The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/security_scan.py executes external security tools such as npm and pip-audit to perform dependency scans. These executions are performed using list-based arguments with subprocess.run(), which avoids shell invocation and mitigates command injection risks.
[EXTERNAL_DOWNLOADS]: Dependency scanning modules rely on npm audit and pip-audit, which connect to official package registries (e.g., npmjs.org, pypi.org) to retrieve vulnerability information. These are well-known technology services used for their intended purpose.
[DATA_EXFILTRATION]: The tool iterates through and reads local project files to perform its analysis. The analysis confirms that no network operations are present in the script that would exfiltrate the contents of these files to external or unauthorized servers.
[PROMPT_INJECTION]: As the skill ingests and displays content from untrusted local files (user code), it has an inherent surface for indirect prompt injection (Category 8). An attacker could place malicious instructions inside a file's comments or code snippets. However, the script mitigates this by using structured output formats (JSON/Text) and truncating snippets, and the agent's internal guardrails typically handle such data-as-instruction scenarios.

code-vuln-audit