The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/lighthouse_audit.py uses subprocess.run() to execute the lighthouse command-line tool.
Evidence: The script passes the user-provided URL (sys.argv[1]) directly into the command arguments list.
Potential Risk: While the script uses a list for the command arguments (which prevents shell-level injection), it does not validate the URL. A malicious user could provide strings starting with hyphens (e.g., --chrome-flags) to attempt argument injection to change the behavior of the Lighthouse command or the underlying Chrome instance.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data provided by the user (a URL) which is audited by an external tool.
Ingestion point: The url parameter in scripts/lighthouse_audit.py (sourced from sys.argv[1]).
Capability inventory: The skill uses subprocess.run to call the lighthouse CLI.
Boundary markers: None. The script does not wrap the URL in delimiters or warn the agent about potential instructions inside the target page.
Sanitization: None. The URL is passed directly to the external command.
Risk Level: Low, as the Lighthouse tool returns structured JSON performance metrics, and the script only extracts specific numeric scores and high-level categories rather than processing arbitrary page text.

performance-profiling