Note

This README was generated by SKILL, get the ZH version from here.
This skill's implementation was fully agent-generated; the developer only tuned its input/output.

---

AST-DRIVEN CODE REVIEWS THAT SKIP THE NOISE!

---

A Claude Code skill that generates optimization reports for Go, Python, and JavaScript/TypeScript via AST analysis, entropy-based secret detection, and a no-op gate

Table of Contents

• Features
• Architecture
• License

Features

/code-reviewer [PROJECT_PATH] [OUTPUT_FILE] · Documentation

• AST-Driven Multi-Language Analysis — Go is parsed via a go/ast helper, Python via the built-in ast module, and JS/TS via a lightweight brace-based structural scan for functions and nesting depth; each degrades gracefully to string scanning when its toolchain is unavailable.
• Entropy-Based Secret Detection — Beyond keyword matching, suspicious strings are scored with Shannon entropy and filtered against UUID/MD5/SHA1/SHA256/MIME-type patterns to cut false positives in credential detection.
• No-Op Gate — When both issue counts and actionable recommendations are zero, the skill skips directory creation and file writes entirely, reporting a single "nothing to do" line instead of an empty report.
• Self-Checked Recommendation Principles — recommendation_principles.md hard-bans wrapping existing abstractions, speculative optimizations, and decorative refactors — every suggestion must anchor to a real file and line number.
• Automatic Go Preprocessing — Every non-test .go file is run through gofmt -s -w before analysis, silently skipped on failure so the pipeline never breaks.

Architecture

Full Architecture

graph TB
    User[User] -->|/code-reviewer| SKILL[SKILL.md<br/>Orchestration]
    SKILL --> Analyze[analyze_code.py<br/>Language Detect + Dispatch]
    Analyze --> Lang[Go / Python / JS-TS<br/>Analyzers]
    Lang --> Gate[No-Op Gate]
    Gate --> Output[.doc/code-reviewer/<br/>Optimization Report]

Loading

License

This project is licensed under the MIT LICENSE.

---

©️ 2026