use sub-agents to search the codebase so you don't fill your own context ("don't assume; verify by searching"). Confirm the story isn't already done.
  4. Implement the story. Keep the diff additive and localized. Obey every guardrail in AGENT.md: preserve all UI element ids and the /ask contract, keep accessibility + reduced-motion, do not edit the hardened core (triage.py / real_run.py verify/PROTO) unless the story says so.
  5. Validate:
    • ./check.sh must pass (never weaken existing tests).
    • python3 -m py_compile lifeline/*.py.
    • For UI stories, confirm the rendered result in the preview.
  6. Set this story's "passes": true in prd.json and commit with a clear message.
  7. If, after this, EVERY story in prd.json has "passes": true, reply with the exact sentinel: <promise>complete</promise> Otherwise, end so the loop runs the next iteration.