The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches a browser process (Chrome, Chromium, or Edge) to analyze the target application. It also starts a local HTTP server to host local files during inspection. These are core functional requirements for runtime diagnosis of web applications.
[DATA_EXPOSURE]: The script scripts/doctor.mjs allows capturing and saving screenshots of the analyzed application to the local filesystem using the --screenshot flag. This involves standard file system write operations for user-requested output.
[INDIRECT_PROMPT_INJECTION]: The skill processes external web content by navigating a browser to user-provided URLs or local HTML files. This creates a surface where a malicious page could attempt to influence the agent through the diagnostic findings.
Ingestion points: Browser navigation to a target URL or local file in scripts/doctor.mjs via the browser harness.
Boundary markers: No specific delimiters are used between the browser-extracted data and the agent instructions.
Capability inventory: The skill can write files (screenshots) and spawn processes (browser).
Sanitization: Diagnostic findings are returned as structured JSON, though the message strings are generated by the logic in assets/scene-doctor.js.
[DYNAMIC_EXECUTION]: The diagnostic script assets/scene-doctor.js is read from the local assets folder and injected into the browser context using Page.addScriptToEvaluateOnNewDocument. This is a standard instrumentation technique for browser-based developer tools.

threejs-scene-doctor