verify-fix

/verify-fix — Bug Fix Verification through Reproduction

Steps

1. Fetch the issue — get the issue details and understand the expected behavior.

2. Check the fix plan — Read .ai/fix-<issue_number>.md if it exists. This tells you:

   • Whether the dev test passed or failed during plan-fix
   • What changes were made and in which files
   • Any known issues or limitations If the dev test status is FAILED, pay special attention to the "Known Issues" section — the fix may be incomplete.

3. Build and patch the product — Follow the patching instructions in CLAUDE.md:

   • Build the changed module(s): mvn clean install -Dmaven.test.skip=true
   • Extract a fresh product pack from the zip
   • Apply JAR patches to repository/components/patches/patch9999/
   • Apply any template or WAR patches
   • Start the server

4. Verify at runtime — You MUST actually test the fix against a running product. Checking code diffs, grepping compiled bundles, or confirming "the build succeeded" is NOT verification. You must observe the correct behavior at runtime.

   For frontend bugs: Use Playwright — follow the "Interacting with the Frontend (Playwright)" section in CLAUDE.md.

   • If a reproduction script exists from the reproduce step (.ai/reproduce-<issue_number>.mjs), run it — the bug behavior should no longer occur.
   • If no script exists, write one following the Playwright guidelines in CLAUDE.md.
   • Save verification screenshots to .ai/screenshots-<issue_number>/verify/.
   • The screenshots must show the correct behavior (e.g., an element that was hidden is now visible).

   For backend bugs: Use curl or REST API calls against the running server.

   • After deploying APIs/products, immediately check the server log for errors before attempting invocation — follow the "Troubleshooting / Log Analysis" section in CLAUDE.md for the exact command and filtering rules.
   • If deployment errors are found, report them immediately — do not proceed to invocation.
   • If deployment is clean, proceed with invocation and compare results.

   What does NOT count as verification:

   • ❌ "The compiled bundle contains the fix logic" — grepping minified JS is not verification
   • ❌ "The build succeeded" — a successful build only means the code compiles
   • ❌ "The admin portal loads (HTTP 200)" — this only means the server is running
   • ✅ Playwright screenshots showing the correct UI behavior after the fix
   • ✅ curl response showing the correct HTTP status/body after the fix
   • ✅ Server logs showing no errors where there were errors before

5. Report — Create .ai/verify-<issue_number>.md:

# Fix Verification Report

**Issue**: <url>
**Verdict**: FIXED | NOT FIXED
**Verification method**: Playwright / curl / server logs

## Reproduction Steps Executed
<what you did — must be runtime steps, not code analysis>

## Result
<what happened — describe observed runtime behavior>

## Evidence
<screenshots in .ai/screenshots-<issue_number>/verify/, curl output, or server log excerpts — MUST be from runtime>

Important Rules

• Server startup: The start command and the log polling loop MUST be in the same Bash tool call with timeout: 200000. Do not split them into separate calls.
• Playwright best practices: Wait for elements rather than using fixed sleeps (page.waitForSelector(), page.locator().waitFor()). Log assertions clearly — print expected vs actual so the output is useful in artifacts.