Review the complete diff as a skeptical senior reviewer before handing work back. Use this after finishing any code change and before declaring it done, before committing a substantial diff or opening a pull request, and whenever the user says "review", "check this", or "is this ready". Includes a lightweight security sweep of the changed lines. Make this the last step of every non-trivial coding task.
SkillScore82Maintained
Automatic
◆Coding pack
Last verified 8/15/2026
What it changes
Passed 17 of 21 test cases with the skill, and 16 of 21 without it.
Without skill77%
With skill82%
Checks it passed that the model alone missed
The review instructions mandate preserving unrelated work without discarding or overwriting it.
The response places critical security vulnerabilities at the top of the findings list before any minor style or documentation issues.
2 fixed16 passed either way3 still failing1 broke
Pass-rate lift
+5 pp
Tokens
+35%
Agent turns
0%
Response time
+15%
Model gemini-3.6-flashMeasured 8/15/2026v218 of 22 cases passed with skill · 21 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 2 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 2
Prompt
I just pushed some security and state-handling refactoring across several modules. I need you to do a thorough code audit of all my local changes without editing or changing any code. Please trace all mutated inputs, inspect our trust boundaries for potential vulnerabilities, check error handling and resource cleanup, and verify test coverage. Provide me with a prioritized list of actionable defects ranked from highest risk to lowest, specifying where each issue occurs, what triggers it, its potential impact, and suggested fixes, followed by any unverified assumptions.
What was checked, without and with the skill
Checked
Without skill
With skill
The response places critical security vulnerabilities at the top of the findings list before any minor style or documentation issues.
Missed
Passed
case-01 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md22 linessingle file
Self Review
Read the actual change as a skeptical maintainer, not from memory.
Select the Mode
Author mode: Review work produced within the current approved implementation. Fix in-scope findings, rerun affected checks, and review the resulting diff again.
Review-only mode: Report findings and evidence without modifying files, formatting, snapshots, dependencies, git state, or external systems. Implement fixes only when separately requested.
Review Process
Inspect repository instructions, status, the complete diff, untracked files, and the original request. Preserve unrelated work.
Trace each changed input through callers, state transitions, outputs, and failure paths. Check empty, boundary, large, Unicode, concurrent, retry, and partial-failure cases where relevant.
Follow changed trust boundaries beyond literal changed lines. Inspect authorization, secrets, injection, filesystem and process execution, unsafe rendering, deserialization, dependency risk, and data exposure.
Check errors and cleanup: failures must surface; resources, transactions, locks, and temporary files must be released.
Check scope and consistency: no debug leftovers, dead code, accidental generated files, unsupported API changes, or unrelated formatting churn.
Check tests: new behavior has meaningful coverage, assertions can fail, and boundary behavior is exercised.
Validate each potential finding against callers, guards, tests, and framework guarantees. Omit speculation.
Report
Lead with independently actionable findings ordered by severity, each with a precise location, trigger, impact, evidence, and remediation direction. Then list material assumptions and verification gaps. If no finding survives validation, say so explicitly.