Code Integrity Audit

Under long sessions or complex tasks, coding agents often degrade silently: leaving stub implementations, swallowing exceptions, faking passing tests, or hardcoding mock returns.

This skill audits git diff with adversarial objectivity before changes are accepted.

Evidence Before Claims

No task is complete on verbal assurance alone.

• Run the actual test, lint, and build commands (npm test, pytest, cargo test, etc.).
• Inspect fresh raw terminal output.
• Never report success without showing command execution evidence.

The 4-Point Diff Scan

Inspect the newly added lines (the green lines in git diff <base>...HEAD):