The Agent Skills Directory

[SAFE]: The skill is entirely instructional, providing a step-by-step workflow for human or AI agents to evaluate the security posture of other software components without executing them.
[SAFE]: No signs of obfuscation, credential harvesting, or unauthorized network operations were detected. The instructions explicitly forbid installing dependencies or providing credentials during the audit process.
[SAFE]: The frontmatter configuration is standard and restricts tool usage to established AI models and the host environment.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted third-party instructions (SKILL.md files), which creates an inherent attack surface for indirect prompt injection.
Ingestion points: Reads external SKILL.md files and referenced scripts/manifests.
Boundary markers: The instructions do not define specific delimiters for the audited content but do instruct the agent to 'Separate confirmed findings from contextual risks' and 'not call a project safe... without evidence'.
Capability inventory: The skill uses standard model capabilities for text analysis and does not invoke subprocesses or network tools.
Sanitization: None observed; the skill relies on the agent's analytical reasoning to identify malicious patterns.

skill-security-audit