The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from the internet and external APIs, creating a vulnerability surface for malicious instructions embedded in third-party data.\n
Ingestion points: Untrusted content from rendered Chromium pages and API responses from the Wayback Machine (archive.org).\n
Boundary markers: Explicit instructions to "Inspect only pages relevant to the task" and suppress unrelated tab titles.\n
Capability inventory: Broad browser control capabilities including evaluate_script, click, and fill_form.\n
Sanitization: Requirements to summarize large evaluate_script outputs, paginate high-volume console/network logs, and filter results.\n- [EXTERNAL_DOWNLOADS]: The skill fetches capture data from the Internet Archive's Wayback Machine (archive.org) via curl for discovery and replay research. This interaction targets a well-known service and is documented neutrally.\n- [COMMAND_EXECUTION]: Utilizes system utilities including curl for API requests to local and remote endpoints, and plutil for verifying Chromium version metadata on macOS as part of its troubleshooting workflow.

chromium-browser