The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file scripts/config.py contains a hardcoded identifier 546c25a59c58ad7 within the SECRET_EXCEPTIONS list. While this is documented as a public key for the Imgur API used to filter false positives during audits, hardcoding authentication-related strings is generally noted as a security concern.\n- [DATA_EXFILTRATION]: The skill performs automated discovery of other skills by scanning the filesystem in scripts/scanner.py. It recursively searches directories defined in SKILL_SEARCH_PATHS, which includes the user's primary skills directory and configuration paths like .claude/skills, to read SKILL.md and Python source files for static analysis.\n- [INDIRECT_PROMPT_INJECTION]: The skill acts as a meta-auditor that processes content from third-party skill files.\n
Ingestion points: scripts/scanner.py reads Markdown documentation and Python scripts from the local filesystem.\n
Boundary markers: The ingestion logic does not employ explicit delimiters or safety warnings when processing external file content.\n
Capability inventory: Filesystem read access, local SQLite database writes, and Markdown report generation.\n
Sanitization: The skill performs static analysis using the Python ast module to avoid execution, but extracted findings and text are presented directly in reports seen by the agent.

skill-sentinel