The Agent Skills Directory

[COMMAND_EXECUTION]: The toolkit includes a Python script (scripts/security_audit_report.py) that performs legitimate command execution to support its auditing functions. It uses the subprocess module to run git clone for fetching remote repositories and launches a headless browser (e.g., Google Chrome, Chromium, or Microsoft Edge) to render HTML reports into PDF format. These commands are implemented securely using argument lists rather than shell strings, which mitigates the risk of command injection.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is the analysis of external, untrusted codebases, which inherently introduces an indirect prompt injection surface.
Ingestion points: The agent is instructed to read files from a target-source directory created by scripts/security_audit_report.py in an isolated temporary workdir.
Boundary markers: The workflow does not specify the use of delimiters or specific instructions to ignore embedded commands when the agent inventories or deep-reads the source code.
Capability inventory: The skill can execute the provided Python scripts, which in turn can execute git and a headless browser. However, these are managed through specific script logic rather than arbitrary execution.
Sanitization: While the skill does not sanitize the input code it reads, it includes a robust redaction engine in scripts/security_audit_report.py to prevent the leakage of secrets, keys, and absolute paths from the analyzed code into the final generated reports.

yao-websecurity-skill