wf-review-code

Reviews a diff for correctness, tests, and conventions, producing a verdict with file:line findings.

Updated May 9, 2026
One-click install
npx skills add https://github.com/23min/aiwf --skill wf-review-code-23min
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: wf-review-code
Source: https://github.com/23min/aiwf/tree/main/internal/skills/embedded-rituals/plugins/wf-rituals/skills/wf-review-code
Command: npx skills add https://github.com/23min/aiwf --skill wf-review-code-23min

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code review quality depends heavily on who reviews and how they are briefed: self-review misses the author's own blind spots, and a bare "review this" prompt yields shallow passes. This Skill provides a structured, adversarial review checklist that classifies findings by kind and urgency so nothing actionable is silently dropped. ## Core Features & Use Cases - Structured diff walkthrough: Checks correctness, edge cases, error handling, project constraints, tests (including a manual branch-coverage audit), conventions, and documentation hygiene. - Two-axis finding classification: Every finding is labeled defect vs. judgment and blocking vs. track-for-later vs. non-issue, each with a file:line reference and a disposition (pin with a test, record a decision, or decline). - Verdict output: Produces approve, request-changes, or questions with a standardized Markdown report format. - Use Case: Before merging a feature branch, dispatch a fresh agent with this Skill and an adversarial brief listing the change's load-bearing claims; the reviewer verifies each claim by measurement and returns blocking findings with locations. ## Quick Start Ask the assistant to review the current branch diff against the stated goal using the wf-review-code checklist and return a verdict with classified findings.

Frequently Asked Questions about wf-review-code

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a pull request with an AI assistant?

Hand the reviewer the diff plus the stated goal or acceptance criteria, then walk each changed file for correctness, tests, and conventions. This Skill structures that pass and outputs a verdict of approve, request-changes, or questions with file:line findings.

How to classify code review findings as blocking or non-blocking?

Classify each finding on two axes: kind (defect verifiable against a test or spec, versus judgment as a design preference) and urgency (blocking before merge, track for later, or non-issue). Blocking defects must be fixed and pinned with a check.

Why is self-review of your own code unreliable?

The same blind spots that produced a defect shape the author's review, so self-review reliably misses what the author did not know to look for. A fresh agent given an adversarial brief and instructed to verify by measurement finds more issues.

Does this code review skill run automated coverage tools?

No. Branch coverage is an agent-performed manual walk of every reachable conditional branch in the diff, because typical mechanical coverage gates are statement-level. The reviewer must perform the walk rather than rely on a tool.

When should I use a codebase health check instead of a diff review?

Use a per-diff review for the change itself; use a codebase-health rubric for structural properties a single diff cannot reach, such as module boundaries and observability. On large or boundary-introducing diffs, run both.