cisco-ai-skill-scanner

What it is and what it does

Cisco AI Skill Scanner is a multi-engine security scanner designed to detect threats in AI Agent Skills—executable skill definitions for LLM agents following OpenAI Codex or Cursor Agent formats. It combines static pattern matching (YAML + YARA rules), behavioral dataflow analysis, and optional LLM-as-a-judge semantic analysis to identify prompt injection, data exfiltration, and malicious code patterns. The tool explicitly disclaims comprehensive coverage: a clean scan does not guarantee security, and novel attacks may evade detection.

The scanner integrates into CI/CD pipelines via SARIF output for GitHub Code Scanning, pre-commit hooks, and configurable exit codes. It supports custom YARA rules, threat taxonomies, and scan policies (strict/balanced/permissive), plus optional cloud integrations (VirusTotal, AWS Bedrock, Google Gemini, Azure OpenAI). A meta-analyzer reduces false positives, and an interactive wizard guides first-time users through CLI options.

Use it for:

• Scan AI agent skills in a GitHub Actions workflow before deployment to catch known threat patterns early.
• Use as a pre-commit hook to prevent developers from committing skills with obvious prompt injection or exfiltration code.
• Audit a repository of existing skills with behavioral and LLM analyzers enabled to identify latent threats across the codebase.
• Generate SARIF reports for GitHub Code Scanning to surface skill security findings alongside other code quality checks.
• Customize scan policies and YARA rules to match your organization's risk tolerance and threat model.
• Integrate VirusTotal or Cisco AI Defense cloud scanning for binary and hash-based threat correlation.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Scans AI Agent Skills for prompt injection, data exfiltration, and malicious code using pattern-based detection, LLM analysis, and behavioral dataflow inspection.

Yes, with conditions. Install if you deploy AI agent skills and want a layered, best-effort threat detection baseline. The tool is actively maintained, has low install friction, and integrates cleanly into CI/CD. However, do not rely on it as your sole security control: the documentation explicitly warns that clean scans do not guarantee safety, and human code review remains essential for high-risk deployments. Suitable for teams building or consuming OpenAI Codex or Cursor Agent skills who need automated early warning.

Install

cisco-ai-skill-scanner on PyPI

pip install cisco-ai-skill-scanner

uv add cisco-ai-skill-scanner

poetry add cisco-ai-skill-scanner

Installing cisco-ai-skill-scanner

Before you install

Low friction: pure Python wheel with no compiled dependencies. Active maintenance—released 11 days ago with 2431 GitHub stars. Requires Python 3.10+. Twenty runtime dependencies are substantial but all standard (anthropic, openai, fastapi, pydantic, yara-x, etc.), indicating a mature, feature-complete tool rather than a minimal proof-of-concept.

License in practice

Apache-2.0 permissive license allows commercial and private use without restriction, modification, or redistribution obligations beyond license notice.

Quickstart

pip install cisco-ai-skill-scanner

skill-scanner scan /path/to/skill

# With LLM analysis and behavioral dataflow:
skill-scanner scan /path/to/skill --use-behavioral --use-llm

Requires Python 3.10 or later. LLM and cloud-based analyzers require API keys (SKILL_SCANNER_LLM_API_KEY, VIRUSTOTAL_API_KEY, AI_DEFENSE_API_KEY) set as environment variables; core static analysis runs without them.

Verify before relying

• Actual false positive rate and detection precision across different threat categories in production use.
• Performance characteristics (scan time, memory usage) on large skill repositories or complex dataflow graphs.
• Whether the LLM consensus mode (multiple runs) materially improves detection accuracy over single-run LLM analysis.

Package facts

Evidence: cisco_ai_skill_scanner-2.0.13-py3-none-any.whl

Keywords: ai-security, anthropic, claude, codex, llm-security, mcp, openai, prompt-injection, scanner, security, skills, static-analysis, threat-detection

Tags

More Testing packages

Further reading