The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The scripts/security_scan.py script reads local file contents, including potential access to sensitive configurations like .env and AWS credentials, to identify hardcoded secrets. This behavior is expected for its role as a security scanner and does not involve unauthorized data transmission.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill utilizes the subprocess module to execute npm audit, allowing it to identify known vulnerabilities in external dependencies. This is a standard security practice and utilizes a well-known service.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code from external projects, creating a potential surface for indirect injection if the agent interprets embedded instructions in the scanned code.
Ingestion points: The security_scan.py script reads code and configuration files from a user-specified directory.
Boundary markers: Absent; the scanner returns raw snippets in its output without delimiters to distinguish untrusted data from instructions.
Capability inventory: The script can read local files and execute specific system commands like npm audit.
Sanitization: The scanner identifies patterns using regex but does not perform sanitization or escaping of the resulting code snippets before returning them to the agent.

vulnerability-scanner