Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsBlogPro
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Authors
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges
  • Chrome Extension
  • Skill Manager

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

ProTermsPrivacyRefunds
Back to skills

Agent Plugin Audit

ASecurity

当要把一个技能/插件/Agent/命令合入或发布前做一次性全量闸门时使用;做结构+质量+脚本+安全+市场合规+生态集成+代码评审的 8 阶段审计并产出 PASS/带警告 PASS/FAIL 的闸门报告(自动修非关键项);不适用于运行时沙箱、单纯安全扫描(用 agent-skill-security-scanner)、CVE 实时联网核验;触发词:审计这个插件、发布前闸门、plugin audit、技能合规检查、上架前检查、8 阶段审计、quality gate

3 stars
0 votes
0 copies
1 views
Added 9/19/2026
ai-agentspythongobashcode-reviewgitsecurity

Works with

cursorcli

Security Analysis

A100/100

Scanned 9/19/2026

$npx -y skills add findscripter/everything-skills --skill agent-plugin-audit --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Agent Plugin Audit?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Agent Plugin Audit
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/findscripter-agent-plugin-audit/badge)](https://www.skillsdirectory.com/skills/findscripter-agent-plugin-audit)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
Files
SKILL.md
---
name: agent-plugin-audit
title: Agent 插件安全审计
description: 当要把一个技能/插件/Agent/命令合入或发布前做一次性全量闸门时使用;做结构+质量+脚本+安全+市场合规+生态集成+代码评审的 8 阶段审计并产出 PASS/带警告 PASS/FAIL 的闸门报告(自动修非关键项);不适用于运行时沙箱、单纯安全扫描(用 agent-skill-security-scanner)、CVE 实时联网核验;触发词:审计这个插件、发布前闸门、plugin audit、技能合规检查、上架前检查、8 阶段审计、quality gate
domain: 安全/appsec
triggers: [审计这个插件, 发布前闸门, 上架前检查, 技能合规检查, 8 阶段审计, plugin audit, quality gate, pre-publish check]
tags: [security, appsec, plugin-audit, publish-gate, quality-gate, marketplace-compliance, ci-gate, skill-validation]
level: 进阶
status: stable
agents: [claude-code, codex, cursor, gemini-cli]
tools: [python3, skill_validator.py, quality_scorer.py, script_tester.py, skill_security_auditor.py, git]
requires: []
related: [agent-skill-security-scanner, agentic-actions-auditor, dependency-auditor, adversarial-code-reviewer]
combines_with: [skill-creator, supply-chain-risk-auditor]
license: MIT
source: alirezarezvani/claude-skills
source_license: MIT
---
# Agent 插件安全审计

## 何时使用

- 在把一个技能 / 插件 / Agent / 命令合入主干或上架插件市场前,要跑一次**端到端发布闸门**,一次覆盖结构、质量、脚本、安全、市场合规、生态集成、代码评审。
- 想要可重复、可进 CI 的「准入/拒绝」判定,并自动修掉非关键问题、只在关键决策(破坏性变更、引入新依赖)上停下来问人。

不该用的边界:
- 只想做**纯安全静态扫描**(危险模式、提示注入、供应链)时,用更聚焦的 `agent-skill-security-scanner`,不必跑全 8 阶段。
- 不做运行时沙箱 / 动态执行,也不联网实时核验 CVE;安全阶段用的是内置静态规则。
- 强依赖本仓自带的校验脚本(`skill_validator.py` / `quality_scorer.py` / `script_tester.py` / `skill_security_auditor.py`);脱离该工具链时按下方「指令」中的等价判据手工执行。

## 步骤

按顺序跑 8 个阶段,遇关键失败即停,非关键问题自动修复,最后出报告。

1. **发现(Discovery)**:确认 `{skill_path}` 存在且含 `SKILL.md`;读 frontmatter 取 `name/description/Category/Tier`;探测子目录(`scripts/`→有脚本、`references/`→有参考、`assets/`→有模板、`agents/`→内嵌 Agent、`skills/`→复合技能、`.claude-plugin/plugin.json`→独立插件);从路径推断 domain;在 `commands/` 找同名命令。先打印发现摘要再继续。
2. **结构校验(Structure)**:跑 `skill_validator.py`,取总分与合规级、失败检查项。**闸门:分数 ≥ 75**。低于则自动补全(缺字段→从正文补、缺小节→补桩标题、缺目录→建空目录加说明),重跑;仍 < 75 记 FAIL 但继续收集后续结果。
3. **质量评分(Quality)**:跑 `quality_scorer.py`,取总分/等级、四维分(文档/代码质量/完整度/可用性)、改进路线项。**闸门:分数 ≥ 60**;低于则把路线项列为行动项。
4. **脚本测试(Scripts)**:有 `scripts/*.py` 才跑 `script_tester.py`。**闸门:所有脚本必须 PASS**,任何 FAIL 即阻断,PARTIAL 触发警告。自动修:`--help` 失败查是否缺 `argparse`;若 stdlib-only 失败,标出该 import 并**询问用户**是否接受该依赖(关键决策)。
5. **安全审计(Security)**:跑 `skill_security_auditor.py --strict`,取裁决(PASS/WARN/FAIL)、CRITICAL/HIGH/INFO 发现。**闸门:CRITICAL 与 HIGH 均须为 0**,任一即阻断(报出文件、行号、模式、修复建议)。**安全问题不自动修**,报告后交用户决定。
6. **市场与插件合规(Marketplace)**:校验 `plugin.json`(仅允许 `name/description/version/author/homepage/repository/license/skills` 字段、版本须与仓库版本一致、`skills` 取约定值、`name` 与目录名一致);校验 `settings.json` 中 `commands` 是否都有对应文件;核对 `marketplace.json` 条目与父域 `plugin.json` 的技能计数。自动修:更新陈旧计数、修版本不一致、删多余字段。
7. **生态集成(Ecosystem)**:核对跨平台索引(如 `.codex/` `.gemini/` 索引)含本技能,缺则重跑同步脚本;校验关联命令的 YAML frontmatter、引用路径与文档导航;校验内嵌 Agent 的 frontmatter 与相对路径;核对 SKILL.md 中对其他技能的引用是否真实存在。自动修:补缺的导航/索引条目。
8. **领域化代码评审(Code Review)**:按 domain 选对应 Agent 视角(如 `engineering/`→cs-senior-engineer、`product-team/`→cs-product-manager、其它→cs-senior-engineer),用其评审标准检查工作流可执行性、脚本正确性、参考准确性、模板可用性、无死链、署名齐全。这是**借用 Agent 的书面视角组织评审,并非派生子 Agent**。

## 指令

```bash
# 用法
/plugin-audit <skill-path>

# 各阶段核心命令(路径相对仓库根;{skill_path} 为待审目标)
python3 engineering/skill-tester/scripts/skill_validator.py {skill_path} --tier {tier} --json     # 阶段2 结构
python3 engineering/skill-tester/scripts/quality_scorer.py {skill_path} --detailed --json          # 阶段3 质量
python3 engineering/skill-tester/scripts/script_tester.py {skill_path} --json --verbose             # 阶段4 脚本
python3 engineering/skill-security-auditor/scripts/skill_security_auditor.py {skill_path} --strict --json   # 阶段5 安全
```

**裁决逻辑**:

| 条件 | 裁决 |
|---|---|
| 全部阶段通过 | **PASS** —— 可合并/发布 |
| 仅有警告、无阻断 | **带警告 PASS** —— 合并前先看警告 |
| 任一阶段有阻断 | **FAIL** —— 列出阻断点与修复指引 |

**阻断项(任一即 FAIL)**:结构分 < 75;质量分 < 60;任一脚本 FAIL;任一 CRITICAL/HIGH 安全发现;`plugin.json` 无效或含非法字段;版本与仓库不一致。

**非阻断(仅警告)**:质量分在 60–75;脚本 PARTIAL;某平台索引缺失(自动修);缺导航条目(自动修);安全 INFO 发现。

## 示例

最终报告(结构示意):

```
PLUGIN AUDIT REPORT: {skill_name}
  Phase 1 — Discovery     OK  {type}, {domain}
  Phase 2 — Structure     OK  {score}/100 ({level})    # 闸门 ≥75
  Phase 3 — Quality       OK  {score}/100 ({grade})    # 闸门 ≥60
  Phase 4 — Scripts       OK  {n}/{n} PASS
  Phase 5 — Security      OK  PASS (0 critical, 0 high)
  Phase 6 — Marketplace   OK  plugin.json valid
  Phase 7 — Ecosystem     OK  Codex + Gemini synced
  Phase 8 — Code Review   OK  {domain} review passed
  VERDICT: PASS — 可合并/发布
  Auto-fixes: {n}   Warnings: {n}   Action items: {n}
```

## 注意事项

- 阶段顺序固定,但**关键失败即停**:脚本非 stdlib-only 依赖、安全 CRITICAL/HIGH 都属关键决策,必须停下问人或交人裁决,不可静默放行。
- **安全发现绝不自动修**——只报告(文件/行号/模式/建议),由人决定整改方式;其余非关键项(缺字段、陈旧计数、缺索引/导航)可自动修后重校。
- `plugin.json` 的 `version` 必须与当前仓库版本一致,且字段白名单严格;多余字段一律删除。
- 本条是**全量发布闸门**,与纯安全扫描 `agent-skill-security-scanner` 分工:前者覆盖结构/质量/脚本/合规/生态/评审 6 个非安全维度,后者把安全静态扫描做深;两者可串联(先安全扫描排雷,再跑全量闸门定准入)。

## 互见

- related:`agent-skill-security-scanner` —— 安装前纯安全静态扫描,本条第 5 阶段的深化与替代选项。
- related:`agentic-actions-auditor` —— 审计 Agent 实际可执行动作的越权与副作用,补足代码评审视角。
- related:`dependency-auditor` —— 依赖供应链与已知漏洞深度核验,承接脚本/合规阶段的依赖发现。
- combines_with:`false-positive-check` —— 复核安全阶段发现,过滤误报后再给最终裁决。

本条采编自 alirezarezvani/claude-skills(MIT)。

Attribution

findscripterfindscripter
View sourceSee grades on GitHubMore from findscripter →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Terse caveman voice: answer first, fluff gone, every technical fact kept. Use for /caveman, "caveman mode", "talk like caveman", "be brief", "less tokens". Stays on until "stop caveman" or "normal mode".

1100021 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

698621 votes

Writing Skills

Create and manage Claude Code skills in HASH repository following Anthropic best practices. Use when creating new skills, modifying skill-rules.json, understanding trigger patterns, working with hooks, debugging skill activation, or implementing progressive disclosure. Covers skill structure, YAML frontmatter, trigger types (keywords, intent patterns), UserPromptSubmit hook, and the 500-line rule. Includes validation and debugging with SKILL_DEBUG. Examples include rust-error-stack, cargo-dep...

3931 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3421 votes

catchup

Recovers the conversation and failed tool calls of a previous Codex, Amp, Claude Code, Antigravity, Cline, Copilot CLI, Cursor, DeepSeek Harness, Grok Build, Kimi, OpenCode, Pi Agent, or ZCode session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", asks to recover/summarize a previous session before continuing, or asks to diagnose or report a catchup failure. Do NOT use for the current conversation, git history, or any non-agent log.

741 votes
View all in ai-agents →