`REREVIEW_PROMPT.txt`

36 lines
You are doing a CRITICAL POST-PATCH REVIEW. The file `CHANGES.md` in this directory lists patches that were just applied to this project. Your job: catch bugs the patcher introduced.

For each entry under `## Applied`, check the file:line cited. Look for:

1. **Regressions**: did the patch break working behavior? (e.g., changed default, broke a callsite that depended on the old fallback, removed a feature accidentally)
2. **Off-by-one or scope errors**: was the surgical patch wider than intended?
3. **Missing call-sites**: did the change cover all places it should have? (e.g., body-guard added at one route but missing at a sibling route with the same risk)
4. **Wrong defaults**: did the patch substitute a value that's worse than the original?
5. **Broken imports / missing requires**: was a helper added but never imported, or imported wrong?
6. **Production breakage**: is there an env var the patch now requires that won't be set in deployed environments?
7. **Lifecycle / cleanup correctness**: if `try/finally` was added, does it actually close the resource? Does it suppress real errors?
8. **Security regressions**: did a fix for one issue introduce another (e.g., escaping that breaks correctness)?

OUTPUT FORMAT (markdown, terse):

## Verdict
One of: SHIP IT / FIX BEFORE SHIP / ROLLBACK. Single line.

## Patches reviewed
Line per Applied item: `file:line — original problem — patch result — STATUS (good/concern/broken)`

## New issues introduced
Each: `file:line — issue — severity (P0/P1/P2)`. If none: `none observed`.

## Missed call-sites (same problem elsewhere)
Each: `file:line — same pattern still present`. If none: `none observed`.

## Recommendations
- Optional. Two bullets max. Skip if SHIP IT.

CONSTRAINTS:
- Read-only. Do not modify any files.
- Cite file:line for everything.
- Cap output at ~500 words.
- If CHANGES.md says "Applied: None", verdict is SHIP IT (nothing to review) and write a one-line note. Skip to end.