ClaudeMap

code-review-gate-skill

Evidence-backed, auditable code review as a skill: static-first, four probes, gate decisions rendered from machine-readable JSON, every finding tied to a location, a contract and a severity — it never modifies the code under review, and ships separate English and Chinese skills (MIT)

Skills & Commandscode reviewstatic analysisquality gateclaude code skill

code-review-gate-skill is kyy-logs' open-source code-review skill (v0.2.4, released 2026-11-10; 447 stars as of writing) that replaces "this looks risky" AI reviews with a report you can audit: every finding points to a location, a contract or code path, and a severity, and every gate decision is rendered from machine-readable JSON. The review runs four probes — behavior-review (does the changed behavior match its actual contract and implementation path), test-coverage (are the tests meaningful, and is incremental coverage measured when data exists) and repo-guard (does the change violate repository or project rules) are required, and impact-analysis (who calls this, what contracts change, the performance and architecture blast radius) is optional. Static-first: it reads the diff, surrounding source, tests and project rules — no build, test run or coverage report required — and it never modifies the code under review. English and Chinese ship as separate skills (skills/code-review-gate and code-review-gate-zh), reports and gate output derive entirely from JSON, and v0.2.x severity is three levels: high, medium, low. The repo carries 116 offline tests across pipeline, NEWS, installation, bilingual, locale and source-freshness checks.

What it does

  • Four probes: behavior-review, test-coverage and repo-guard are required; impact-analysis (callers, contract changes, performance and architecture blast radius) is optional
  • Auditable output: every finding points to a location, a contract or code path and a severity; every gate decision is rendered from machine-readable JSON
  • Static-first: reads the diff, surrounding source, tests and project rules — no build, test run or coverage report required, and it never modifies the code under review
  • Bilingual by design: skills/code-review-gate (English) and code-review-gate-zh (Chinese); install one language per project
  • v0.2.4 stable (2026-11-10), backed by 116 offline tests across pipeline, NEWS, installation, bilingual, locale and source-freshness checks

When to use it

For teams that want AI review as a pre-merge gate: the report's findings can be checked one by one, gate decisions derive from JSON and plug into automation, and audit-hungry compliance settings get a trail to read.

Getting started

# inside your project (use --agent claude-code for Claude Code; it installs to .claude/skills/<skill>/):
npx -y skills add kyy-logs/code-review-gate-skill --agent codex --skill code-review-gate --yes

# Chinese skill:
npx -y skills add kyy-logs/code-review-gate-skill --agent codex --skill code-review-gate-zh --yes

# Codex installs to .agents/skills/<skill>/; the Python installers also accept --lang en|zh-CN

Things to know

Not a replacement for tests, CI or human review, and a clean report does not mean the change is correct. Language coverage is narrow by design: the default repo-guard workflow pattern-scans .java, .ts, .tsx and .vue files with a Java/Spring-leaning rule pack — selecting an extension does not establish complete framework coverage, and AI semantic checks depend on the model plus available contracts, source and tests. v0.2.x severity is three levels (high, medium, low): records from the earlier four-level version, including any critical value, are rejected by collect-metrics — re-run the review or migrate the records first. Requirements: Claude Code with Agent Skills or Codex, Python 3.9+ with PyYAML, Bash 3.2+, Git, awk, grep and sed, plus Node.js 22.20+ for the npx install entry. A controlled comparison against single-agent review has not been established. MIT. 447 stars as of writing (2026-11-10).

Repository

kyy-logs/code-review-gate-skill

Charted

Related on the map

  • An instruction-only skill that makes agents research before they ask, plan before they build, and verify before they call anything done.

    • workflow
    • verification
    • quality
    • skill
  • Guizang's agent skill turns your repo's recent updates into a finished promo video — real components, code-rendered animation, an original code-written soundtrack — no video editor involved.

    • agent-skill
    • video
    • promo
    • marketing
  • An agent skill that geolocates a photo with nothing readable in it — OSM geometry, elevation skylines and satellite tiles turn “where was this taken?” into a measured answer with an error radius.

    • agent-skill
    • geolocation
    • osint
    • openstreetmap