The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/security_scan.py uses subprocess.run to execute the npm audit command on a specified project path. This is a controlled execution of a standard security utility used for its intended purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes external project files which constitutes an ingestion surface for untrusted data. A malicious actor could theoretically place instructions in comments within those files to attempt to influence the agent's analysis.
Ingestion points: The skill reads source code and configuration files from the user-provided project_path during the scanning process (scripts/security_scan.py).
Boundary markers: The SKILL.md workflow explicitly instructs the operator to "keep provenance and source boundaries explicit in the working notes," providing a manual boundary for the agent.
Capability inventory: The skill can read local files and execute a predefined npm audit command via the provided Python script (scripts/security_scan.py).
Sanitization: No explicit sanitization or filtering of the file contents is performed before the agent processes the generated scan results.
[SAFE]: Static analysis markers for dangerous functions like eval, exec, and subprocess with shell=True are false positives. These strings are contained within regex pattern lists used by the script to find vulnerabilities in other code and are not executed by the skill itself. References to .env files are also used as search criteria for identifying hardcoded secrets in the target project, not for exfiltration.

vulnerability-scanner