Repolode
Submit
cisco-ai-defense

cisco-ai-defense/skill-scanner

A best-effort security scanner for AI Agent Skills that detects prompt injection, data exfiltration and malicious code patterns.

View on GitHubWebsite
Stars
2,701
Forks
342
Open issues
2
Last commit
yesterday

What is it?

Skill Scanner checks AI agent skills for risky content before they are installed or shared. It combines pattern-based detection with YAML and YARA-X rules, AST and dataflow analysis of Python code, and an optional LLM judge that reviews findings semantically. Results can be reported as JSON, Markdown, SARIF, or an interactive HTML report.

The scanner targets Codex and Cursor agent skill formats, and with a lenient flag also handles Claude Code command files and flat markdown skill repositories. It can scan a single skill, a directory of skills, or a GitHub repository, and offers a CLI, a Python SDK, a REST API, and a pre-commit hook. Policy presets such as balanced, low-noise and quiet control how many findings reach review.

Detection is explicitly best-effort: a scan with no findings does not certify a skill as safe, and the project recommends pairing results with manual review for production use.

Key features

  • 01Combines YAML and YARA-X pattern rules with AST and dataflow analysis.
  • 02Optional LLM judge for semantic review of findings, with multiple provider options.
  • 03Outputs JSON, Markdown, SARIF and interactive HTML reports.
  • 04Provides a reusable GitHub Actions workflow and a pre-commit hook.
  • 05Supports policy presets and custom YAML policies to tune noise.

Use cases

  • Vetting third-party agent skills before installing them into a coding assistant.
  • Gating skill changes in CI with SARIF results surfaced in GitHub Code Scanning.
  • Scanning a team's skill repository on every commit through pre-commit.
  • Auditing a directory of skills in bulk with cross-skill overlap checks.

Quick start

  1. 1

    Install the scanner with uv

    uv pip install cisco-ai-skill-scanner
  2. 2

    Set the LLM judge API key

    export SKILL_SCANNER_LLM_API_KEY="your_api_key"
  3. 3

    Scan a skill with the LLM judge and the balanced preset

    skill-scanner scan /path/to/skill --use-llm --policy balanced --fail-on-severity high

Things to know

  • Best-effort detection only; a scan with no findings does not guarantee a skill is safe.
  • Recommended setups need an LLM judge API key, since rules alone catch only about 8% of held-out malicious skills.
  • License is listed as NOASSERTION in metadata, so license terms should be checked before use.

In collections

AI agent skill librariesSecurity scanners, audits & threat intel toolsTesting frameworks & test automation tools

Topics

#agent-skills#security#prompt-injection#static-analysis#llm#yara#ci-cd#agent

FAQ

What is skill-scanner?
A best-effort security scanner for AI Agent Skills that detects prompt injection, data exfiltration and malicious code patterns. Skill Scanner checks AI agent skills for risky content before they are installed or shared. It combines pattern-based detection with YAML and YARA-X rules, AST and dataflow analysis of Python code, and an optional LLM judge that reviews findings semantically. Results can be reported as JSON, Markdown, SARIF, or an interactive HTML report.
Is skill-scanner free to use? What license does it have?
skill-scanner is open source under the Other license. Check the license terms before using it in commercial products.
Is skill-scanner still maintained?
Actively maintained. The last commit to skill-scanner was on Nov 6, 2026.
What can I use skill-scanner for?
Vetting third-party agent skills before installing them into a coding assistant. Gating skill changes in CI with SARIF results surfaced in GitHub Code Scanning. Scanning a team's skill repository on every commit through pre-commit. Auditing a directory of skills in bulk with cross-skill overlap checks.
How do I install skill-scanner?
Install the scanner with uv: uv pip install cisco-ai-skill-scanner Set the LLM judge API key: export SKILL_SCANNER_LLM_API_KEY="your_api_key" Scan a skill with the LLM judge and the balanced preset: skill-scanner scan /path/to/skill --use-llm --policy balanced --fail-on-severity high
Where can I get skill-scanner?
The source code and installation instructions are on GitHub: https://github.com/cisco-ai-defense/skill-scanner

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects