The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates browser control by executing local commands through the 'cli-jaw' and 'agbrowse' binaries. It also includes an 'evaluate' command that allows for the execution of JavaScript directly within the browser context to perform diagnostics or extract data.- [EXTERNAL_DOWNLOADS]: The documentation references several external tools and libraries required for full functionality, including 'cliclick' for mouse control, 'camoufox' for stealth browsing, 'playwright-core', and 'yt-dlp' for media processing. It also identifies 'r.jina.ai' as a service for reading and simplifying web page content.- [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by design as it fetches and interprets content from external websites. * Ingestion points: Untrusted data enters the agent context via 'cli-jaw browser fetch', 'navigate', 'text', and 'snapshot' commands in SKILL.md. * Boundary markers: There are no explicit delimiters or specific instructions provided to the agent to ignore potentially malicious instructions embedded within the fetched web content. * Capability inventory: The skill has access to subprocess execution ('cli-jaw'), browser-side JavaScript execution ('evaluate'), and filesystem writes for saving screenshots. * Sanitization: The skill does not implement any sanitization or filtering of the content retrieved from the web before processing it.

browser