The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the browse CLI tool via npm install -g browse to perform its core functions. It also optionally suggests the use of jq for querying data. These are standard tools in the browser automation and data processing ecosystem.
[COMMAND_EXECUTION]: The Node.js scripts use spawn and execFileSync to execute the browse CLI. These calls are implemented using standard practices to prevent shell injection, passing arguments as arrays rather than via a shell string.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted data from external websites (DOM trees, console logs, and network traffic).
Ingestion points: Files generated in .o11y/<run-id>/dom/ and .o11y/<run-id>/cdp/ via snapshot-loop.mjs and start-capture.mjs.
Boundary markers: The captured content is stored in raw formats (.html, .ndjson) without additional boundary markers or sanitization within the skill itself.
Capability inventory: The skill includes file writing and command execution (via browse CLI).
Sanitization: None; the skill records raw data as seen by the browser.
Note: While the skill itself is safe, an AI agent reading the logs generated by this skill could be susceptible to prompt injection instructions embedded in the captured website content.

browser-trace