review

Reviews diffs, branches, or PRs with verified findings ranked by severity.

9|1|Updated Aug 28, 2026
One-click install
npx skills add https://github.com/RisorseArtificiali/skills --skill review-risorseartificiali
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: review
Source: https://github.com/RisorseArtificiali/skills/tree/main/skills/review
Command: npx skills add https://github.com/RisorseArtificiali/skills --skill review-risorseartificiali

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code reviews often produce unverified or low-value findings that waste the author's time investigating false positives, while critical bugs get buried under nitpicks. This Skill runs an on-demand leveled review where every reported finding is verified before it reaches you. ## Core Features & Use Cases - Verified Findings Only: Every finding is re-checked at its cited file and line; unverifiable claims are dropped, and at high level a skeptic subagent must reproduce each issue. - Leveled Review Depth: Choose low/medium for a few high-confidence findings or high for broader coverage with correctness and quality lenses. - Flexible Targets: Review uncommitted changes, a branch, a PR number, a commit range, or specific paths, with optional --fix and --comment modes. - Use Case: After finishing a feature branch, ask for a high-level review of the diff; you receive a ranked list of confirmed bugs with reproduction evidence, or a plain statement that nothing was found. ## Quick Start Ask the agent to review my current diff at medium level and report only verified findings ranked by severity.

Frequently Asked Questions about review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a git diff or pull request with an AI agent?▼

Point the review at your target: uncommitted changes, a branch, a PR number, or specific paths. The skill dispatches a fresh reviewer subagent, verifies each finding at its cited location, and reports results ranked most-severe first.

What is the difference between low, medium, and high code review levels?▼

Low and medium use one reviewer and return few high-confidence findings verified by re-reading cited code. High uses two reviewers (correctness and quality lenses) plus a skeptic subagent per finding that must reproduce each issue or label it PLAUSIBLE.

When should I use adversarial-code-review instead of a quick review?▼

Use the heavyweight gate when a branch is about to merge, the change touches auth, data migrations, concurrency, or public APIs, or repeated reviews keep surfacing confirmed findings. The quick review is for on-demand, mid-work checks.

Can the review automatically fix the findings it reports?▼

Yes, the --fix option applies confirmed findings to the working tree with the smallest resolving change, then re-runs tests covering the touched code. It reports what was fixed versus skipped and never commits unless asked.

Does the review post comments to GitHub pull requests automatically?▼

No. The --comment option drafts inline PR comments and shows you the exact text first. Nothing is posted to any external surface without your explicit approval of the content.