Agent code review
Skill Amey-Thakur/AI-SKILLS/skills/ai-coding-tools/agent-code-review
Use an agent to review a diff productively, with the right scope, framing, and scepticism about its findings. Use when reviewing your own changes before requesting human review.From its SKILL.md
npx -y skills add Amey-Thakur/AI-SKILLS --skill agent-code-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
1.8 KB, as published. Nobody here has run it
Agent code review
Agents catch a real class of defects consistently and produce confident false positives just as consistently. The value comes from scoping the review and verifying findings rather than accepting the list.
Method
- Review the diff, not the repository. A focused review of what changed produces specific findings; a whole-codebase review produces generic advice (see code-review).
- Give the intent of the change. What it is meant to do, so the review can judge whether it does (see agent-context-setup).
- Ask for one dimension at a time. Correctness, then security, then performance. A single pass over everything covers each shallowly (see agent-quality-council).
- Require evidence for each finding. The specific line and a concrete failure scenario, since findings without them are usually pattern-matching (see agent-generate-and-verify).
- Verify before acting. Confirm the issue is real in your codebase, because plausible-sounding findings about code that does not behave that way are common.
- Use it before human review, not instead. It clears mechanical issues so human reviewers spend attention on design.
- Feed recurring findings into the instruction file. A rule the agent keeps flagging belongs written down (see agent-instruction-files).
Boundaries
Agent review supplements human review and cannot judge whether the change is the right thing to build. It lacks the system context a reviewer holds. False positives waste time and erode trust, which is why verification is part of the workflow.