The Agent Skills Directory

[SAFE]: The skill implements visual analysis tools that interact with a local Chrome instance via the DevTools Protocol. All actions are consistent with its described purpose.\n- [COMMAND_EXECUTION]: Node.js scripts automate browser control, screenshot capture, and image processing. These commands are localized to the user's environment.\n- [REMOTE_CODE_EXECUTION]: The skill injects the vet.js script into browser tabs to perform structural analysis. The injection is controlled via CDP and uses code bundled within the skill.\n- [EXTERNAL_DOWNLOADS]: The skill relies on standard, well-known Node.js packages (sharp, ws) for its functionality. No unverified external code is executed.\n- [DATA_EXFILTRATION]: Network activity is confined to the local loopback interface for CDP communication. No data is sent to external servers.\n- [SAFE]: Regarding indirect prompt injection: Ingestion occurs through browser DOM reading in vet.js; no explicit boundary markers are required for visual testing; the capability inventory is limited to local file writes and browser interaction; and data is properly serialized using JSON.stringify before being returned to the agent.

vet-generator