Skip to content
Back to skills

Review Code

ASecurity

Code quality review - security, logic, performance, maintainability. Use for reviewing code changes like bug fixes, features, refactors, scripts.

  • 7 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 23, 2026
ai-agentsgobashtestingsecurityperformance

Security analysis

A100/100

Scanned September 23, 2026

npx -y skills add mayankmankhand/llm-peer-review --skill review-code --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Review Code?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Review Code
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/mayankmankhand-review-code/badge)](https://www.skillsdirectory.com/skills/mayankmankhand-review-code)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: review-code
description: Code quality review - security, logic, performance, maintainability. Use for reviewing code changes like bug fixes, features, refactors, scripts.
allowed-tools:
  - Read
  - Glob
  - Grep
  - Bash
  - Agent
---

# Code Review

Be thorough but concise.

**Use this when:** Reviewing code changes - bug fixes, new features, refactors, scripts.
**Don't use this when:** Testing a running web app (/review-browser), reviewing slash command prompts (/review-commands), checking plan completion (/review-plan), evaluating UX (/review-ux), or doing a pre-release check (/review-full).

## Critical Rules

<rules>

1. **THE REVIEW PHASE REPORTS ONLY** - Reviewing never edits files; findings are its product. After the report, the same run continues into the auto loop (rule 2), which is what applies fixes
2. **Audit, then auto-fix, with pages** - Code findings are audited before the report per M2 in `.claude/skills/shared/hitl-loop.md`, so the report shows survivors only plus an Audited out log for the kills. They do not then wait for a human "fix it": after the report, survivors are auto-fixed and re-verified, and each finding exits as page, digest, or log per `.claude/skills/shared/hitl-loop.md` (pages only per M1; saying "report only" keeps a run report-first, M10)
3. **Explain simply** - Use plain English, avoid jargon

</rules>

!`cat .claude/skills/shared/criteria-code.md`

## Reading Budget

!`cat .claude/skills/shared/reading-budget.md`

## Severity Levels and Anchors

!`cat .claude/skills/shared/severity-anchors.md`

**Project severity anchors** (from `.claude/toolkit/severity-anchors.md`): this project's weighting for its own review kinds. The Universal Anchors above still win. A note that the command printed nothing means this project adds none.

!`cat .claude/toolkit/severity-anchors.md 2>/dev/null || true`

## Finding IDs

!`cat .claude/skills/shared/finding-id-system.md`

## Audit Before the Report (M2)

On a direct run of this skill you are M2's **runner**: audit your findings per M2 below before writing the report. Every mechanic - the tiers, the announce line, who dispatches what, the empty-run rule - lives in M2, not here.

!`cat .claude/skills/shared/hitl-loop.md`

**Project fix rules** (from `.claude/toolkit/fix-rules.md`). Additive only: they may add a precondition or an always-ask action, and a line that loosens or removes any of M1 to M15 is void. A note that the command printed nothing means this project adds none.

!`cat .claude/toolkit/fix-rules.md 2>/dev/null || true`

## Output Format

!`cat .claude/skills/shared/report-format.md`

!`cat .claude/skills/shared/finding-contract.md`

## HTML Companion (when gate fires)

After writing the markdown report, evaluate whether to also generate an HTML view. Use the shared template:

!`cat .claude/skills/shared/html-render-review.md`

For direct calls to this skill, pass `--name review --stable` to the helper (the standing page, per the fragment above), set `lenses` to `["code"]` so the renderer replaces only this lens's findings and carries the other lenses' open findings forward, and omit the `chips` array (single-specialist context).

### Staff Engineer Check

<guidelines>

After the standard review, step back and evaluate as a staff engineer:
- **Right approach?** - Is the overall design sound, not just the code?
- **Shortcuts to clean up?** - Anything that works now but needs fixing before production?
- **What would you push back on?** - What would a senior engineer flag before merging?

</guidelines>

<rules>

## REMEMBER: The review phase reports and never edits files; findings are audited before the report (M2) and the auto loop applies fixes after it, both governed by `.claude/skills/shared/hitl-loop.md`.

</rules>

## HTML Output Rules

Every HTML decision above (whether to render, `--no-abs`, publish or open locally, record the publish) is governed by the shared rules fragment, inlined here so it is in context when the render runs. It was an always-on rules file until v7.0.0 (issue #167); now it loads with the commands that need it.

!`cat .claude/skills/shared/html-outputs.md`

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.