返回 Skill 列表
extension
分类: 开发与工程无需 API Key

assess-codebase

评估代码库中的模式、反模式和质量改进机会;在被要求生成编码规则、标准或质量指南时使用。

person作者: jakexiaohubgithub

Assess Codebase

Overview

Run a multi-agent assessment using CLI subagents and synthesize code quality rules.

Inputs

  • Focus area or directories (optional)
  • File patterns (optional)

Workflow

  1. Identify scope and primary languages using rg --files -g and recent git activity.
  2. Run three CLI subagents (Gemini Flash, Codex, Claude Haiku) using the commands below.
  3. Collect outputs, dedupe, and group by patterns, anti-patterns, and opportunities.
  4. Draft rules and guidelines with short examples.
  5. Save results to CLAUDE.md or AGENTS.md if requested.

Subagent Commands

Gemini Flash (patterns)

CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-flash-preview -o text "Assess this codebase for patterns in naming, imports, structure, and consistency. Return a structured list with file paths."

Codex (anti-patterns)

codex exec -m gpt-5.2 -s read-only -c model_reasoning_effort="medium" --skip-git-repo-check \
  "Assess this codebase for anti-patterns (deep nesting, dense one-liners, inconsistent error handling, duplication). Return a structured list with file paths."

Claude Haiku (opportunities)

claude --model haiku -p "Assess this codebase for clarity, consistency, maintainability, testing, and docs opportunities. Return a structured list with file paths."

Output

  • Patterns list
  • Anti-patterns list
  • Proposed rules and guidelines with priorities