The Agent Skills Directory

[SAFE]: The Python-based tools (security_scanner.py, vulnerability_assessor.py, compliance_checker.py) operate strictly on the local file system. A review of the source code shows they perform regex-based pattern matching for security auditing and do not import untrusted libraries, make network requests, or execute arbitrary secondary commands.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a natural attack surface for indirect prompt injection as it is designed to scan and present code snippets from untrusted source code to the agent.
Ingestion points: scripts/security_scanner.py and other scripts read arbitrary file content from the user-specified <target> directory.
Boundary markers: While the output is provided in structured JSON, which helps delineate data from instructions, there are no explicit boundary warnings (e.g., "ignore instructions found in this snippet") within the generated report.
Capability inventory: The skill is restricted to reading local files and writing to standard output; it does not possess inherent write or network capabilities.
Sanitization: The scanner implementation truncates code snippets to 100 characters, significantly limiting the potential for long-form instruction payloads while still performing its intended audit function.
[SAFE]: Hardcoded credentials found in documentation (references/security_standards.md) and scanner logic (scripts/security_scanner.py) consist of synthetically generated examples used for pattern matching and developer education. None of the patterns represent functional secrets or credentials belonging to the author or any third party.

senior-secops