code-review-action: Skill for Claude Code

Code Review

Provide a code review for the selected target.

Dependencies

• code-simplify — supply the complete simplification rubric for that review lens.
• security-audit — supply the complete security rubric and attack-class catalogue for that review lens.

/code-review [low|medium|high|xhigh|max|ultra] [fix] [comment] [<target>]

Arguments

Arguments may appear in any order, and the -- prefix on a mode is optional: fix and --fix mean the same thing, as do comment and --comment. /code-review medium fix is a valid invocation.

Read the tokens as follows:

• A token matching an effort level sets the effort.
• A token matching fix or comment, with or without a leading --, enables that mode.
• Any remaining token is the <target>. To review a ref whose name collides with a reserved word, pass it as a fuller ref such as refs/heads/fix.

Ask for anything the invocation did not state, before doing anything else. The invocation states two things independently, and each unstated one is a question:

• Effort — stated by an effort token. Otherwise ask which level to run.
• Modes — stated by the presence of any mode token, which turns the named modes on and leaves the unnamed ones off. When no mode token appears at all, ask which modes to enable.

So a bare /code-review asks both, /code-review high asks only about modes, /code-review fix asks only about effort, and /code-review high fix asks nothing and runs. A <target> never triggers a question; it has a defined fallback when absent.

When asking for effort, list every accepted value explicitly and describe its actual coverage and validation depth:

• low — Run one Rules lens and one Bugs lens with inline validation for a quick review of small, low-risk changes.
• medium — Run one Rules, Bugs, Contracts & comments, History, Simplification, and Security lens for a routine contextual review, with inline validation.
• high — Run one Rules, two independent Bugs, one Contracts & comments, one History, one Prior PRs, one Simplification, and one Security lens, then have one standard validator try to refute every finding.
• xhigh — Run two independent Rules and Bugs lenses plus one Contracts & comments, History, Prior PRs, Simplification, and Security lens, then use two standard validators per finding with majority rule.
• max — Run two Rules, three Bugs, two Contracts & comments, two History, two Prior PRs, two Simplification, and two Security lenses in one pass, then use three deep refuters per finding with majority rule.
• ultra — Repeat the max cohort until two consecutive rounds find nothing new, using the same three deep refuters per finding.