skill-scanner

To scan agent skills for prompt injection and malicious code, run the built-in scanner on a skill directory using uv run ${CLAUDE_SKILL_ROOT}/scripts/scan_skill.py <skill-directory>. It analyzes SKILL.md, references, and bundled scripts to detect security risks before adoption.

Before adopting agent skills, check for prompt injection, dangerous code, secret exposure, excessive permissions, and supply chain risks. Scanning also enforces frontmatter validation, obfuscation checks, and URL extraction to generate structured findings for human review.

Yes, the agent skill scanner detects supply chain risks and performs obfuscation checks by analyzing bundled scripts and references. It applies risk assessment of permissions and scripts to identify dangerous code patterns and potential malicious payloads.

You can validate SKILL.md frontmatter and detect exposed secrets by running the scanner script. It enforces frontmatter validation and applies secret detection techniques across skill directories to extract keywords and identify sensitive data exposure.

You need pyyaml installed and uv to execute the scanner script for static analysis on agent skills. The scanner processes skill directories containing SKILL.md files, references, and scripts to generate structured findings and risk signals.

Static analysis identifies prompt injection in SKILL.md files by applying prompt-injection analysis and risk assessment of permissions. The scanner evaluates frontmatter, extracts URLs, and checks for obfuscation to generate structured findings for human review.