Code Review Skills
The evidence-code-review skill for Codex and Claude Code: evidence-based change review with MUST, SHOULD, BETTER and NITS levels
- 89 ★ on the repository
- Last commit
Minimal access
Why this level
- The skill reads the repository and writes review text, changing no files and widening no agent permissions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/code-review-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill reads the repository and writes review text, changing no files and widening no agent permissions.
This is third-party code. Review the repository files before installing.
What it does
The `evidence-code-review` skill runs a change review as a chain of evidence. The agent first sets the task scope and repository rules, then traces changed behaviour through every affected code path, including those where no line changed. Checks cover correctness, interfaces, design, security, reliability, tests and documentation, with depth set by the scale of the change. Every comment starts with an action level and a viewpoint. `MUST` is a demonstrated issue to fix before merge, `SHOULD` is a concrete threat, `BETTER` gives an optional alternative and `NITS` marks a minor point. A viewpoint such as `Functionality`, `Test`, `Simplicity` or `Document` explains the concern. The author draws on ideas from a Japanese book on code review and adapts them independently. The skill changes no files unless you ask separately.
Who it is for. Developers and team leads who hand the first review pass of a branch to Codex or Claude Code and want comments with a code location and evidence.
Good fit when
- A branch is ready to merge and you want a first review pass with a file and line on every finding
- A change touches contracts between modules and the impact beyond the edited lines matters
- A team settles on one comment format with severity levels and viewpoints
- Before handoff the tests and documentation for a change are checked together with the code
Not a fit when
- You want found issues fixed automatically, while the skill modifies nothing by default
- You need a guarantee that every defect is found, which the author says is not provided
- The repository has no rules or tests and the review must run with no human involved
Example request
/evidence-code-review Review the changes on this branch against main, focusing on correctness, regressions and tests. Do not modify files.Limitations
Results depend on the model, agent version, available tools, review scope and project information. A strong model helps but cannot guarantee completeness. The skill works together with tests and human review. Put project rules in the repository `AGENTS.md` or `CLAUDE.md`, because the installed package is overwritten on update. Large changes use many tokens. The skill grants the agent no extra permissions, and file access is set by the host agent.
How to disable. Delete the `evidence-code-review` folder from `~/.codex/skills` or `~/.claude/skills`. Removal details are in `docs/INSTALLATION.md`.
FAQ
How do I install the skill in Claude Code?
Run `git clone --depth 1 --branch v1.0.0 https://github.com/akkie76/code-review-skills.git`, then `mkdir -p ~/.claude/skills` and `cp -R dist/claude-code/evidence-code-review ~/.claude/skills/`. Start a new session.
What do MUST, SHOULD, BETTER and NITS mean?
`MUST` marks a demonstrated issue to resolve before merge. `SHOULD` flags a concrete threat that can be deferred by an explicit decision. `BETTER` gives an optional alternative with a clear benefit. `NITS` tags a minor non-blocking fix.
Does the skill edit code during a review?
No, the review runs without modifying files until you ask for changes separately. The README sample prompt includes the phrase "Do not modify files".
Where should project rules live?
In `AGENTS.md`, `CLAUDE.md` or a document they reference: version constraints, lifecycle and concurrency rules, architecture conventions and verification commands. Do not put them in the installed package.
How does it differ from code-review-skill?
This skill builds comments in a level, viewpoint, location, evidence shape and is pinned to a release with a documented output format.
Discussion
No questions or reviews yet.
Related
SkillsTwo skills for Codex and Claude Code: a read-only repository audit and an experiment loop with saved baselines.
Eleven skills to verify AI-written code: change blast radius, defect scan, review report, test cases and browser flow replay
A skill that reads a repository once and installs checks: code rules become gates that fail a commit or a CI run.
/pr-review sends a pull request through five independent reviewers, cross-checks findings and builds one report.
A CLI and MCP for testing agent skills: repeatable YAML evals, report comparison, regression detection and skilled-improve
An adversarial code-audit skill: Hunter finds bugs, Skeptic challenges them, Referee rules, and no source edit happens without your permission
README in short
The README covers release-tag installation for Codex and Claude Code, the comment format with levels, how to store project context and the limits. The evaluation in the repository uses a small synthetic fixture set.