菜单
Scan and validate AI agent skills against the OWASP Agentic Skills Top 10 (AST10) security framework. Detects malicious skills, prompt injection, data exfiltration, supply chain risks, and cross-platform metadata loss. Provides CI/CD-ready security gating with SARIF output.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/JPeetz/agent-skills --skill agentic-security-scanner命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Design-first API development skill. Generates OpenAPI 3.1 specifications, enforces REST design best practices, validates endpoints, handles versioning, pagination, error formatting, authentication patterns, rate limiting, and idempotency. Activates when users say "design an API", "create OpenAPI spec", "API endpoint", "REST API design", "API contract", "Swagger/OpenAPI doc", "API versioning", "rate limiting", or "API security". Covers REST, GraphQL schema design, and gRPC proto generation with cross-protocol consistency.
AI-powered GraphQL API design, implementation, and optimization. Covers schema-first design, resolver architecture, query optimization with DataLoader for N+1 prevention, mutation patterns with idempotency, real-time subscriptions, Apollo Federation for distributed graphs, security hardening (depth limiting, rate limiting, authz), and production performance (persisted queries, caching, CDN integration). Primary keyword clusters: GraphQL schema design best practices, Apollo Federation subgraph patterns, DataLoader N+1 query optimization, GraphQL security depth limiting rate limiting, GraphQL persisted queries performance, GraphQL subscription real-time patterns, GraphQL error handling union types, GraphQL pagination relay cursor connection, GraphQL caching strategies production, GraphQL resolver architecture patterns. Designed for agentic platforms — Claude Code, Codex, Cursor, Gemini CLI, OpenClaw, GitHub Copilot, Windsurf, and OpenCode.
Use this skill when an agent needs to call a website's hidden or undocumented API: capture real browser requests into a HAR file, derive the exact request shape, build a clean replayable client, verify it outside the browser, and reuse the verified client. Activates when a backend has no documented API, or when you must reproduce the precise request (method, URL, headers, JSON body) of an action you can only perform in a browser. Authorized use only — never to bypass auth or bot detection.
| name | agentic-security-scanner |
| description | Scan and validate AI agent skills against the OWASP Agentic Skills Top 10 (AST10) security framework. Detects malicious skills, prompt injection, data exfiltration, supply chain risks, and cross-platform metadata loss. Provides CI/CD-ready security gating with SARIF output. |
| version | 1.0.0 |
| platforms | ["openclaw","claude","codex","cursor","gemini","copilot","opencode","windsurf"] |
| author | {"name":"Skill Foundry (Forge)","source":"OWASP Agentic Skills Top 10, SkillSpector Scanner"} |
| license | MIT |
| risk_tier | L1 |
| tags | ["security","owasp","scanner","ast10","compliance","supply-chain","ci-cd"] |
| requires | {"binaries":["jq","python3"]} |
Scan and validate AI agent skills against the OWASP Agentic Skills Top 10 (AST10) — the definitive security framework for agent skill ecosystems. Detect malicious skills, prompt injection sinks, data exfiltration paths, supply chain risks, and cross-platform metadata loss before skills are installed or published.
Use this skill when:
The OWASP Agentic Skills Top 10 (AST10) defines the 10 most critical
security risks in AI agent skills across all major platforms:
OpenClaw (SKILL.md), Claude Code (skill.json), Cursor/Codex
(manifest.json), and VS Code (package.json).
Skills that contain hidden destructive commands, backdoors, or malware.
Skills with instructions that can be overridden by user input, causing the agent to execute unintended actions.
Skills that send sensitive data (env vars, API keys, source code) to external endpoints without disclosure.
Skills requesting broader permissions than their stated purpose requires.
Skills depending on unverified packages, unpinned versions, or compromised repositories.
requires.binaries, requires.packages,
check version pinning, verify dependency integritySkills executing scripts without sandboxing, input validation, or output verification.
Skills without content hashes, signatures, or provenance records.
content_hash, signature fields present
and valid in skill metadataSkills reading or writing files outside declared paths, using wildcards, or accessing identity files (SOUL.md, MEMORY.md).
permissions.files.read and deny_write for
wildcards, verify no access to protected pathsSkills opening network connections without declared domain allowlists.
permissions.network.allow and deny fields,
verify no catch-all patternsSecurity metadata weakened or lost when porting skills between platforms.
Run the scanner to produce a 0-100 risk score and AST01-AST10 mapping:
python3 scripts/scan_skill.py --skill-dir <path> --format sarif --output results.sarif
The scanner checks:
| Score | Risk Tier | Action |
|---|---|---|
| 0-20 | L0 (Safe) | Allow installation |
| 21-40 | L1 (Low) | Review recommended |
| 41-60 | L2 (Elevated) | Manual review required |
| 61-80 | L3 (High) | Block until remediated |
| 81-100 | Critical | Reject permanently |
GitHub Actions gate:
- name: Scan Skills
run: |
for skill in skills/*/; do
python3 scripts/scan_skill.py --skill-dir "$skill" \
--format sarif --output "results/$(basename $skill).sarif"
done
- name: Upload SARIF
uses: github/codeql-action/upload-sarif@v3
with:
sarif_file: results/
- name: Gate on Score
run: |
python3 scripts/gate_score.py --max-score 60 --results-dir results/
For each finding, the scanner produces:
# Clone to your agent's skills directory
git clone https://github.com/skill-foundry/agentic-security-scanner.git \
~/.agents/skills/agentic-security-scanner
# Or via npx
npx skills add skill-foundry/agentic-security-scanner
SKILL.md YAML frontmatter with permissions blockskill.json and .claude/skills/ directory structuremanifest.json with tool declarationspackage.json extension contributions