The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to manage the local environment, including checking port availability with lsof, starting the application via npm run dev, and ensuring process cleanup with kill and trap mechanisms.
[PROMPT_INJECTION]: The skill processes untrusted content by rendering the application's UI surfaces. While this creates an indirect injection surface if the rendered page contains malicious logic, the risk is minimized through mandatory browser isolation (using a fresh --user-data-dir), headless execution, and the absence of access to the user's primary browser profile or credentials.
[SAFE]: The skill implements significant safety guardrails, including 'Hard Rules' that prevent persistence, daemonization, and the use of third-party services. All evidence is stored locally within the .captain/ directory, which is explicitly excluded from version control.

browser-evidence