The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/security_scan.py executes external tools npm audit and pip-audit using subprocess.run. While these are standard security utilities, the use of subprocess calls represents a privileged capability. The implementation uses list-based arguments which mitigates shell injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a user-specified directory and presents snippets of that data back to the agent. This creates a surface for indirect prompt injection if the scanned files contain malicious instructions.
Ingestion points: scripts/security_scan.py reads file content using walk_files and read_lines functions for analysis.
Boundary markers: Absent. The scan output does not use specialized delimiters or instructions to prevent the LLM from obeying directives found within code snippets.
Capability inventory: The skill can execute shell commands (standard audit tools) and read wide range of local filesystem content.
Sanitization: Snippets extracted from codebase files are truncated to 200 characters for display, but no systematic sanitization for prompt injection markers is performed.

code-vuln-audit