The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface within its report generation logic.
Ingestion points: The visualize.py script ingests data from user-provided Lighthouse JSON files via the --lighthouse argument or from a user-supplied URL string via the --url argument.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed JSON or URL data.
Capability inventory: The skill possesses the capability to execute python3 scripts that write to the local file system and interact with the system's default web browser using webbrowser.open().
Sanitization: The script fails to sanitize the url field (and others) before interpolating them into the HTML report template (f'<p class="url">{url}</p>'). This allows for potential XSS or injection if the source data is controlled by an attacker.
[COMMAND_EXECUTION]: The skill requires the agent to execute a local Python script (visualize.py). This script performs automated local actions, specifically using the webbrowser library to open generated files in the host system's web browser, which is an external side effect.
[EXTERNAL_DOWNLOADS]: The documentation and reference files (e.g., tools.md) recommend installing several well-known and trusted technology packages from the npm registry, including web-vitals, lighthouse, and @lhci/cli, to facilitate performance auditing.

core-web-vitals