The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell scripts (scripts/run-iteration.sh) and calls the claude CLI and cdp.mjs driver to perform browser automation tasks. These commands are integral to the skill's purpose of automating browser navigation and are scoped to an isolated Chrome profile.
[EXTERNAL_DOWNLOADS]: The skill references and downloads configuration/guidelines from the chrome-cdp sibling skill and potentially external URLs if defined in a task. These operations target local workspace directories or are part of the intended automation workflow.
[SAFE_PRACTICE]: The skill explicitly uses an isolated Chrome profile (~/.chrome-debug-profile) to ensure that the automation loop does not interact with or expose the user's primary browser data, cookies, or sessions.
[INDIRECT_PROMPT_INJECTION]: As a browser automation tool, the skill ingests data from external websites. It includes instructions for the inner agent to verify page state using snap and eval rather than blindly following page content, and the design relies on the outer agent (the user/main session) to manually review traces, which serves as a human-in-the-loop security boundary.

autobrowse-cc