Code Reviewer Pro

What this skill does

• Starts with a PR overview (TL;DR, affected modules, risk areas) — lowers cognitive load
• Line-/diff-precise comments with a concrete fix, where possible as a committable suggestion block (apply in one click)
• Risk-based: reviews auth, queries, removed validations, money/data paths & incident areas more deeply, trivia quickly
• Finds bugs, edge cases and security holes (OWASP Top 10 / STRIDE)
• Checks dependencies/supply chain (CVEs, license, hallucinated packages) and AI-generated code (plausible-but-wrong)
• Surfaces missing & brittle tests and untested error paths
• Signal over noise: leaves mechanical nits to the linter, keeps nits short; high recall on bugs/security
• PR hygiene incl. size rule of thumb (≤ ~400 lines, else split), commit messages, scope
• Language presets (Python, JS/TS, Go, Java, SQL …), focus modes (security/tests/triage/architecture)
• States limits (static, not executed) and gives a clear verdict: LGTM / Request Changes

Description

Code-Reviewer Pro is a skill for AI code review and pull request review — it reviews your code line by line for bugs, security flaws, missing tests and maintainability. A deep, dev-focused code-review specialist for GitHub and GitLab. Works to the Google code-review standard (code health over perfection, blocking vs. nit, issue not person) and gives line-/diff-precise comments in Conventional Comments style (issue / suggestion / question / nitpick / praise), paste-ready for GitHub/GitLab. Review categories: correctness & bugs, edge cases, security (OWASP Top 10 / STRIDE, input validation, no secrets in code), tests (missing/brittle tests, untested error paths, coverage gaps), style & readability (naming, complexity, duplication), maintainability & structure, performance (N+1, hotspots), error handling & logging, API/interface design and PR hygiene (scope, commit messages, PR description). Language presets for Python, JS/TS, Go, Java/Kotlin and SQL with their typical pitfalls; other languages idiomatically. Optional focus modes: security, tests, quick triage (blockers only), architecture/design. Every review states its limits (static, not executed; only the shown excerpt) and ends with a clear verdict (LGTM / Request Changes). Surfaces vulnerabilities solely to fix them — no malicious code. For non-code reviews it points to the Review Architect, for deep statistics to the A/B Test Auditor.

Examples

What it does not do

• Writes no malicious code and builds no exploits — vulnerabilities only to fix them
• No sugar-coating and no tearing-down — always with a fix, blocking separated from nit
• Doesn't run the code (static review)
• Invents no secrets and passes none on
• Won't review non-code artifacts (use the Review Architect for that)

Compatibility & tech

Tested (internal)

3 scenarios

Recommended runtime

Strongest available model; most effective in agentic tools (Claude Code/Codex/Cursor) with repo access.

Modes

PR/diff review · File/function · Security focus · Test focus · Quick triage · Architecture/design

Inputs

Pull request / diff · File · Function/snippet · optional: language, purpose, test status, coding guidelines

Output format

PR overview (TL;DR/modules/risk) + summary & verdict (LGTM/Request Changes), line-level Conventional Comments with severity & fix (committable suggestions), test gaps, 'not checked/assumptions', next step

Subcategory

Code quality & review

License

Proprietär

Security profile

What you get

• code-reviewer-pro-1.4.0/4 files
  • .claude-plugin/marketplace.json
  • plugins/code-reviewer-pro/skills/code-reviewer-pro/4 files

    • SKILL.md
    • manifest.json
    • references/2 files

      • languages.md
      • review-checklist.md
  • .agents/skills/code-reviewer-pro/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
  • LICENSE.txt

Installation

Also works as a chat prompt

Note

A focused dev specialist (deeper than the Review Architect's code module). Best with language/framework, the code's purpose and test status. For reviews outside code → Review Architect.

Changelog

• v1.4.022.06.2026Platform-native output: now actively uses your platform's native formatting — tables, checklists, code blocks and clickable/button options, with a numbered fallback on non-interactive platforms. Plus internal metadata that prepares the upcoming agent builder.
• v1.3.018.06.2026Reworked to the AUTHORING v2 standard: realistic input/answer examples (incl. a calibration case), copy checklist(s) and internal evals; cross-references added.
• v1.2.018.06.2026Added references/languages.md: deeper detail reference via progressive disclosure; SKILL.md stays lean.
• v1.1.010.06.2026From web research (modern 2026 practices): PR overview/walkthrough first, committable suggestion blocks, risk-based depth, dependencies/supply chain (CVE/license/hallucinated packages) + AI-code check, signal-over-noise (mechanics to the linter, high recall on bugs/security), PR size rule of thumb (≤ ~400 lines).
• v1.0.010.06.2026Deep code-review specialist: line-precise Conventional Comments, OWASP/STRIDE security, test gaps, style/maintainability/performance, PR hygiene, language presets (Python/JS-TS/Go/Java/SQL), focus modes, LGTM/Request Changes verdict, limits stated; Google code-review better-not-perfect principle.

Frequently asked questions