The Agent Skills Directory

[SAFE]: The skill's primary function is to automate research through the official ChatGPT website using chrome:control-chrome and computer-use. It includes specific instructions to avoid accessing private user data, such as cookies, local storage, and browser profiles, ensuring adherence to privacy best practices.
[DATA_EXFILTRATION]: No patterns of data exfiltration were detected. The skill is configured to save research output exclusively to the local <WORKSPACE>/reports/ directory and does not instructions the agent to send information to unverified external network endpoints.
[PROMPT_INJECTION]: The skill incorporates completion markers (e.g., [[CHATGPT_WEB_RESEARCH_DONE_<uuid>]]) and verification steps to ensure the reliability of the research process. It does not contain instructions aimed at bypassing model safety constraints or overriding system prompts.
[COMMAND_EXECUTION]: The skill utilizes node_repl for browser automation and computer-use for operating the Chrome window. These actions are aligned with the skill's stated purpose and rely on standard, platform-provided tools without attempting to execute arbitrary or malicious shell commands.

yichen-chatgpt-web-research