agentleFS
Flowerf19Flowerf19/agents-skills/code-reviewer/SKILL.md

code-reviewer

Skill3 starsChanged 4 days agoMITView on GitHub

Review code changes for bugs, security, compatibility, scope, and missing checks. Use when the user asks to review or audit a diff, pull request, or uncommitted changes, or when an approved implementation needs review. Review only; do not edit files.

182 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use git-doc-updater, git-precommit and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • u9401066Skill

    git-doc-updater

    Auto-check and update key documentation before Git commits to keep docs in sync with code. Triggers: docs, 文檔, 更新文檔, sync docs, release, 發布.

    Also in 5 projects that use this

  • u9401066Skill

    git-precommit

    Orchestrate pre-commit workflow including Memory Bank sync, README/CHANGELOG/ROADMAP updates. Triggers: GIT, gc, push, commit, 提交, 準備 commit, 要提交了, git commit, pre-commit, 推送.

    Also in 5 projects that use this

  • JeffallanSkill

    shopify-expert

    Builds and debugs Shopify themes (.liquid files, theme.json, sections), develops custom Shopify apps (shopify.app.toml, OAuth, webhooks), and implements Storefront API integrations for headless storefronts. Use when building or customizing Shopify themes, creating Hydrogen or custom React storefronts, developing Shopify apps, implementing checkout UI extensions or Shopify Functions, optimizing performance, or integrating third-party services. Invoke for Liquid templating, Storefront API, app development, checkout customization, Shopify Plus features, App Bridge, Polaris, or Shopify CLI workflows.

    Also in 5 projects that use this

  • JeffallanSkill

    spec-miner

    Reverse-engineering specialist that extracts specifications from existing codebases. Use when working with legacy or undocumented systems, inherited projects, or old codebases with no documentation. Invoke to map code dependencies, generate API documentation from source, identify undocumented business logic, figure out what code does, or create architecture documentation from implementation. Trigger phrases: reverse engineer, old codebase, no docs, no documentation, figure out how this works, inherited project, legacy analysis, code archaeology, undocumented features.

    Also in 6 projects that use this

  • +4 more from u9401066
  • +4 more from Jeffallan

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 155 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If no range was given, state which changes you selected.

    line 16
  2. Distinctive:

    Check current code against accepted requirements, not stale document claims.

    line 17
  3. Distinctive:

    If the target or requirement is unclear, ask or state the limits of the review.

    line 18
  4. Distinctive:

    Check the parts relevant to the task:

    line 22
  5. Distinctive:

    Look for an existing guard or test that disproves the concern.

    line 43
  6. Distinctive:

    Put findings first, most severe first.

    line 48
  7. Very specific:

    If there are no confirmed findings, say so.

    line 59
  8. Distinctive:

    Stop after the review for a review-only request.

    line 61

“Common” is measured: kept by at least 6 in 10 of the 155 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Code reviewer

Report confirmed problems in the actual change and explain their impact.

Review is read-only. Do not fix code, rewrite tests, or change plan status. A verdict does not grant permission to implement, commit, or merge. Report unavailable independent review as outstanding; self-review does not replace it.

Step 1: Set the review scope

  1. Identify the diff or commit range, requirements, approved scope, and check results.
  2. If no range was given, state which changes you selected. Keep unrelated existing edits outside the verdict.
  3. Check current code against accepted requirements, not stale document claims.
  4. If the target or requirement is unclear, ask or state the limits of the review. Do not invent a specification.

Step 2: Check the change

Check the parts relevant to the task:

  • Requirements: requested behavior is complete; proposed business rules have not become defaults without approval.
  • Correctness: inputs, errors, event order, and state changes work as required. Check concurrent updates when relevant. For heuristics, check their basis and counterexamples.
  • Security and data: input validation, permissions, tenant separation, secrets, parsing, and deletion remain safe.
  • Compatibility: callers, public imports, APIs, configuration, stored data, and migrations still match the accepted behavior.
  • Structure and scope: rules and external calls stay in their existing modules. Shared data definitions stay at their source. Do not demand new layers or refactors just for style.
  • Method comments: each new or changed method or function has a short comment directly above it when its behavior, intent, rule, or limit is not clear from the code, in the language's comment syntax (# in Python, // in TypeScript or Go). The comment says in plain words what the method does, and why it exists when the name does not say it. A one-line docstring in the same place counts when the file already uses docstrings for every method. A missing useful comment is a Minor finding. Do not flag a missing comment when the code is self-explanatory.
  • Object design: where the code is object-oriented, each class holds one clear concept, internal state stays hidden behind methods, and inheritance marks a real is-a relation. A class that mixes unrelated jobs is an Important finding.
  • File size: a changed file over 500 lines must hold one clear responsibility. Flag mixed responsibilities as Important. Flag growth past 500 lines with no stated reason as Minor.
  • Plain writing: documents, comments, and reports follow the "Plain writing" section in AGENTS.md (ISO 24495-1). Text the reader cannot understand on the first read is a Minor finding.
  • Integration: follow moved or changed code through its callers to the actual product flow. Check API fields, errors, event order, and data source tracking where affected.
  • Tests: assertions cover required behavior and important failures. Changed expectations have a valid reason. Compare reported results with the evidence. Run safe, relevant checks when useful and report what you ran.
  • Required checks: task prerequisites and checks before the next phase were respected. Skipped required checks remain outstanding. Mock tests do not prove real service or model behavior.

For a benchmark, also check where the measurements came from and whether the compared runs used compatible conditions.

Step 3: Validate each finding

  1. Find a case that can actually occur under the requirements.
  2. Show the evidence or reproduce the problem. Explain the effect on a caller or user.
  3. Look for an existing guard or test that disproves the concern.
  4. Report only findings that survive these checks. Put unresolved concerns under open questions. Do not fill a finding quota or invent failures outside the supported behavior.

Step 4: Return the review

Put findings first, most severe first. Write each finding in plain words: name the problem, then its impact, then the smallest fix. For each, include:

  • [Critical / Important / Minor] and path:line.
  • The problem, evidence, concrete impact, and smallest correction.
SeverityMeaning
CriticalSecurity failure, data loss, or failure of the main product flow
ImportantWrong required behavior, broken compatibility, or a meaningful missing check
MinorConfirmed issue that does not block acceptance; not a personal style preference

If there are no confirmed findings, say so. End with Approve, Approve with fixes, or Request changes. Include untested risks and open questions. The verdict describes the reviewed change, not permission to deploy it.

Stop after the review for a review-only request.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.