Skip to content
Back to search

code-review-skills

Evidence-based code review skills for Codex and Claude Code

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/akkie76--code-review-skills/badge.svg)](/listings/akkie76--code-review-skills)

HTML

<a href="/listings/akkie76--code-review-skills"><img src="/api/listings/akkie76--code-review-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 125 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # Code Review Skills
  2. [日本語](README.ja.md)
  3. An evidence-driven code-review workflow for Codex and Claude Code.
  4. This Skill is based on ideas from [*コードレビューの教科書*](https://gihyo.jp/book/2026/978-4-297-15768-5)
  5. and independently adapts them for AI coding agents. It does not reproduce or
  6. replace the book.
  7. **Version:** `v1.0.0` · **License:** MIT
  8. The documented review-output format is stable from 1.0.0. Review decisions are
  9. still model- and context-dependent: this Skill does not guarantee that every
  10. defect will be found or that repeated reviews will produce identical findings.
  11. ## Get started
  12. Clone the release tag so the installed package has a known version:
  13. ```sh
  14. git clone --depth 1 --branch v1.0.0 https://github.com/akkie76/code-review-skills.git
  15. cd code-review-skills
  16. ```
  17. ### Codex
  18. ```sh
  19. mkdir -p ~/.codex/skills
  20. cp -R dist/codex/evidence-code-review ~/.codex/skills/
  21. ```
  22. Then start a new Codex task and ask, for example:
  23. > Use the `evidence-code-review` skill to review the changes on this branch against `main`. Focus on correctness, regressions, and tests. Do not modify files.
  24. ### Claude Code
  25. ```sh
  26. mkdir -p ~/.claude/skills
  27. cp -R dist/claude-code/evidence-code-review ~/.claude/skills/
  28. ```
  29. Start a new Claude Code session and invoke the Skill, for example:
  30. ```text
  31. /evidence-code-review Review the changes on this branch against main. Focus on correctness, regressions, and tests. Do not modify files.
  32. ```
  33. For project-only installation, Windows instructions, verification, updates,
  34. removal, and troubleshooting, see the [installation guide](docs/INSTALLATION.md).
  35. ## What it reviews
  36. - Establishes the requested scope and repository-specific rules.
  37. - Traces changed behavior through affected code paths, not only changed lines.
  38. - Checks correctness, interfaces, design, security, reliability, tests, and
  39. documentation according to risk.
  40. - Reviews code without modifying it unless you separately ask for changes.
  41. ## What a review looks like
  42. Actionable comments start with an action level and viewpoint:
  43. - `MUST`: a demonstrated issue that must be resolved before merge.
  44. - `SHOULD`: a concrete risk that normally should be addressed, but may be
  45. deferred by an explicit decision.
  46. - `BETTER`: an optional alternative with a specific benefit.
  47. - `NITS`: a minor, non-blocking correction.
  48. The viewpoint explains the concern, such as `Functionality`, `Test`,
  49. `Simplicity`, or `Document`. A finding should give its location and concrete
  50. evidence, explain when it occurs and its impact, and suggest a proportionate
  51. direction.
  52. ```text
  53. MUST(Functionality): Advance the page before requesting the next result set
  54. Location: `src/client.ts:42`
  55. When the API returns a full page, this loop requests the same page again because
  56. `page` never changes. That duplicates results and increases request volume;
  57. advance the page before the next request.
  58. ```
  59. This is an illustrative example. See the [review comment convention](docs/REVIEW_COMMENTS.md)
  60. for the full action-level, viewpoint, evidence, and communication rules.
  61. ## Project-specific context
  62. Languages, frameworks, architecture, and test commands differ by repository.
  63. Put the information needed for an accurate review in that repository's
  64. `AGENTS.md`, `CLAUDE.md`, or a project document referenced by those files. Useful
  65. context includes version constraints, lifecycle or concurrency rules,
  66. architecture conventions, and verification commands. Do not store project
  67. information in the installed package; updates may replace it.
  68. ## Limits and expectations
  69. - Results depend on the model, agent version, available tools, review scope, and
  70. project information. A stronger model may help, but does not guarantee a
  71. correct or complete review.
  72. - Use this Skill alongside tests and human review; it is not a substitute for
  73. either and cannot guarantee zero missed defects or false positives.
  74. - The Skill does not grant additional permissions. File access and approval
  75. behavior are controlled by the host agent and its settings.
  76. - Reviewing large changes may consume substantial model tokens.
  77. - The [latest evaluation report](tests/results/2026-10-08-issue-24-v2-evaluation.md)
  78. describes a small synthetic fixture campaign, not production review accuracy;
  79. see the [evaluation guide](tests/README.md) for context and other reports.
  80. ## Updates and support
  81. - The [versioning policy](docs/VERSIONING.md) describes the stable output
  82. contract and how changes map to versions.
  83. - See the [installation guide](docs/INSTALLATION.md) for updating or removing
  84. an installation.
  85. - Evidence-backed bug and accuracy reports are welcome under the
  86. [support policy](SUPPORT.md). Pull requests require prior agreement.
  87. - Report suspected vulnerabilities privately as described in
  88. [SECURITY.md](SECURITY.md).
  89. ## License
  90. The original material in this repository is licensed under the
  91. [MIT License](LICENSE).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.