Token导航 LogoToken导航TokenDH.com
研究检索需要联网github未标认证来源可访问许可证需确认审计通过

deep-codebase-audit深度代码库审计

Agent Skill

用于辅助安全审计、权限检查、凭据风险、认证流程和常见漏洞排查。它适合让 Agent 梳理敏感配置、检查依赖风险、分析鉴权逻辑或生成安全复核清单。使用时不能把工具输出直接当最终结论,涉及密钥、令牌、用户数据或生产系统时,应先确认最小权限、脱敏方式和操作边界。

总安装

250

周安装

10

GitHub Stars

6

下载量

81
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:deep-codebase-audit(深度代码库审计)
来源仓库:https://github.com/kambleakash0/agent-skills
仓库路径:skills/deep-codebase-audit
安装命令:
npx skills add https://github.com/kambleakash0/agent-skills --skill deep-codebase-audit
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/kambleakash0/agent-skills --skill deep-codebase-audit

简介

用于提升代码库可测试性与可修改性,识别模块耦合与架构摩擦点。

  • 通过深度遍历文件与引用关系,建议改进模块划分与接口设计。
  • 适用于代码混乱、难以变更或“未准备好对接 AI”的项目场景。
  • 输出包含改进建议但非最终结论,需人工确认改动影响与测试覆盖。
  • deep-codebase-audit 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

Improve Codebase Architecture

This skill explores a codebase the way an AI agent experiences it: by reading files, following references, and trying to understand how to make safe changes. It surfaces architectural friction (shallow modules, tangled dependencies, missing seams) and suggests deep-module improvements that make the codebase easier to test, reason about, and modify with AI.

References for this skill are kept in references.md

When to Use

Use this skill when:

  • The user says the codebase feels messy, hard to change, or “not ready for AI”.
  • You want to improve testability and make future TDD or autonomous agents more effective.
  • You’re about to invest heavily in AI‑driven changes and want to reduce risk first.

If the user just wants to ship a small feature, consider using /tdd or similar execution skills instead.

Overall Approach

The goal is not to apply a rigid checklist, but to explore organically and report back what hurts and where to deepen modules.

  1. Clarify goals and constraints

- Ask what “better architecture” means in this context: easier AI changes, better tests, less cognitive load, fewer regressions, etc. - Ask about constraints: tech stack, team experience, time horizon (quick wins vs deeper re‑architecture).

  1. Map the codebase like an AI

- Start from natural entry points: main app, top‑level routes, CLI commands, or “hot” areas the user mentions. - Follow imports and references to build a mental map of major modules and how they relate. - Notice where behavior is spread across many shallow modules versus encapsulated in a few deep ones.

  1. Identify architectural friction Look for and note:

- Shallow modules: Lots of tiny helpers that do very little each, forcing callers (and AI) to stitch behavior together. - Cross‑cutting concerns scattered everywhere (logging, auth, validation) instead of grouped behind boundaries. - Hidden contracts: Implicit assumptions encoded in call order, parameter combinations, or shared mutable state. - Hard‑to‑test areas: Code that touches many systems at once, with no obvious narrow seam to test.

  1. Spot opportunities for deep modules

- From the friction you find, propose deep module seams: places where you could introduce or strengthen modules with simple APIs and rich internals. - Think in terms of “lumps of responsibility” an AI could own: “billing calculator”, “metrics aggregator”, “notification scheduler”, etc. - For each candidate module, sketch what a clean, stable interface might look like and what behavior it would hide.

  1. Evaluate AI‑friendliness For each key area:

- How easy would it be for an AI (or new human) to find the right module and call it safely? - Is there a clear public interface and a matching set of tests? - Does the current shape encourage progressive disclosure of complexity: simple on the outside, complex only when you dive in?

  1. Propose practical refactors

- Turn your findings into concrete, incremental refactors: deepen a module, extract a seam, consolidate duplicated logic, wrap a messy call pattern in a single function. - Prioritize changes that will unlock better tests, clearer interfaces, or safer AI modifications. - Suggest a small number of vertical slices (e.g., “make the metrics subsystem deep‑module‑friendly”) instead of a total rewrite.

  1. Summarize and hand off

- Write a short architecture report: current pain points, candidate deep modules, suggested refactors, and expected benefits for AI and humans. - Recommend follow‑up skills: /tdd for test‑driven refactors, /spec-writer and /slice-the-spec for larger architectural initiatives.

Behavior and Rules

  1. Stay descriptive before prescriptive. Spend time describing what the codebase feels like to work with before jumping to solutions.
  2. Prioritize leverage. Focus on changes that will make many future changes easier, not micro‑tweaks.
  3. Think in deep modules. Whenever you see scattered logic, ask “what deep module could own this?”.
  4. Respect constraints. If the user says they only have time for quick wins, adjust recommendations accordingly.
  5. Avoid framework dogma. Use the existing language and framework idioms; don’t reinvent the project around a new architecture trend.
  6. Write for humans and agents. Explanations and recommendations should be understandable both to human engineers and future AI agents reading the report.

Example

User: “This Typescript backend feels impossible to navigate and test. Make it more AI‑friendly.”

You (high level):

  1. Ask what “AI‑friendly” means to them (e.g., safer autonomous changes, fewer regressions).
  2. Explore the codebase from the main entrypoints, mapping major modules and noting shallow, scattered logic.
  3. Identify friction: cross‑cutting concerns tangled through routes, no clear deep modules for billing, notifications, or metrics.
  4. Propose 3–5 deep module candidates with sketched interfaces and concrete refactors to get there.
  5. Summarize in a short report and suggest follow‑up TDD work to implement the refactors safely.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

35.36%
按下载量换算29

Claude

32.72%
按下载量换算27

Cursor

20.09%
按下载量换算16

Gemini CLI

9.18%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills