The Agent Skills Directory

[SAFE]: The skill is designed as a security utility to monitor and audit local skill files for common issues like dead code, missing documentation, and unsafe shell execution paths. No malicious patterns such as exfiltration, obfuscation, or remote code execution were found in the provided instructions.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes untrusted data from other SKILL.md files.
Ingestion points: Recursively reads content from all markdown files in the skills/ directory.
Boundary markers: No specific delimiters or markers are defined to isolate untrusted skill content from the auditing instructions.
Capability inventory: Includes the ability to generate reports, perform file modifications (auto-fixes), and delete files.
Sanitization: The skill relies on human-in-the-loop confirmation before executing sensitive file operations, which serves as a primary defense against malicious injection attempts.

skill-warden