review
Read-only protocol for reviewing code, documents, pull requests, or actions taken by the AI, using structured emoji indicators.
Read-Only Review Protocol
When requested to review, check, inspect, or look at files, code, changes, or pull requests, follow this protocol.
Critical Rules
- READ-ONLY: Do NOT make any edits to the target files, directories, or databases unless explicitly asked to do so.
- NO MODIFICATION: Do NOT run commands that edit files, rename directories, or delete code during the review.
When reviewing something you did: If you make a mistake, go idle, hallucinate, or screw up a tool call, look at your conversation history to see exactly what you did, figure out what happened, and accurately explain it to the user without making anything up.
- ANTI-HALLUCINATION DURING REVIEW: When reviewing actions you or another AI took, hallucinating the explanation is FAR WORSE than a normal hallucination. The user is asking you to review precisely because they need an accurate account of what happened. If you fabricate an explanation, you are actively destroying the user's ability to trust any AI in this conversation. Before presenting your explanation: (a) mechanically trace the actual tool calls and their outputs step-by-step in the transcript, (b) verify every factual claim against actual evidence (file contents, timestamps, command outputs), (c) if you are uncertain about any part of your explanation, say so explicitly rather than filling gaps with plausible-sounding guesses, (d) never attribute actions to external causes (other AIs, bugs, scripts) without concrete proof.
Formatting Guidelines
When presenting your review:
- Enumerate the files, functions, or changes being reviewed.
- Link to every file using GitHub markdown links (e.g.
[filename](file:///absolute/path/to/file)). - Prefix each reviewed item or file with one of the following emojis to concisely summarize the status of that item:
- ✅ Check Mark: Perfectly OK. No issues, regressions, or changes suggested.
- ⚠️ Warning: The item is functional/mostly fine, but has questions, potential optimizations, or edge cases that need to be pointed out.
- 🛑 Stop Sign: A definite issue or bug is present that must be fixed before proceeding.
- Provide a clear explanation explaining the mechanical reasoning behind your rating.
- Provide a summary block at the very end of the review:
- If everything is OK: "Review Summary: Everything looks perfect."
- If there are warnings/issues: Summarize them concisely (e.g., "Review Summary: Mostly OK, but review [item X]" or "Review Summary: Critical issues found in [item Y]").