AI Code Change Auditor

The problem

AI agents often claim a task is complete while leaving behind TODOs, omitting necessary tests, or making unrelated changes to your codebase. Reviewing these "black box" patches manually is time consuming and prone to missing subtle scope drift or false completions.

What it does

• Detects false completions where the AI claims a fix but the diff lacks the actual implementation.
• Identifies scope drift and "overreach" where the agent modified files unrelated to the original ticket.
• Flagging missing validation, such as skipped edge cases, missing migration scripts, or tests that don't actually assert the requested behavior.
• Categorizes patches into actionable states like PASS, NEEDS_FIX, or PRODUCT_REVIEW_REQUIRED.
• Generates a structured Fix Verification Receipt to document evidence, reviewed files, and outstanding risks.

Why this beats prompting it yourself

General prompts often miss "hallucinated" progress where an agent says a test passed that wasn't actually run. This skill enforces a rigorous audit procedure that specifically looks for agentic failure modes like happy-path-only fixes and placeholder code that standard reviews often overlook.

Use cases

• Reviewing a PR generated by an automated coding agent before merging to main.
• Auditing a complex bug fix to ensure the agent didn't break adjacent call sites.
• Verifying that a refactor stayed within the requested architectural boundaries.
• Generating a verification receipt for compliance or team handoff after AI-assisted development.

Known limitations

This is a change auditor, not a deep security or performance profiler. It cannot verify UI behavior without provided screenshots or logs.