The Agent Skills Directory

[SAFE]: The skill provides defensive security auditing functionality, including static analysis patterns for common web and system vulnerabilities. The implementation uses standard Python libraries (re, dataclasses) to define detection signatures.
[INDIRECT_PROMPT_INJECTION]: The skill functions by ingesting and analyzing untrusted code generated by AI agents. This creates a potential attack surface for indirect prompt injection if the scanned content contains instructions intended to mislead the agent performing the review. However, the logic is based on static pattern matching which provides a defensive layer against such vectors.
Ingestion points: The scan_file method processes the content of external files, and scan_agent_threats processes tool_calls data.
Boundary markers: No specific delimiters are defined in the provided code snippet to separate untrusted code from the agent's instructions.
Capability inventory: The scanner performs passive regex-based analysis and does not execute or evaluate the content it scans.
Sanitization: The scanning logic itself serves as the primary sanitization and audit mechanism for the ingested data.

agent-security-scanning