The Agent Skills Directory

[SAFE]: The skill conducts read-only security audits of project files, which is its primary stated purpose. It includes explicit instructions to redact discovered secrets (e.g., API keys, tokens) by showing only the first four characters before including them in any reports.- [SAFE]: The skill uses platform-specific directives like 'AUTONOMOUS MODE' to manage the agent's interactive behavior during the scanning process, ensuring a continuous workflow without unnecessary interruptions.- [SAFE]: The skill implements a self-evolution telemetry feature that logs execution metadata to a local directory ('~/.claude/projects/') associated with the target platform's project memory features.- [SAFE]: The skill manages a surface for indirect prompt injection as it processes untrusted project source code and configuration files.
Ingestion points: The skill reads various project artifacts including source code, dependency manifests (e.g., 'package.json', 'requirements.txt'), and framework configurations.
Boundary markers: The instructions do not specify explicit delimiters or isolation markers for the audited content, relying on the agent's context window.
Capability inventory: The skill's capabilities are restricted to file system read operations and appending metadata to a specific telemetry file; it explicitly forbids running exploits or modifying code.
Sanitization: The skill performs output-level sanitization by redacting credentials and sensitive identifiers discovered during the audit.

owasp