# Firzus Agent Skills Code Review > Code Review - Skill: `tomevault-io/firzus-agent-skills-code-review` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/firzus-agent-skills-code-review` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/firzus-agent-skills-code-review/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/firzus-agent-skills-code-review --- # Code Review Use this skill to review changed code with a senior-engineer stance: find real bugs and risks introduced by the change, avoid noise, and lead with actionable findings. ## Scope Review only the changed behavior unless the user explicitly asks for a broader audit. Do not treat pre-existing issues as findings unless the change makes them newly reachable or worse. For GitHub PRs, require `gh` to be installed and authenticated. Use `gh` for PR metadata, PR diffs, previous PR comments, and posting comments when requested. Use `git` for local diffs, branch history, `blame`, and full commit SHAs. Do not run builds, typechecks, linters, or test suites as part of review unless the user asks. CI should catch compiler, formatter, import, and basic lint failures. ## Workflow 1. Identify the review target: - GitHub PR: inspect it with `gh pr view` and `gh pr diff`. - Local branch or diff: compare against the requested base branch, merge base, or staged/unstaged changes. - If the target is unclear, ask one concise clarifying question. 2. Check whether review should proceed: - Skip closed or draft PRs unless the user explicitly asks. - Skip trivial generated or automated changes unless the user asks. - If you already reviewed the same PR and no relevant changes were added, say so instead of posting duplicate feedback. 3. Gather repository guidance: - Read relevant `AGENTS.md`, `CLAUDE.md`, Cursor rules, contribution docs, and review guidelines that apply to the modified files. - Use guidance only when it is specific enough to support a finding. 4. Review through five independent lenses. Use parallel subagents when the runtime supports them; otherwise perform separate passes yourself: - **Guideline compliance**: check changed code against applicable repo instructions. - **Obvious bugs**: look for correctness issues, broken edge cases, regression risks, resource leaks, bad error handling, and unsafe state. - **Historical context**: use `git blame`, recent commits, and file history to understand why nearby code exists. - **Prior PR context**: when available, inspect previous PRs or comments touching the same files with `gh`. - **Local contracts**: verify changes respect nearby comments, invariants, tests, API contracts, and documented assumptions in modified files. 5. Score each potential issue before reporting it: - `0`: false positive, pre-existing issue, or unsupported speculation. - `25`: possible issue, but not verified. - `50`: real issue, but minor or unlikely to matter in practice. - `75`: important and likely real, with strong evidence. - `100`: certain, reproducible, and directly caused by the change. For automated PR comments, report only issues scored `80` or higher. In chat, lead with high-confidence findings and put uncertain observations under open questions or residual risk. ## False Positives To Filter Do not report: - Pre-existing issues not introduced or worsened by the change. - Pedantic style nits, formatting, imports, or issues a linter/typechecker should catch. - General quality concerns not tied to a concrete risk. - Missing tests unless the absence creates a clear regression risk for changed behavior. - Intentional behavior changes that match the PR goal. - Guideline issues where the relevant instruction does not explicitly apply. - Problems on untouched lines, unless the changed lines make them newly fail. ## Output Follow the host agent's normal response conventions. Do not add vendor-specific signatures, attribution footers, or emoji unless the user asked for that format. Lead with findings, ordered by severity. For each finding include: - The affected file or code location. - The bug or risk in one concise sentence. - Why it matters and what condition triggers it. - The relevant evidence: changed code, guideline text, history, or prior PR context. - A concrete fix direction when it is not obvious. After findings, include open questions or assumptions. Keep summaries brief and secondary. If no issues are found, say that clearly and mention any review gaps such as unavailable PR metadata, missing `gh` authentication, or tests not run. When posting a GitHub PR comment with `gh pr comment`, keep it concise and use full commit SHAs in code links. GitHub links must use this shape: ```text https://github.com/owner/repo/blob//path/to/file.ext#L10-L15 ``` --- > Source: [Firzus/agent-skills](https://github.com/Firzus/agent-skills) — distributed by [TomeVault](https://tomevault.io).