The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code and dependency files from a target directory. While the provided scripts use regex-based scanning and do not currently interpolate results into subsequent LLM prompts, the processing of external data represents a potential attack surface for malformed code targeting static analyzers.
Ingestion points: scripts/security_scanner.py, scripts/vulnerability_assessor.py, and scripts/compliance_checker.py all traverse and read the content of files from a user-specified target directory.
Boundary markers: No specific boundary markers or 'ignore' instructions are used when reading input files; they are processed as raw text for pattern matching.
Capability inventory: The scanning scripts perform filesystem reads and can write results to output files (via JSON export). They do not execute the scanned code or perform network operations.
Sanitization: The skill's reference documentation (references/security_standards.md) provides extensive guidance on input validation and output encoding, although the scanner scripts themselves rely primarily on regular expression matching.
[EXTERNAL_DOWNLOADS]: The skill's documentation and YAML examples reference several well-known third-party security tools and GitHub actions.
Evidence: Mentions of gitleaks, detect-secrets (Yelp repository), trufflehog, syft, cosign, and cyclonedx-cli are included as standard security recommendations. These references target official repositories of trusted organizations and well-known services.
[SAFE]: The skill includes hardcoded examples of security vulnerabilities (e.g., sample API keys and SQL injection patterns) in SKILL.md and scripts/security_scanner.py. These are explicitly labeled as unsafe examples or used as detection patterns for the scanner, posing no risk to the execution environment.

senior-secops