The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses the /api/auth/session endpoint on chatgpt.com to verify user login status. It contains strict instructions to only use the boolean result of this check and explicitly forbids displaying or storing sensitive session fields, tokens, or cookies.
[DATA_EXFILTRATION]: The skill allows the agent to upload up to 10 local files to the ChatGPT interface. It mandates preflight checks for absolute paths and file existence, ensuring files are handled through standard BrowserOS MCP mechanisms.
[COMMAND_EXECUTION]: The skill uses JavaScript execution within the browser (via BrowserOS MCP) to manipulate the DOM for file uploads and to perform session checks. These actions are limited to the chatgpt.com domain.
[PROMPT_INJECTION]: The skill processes data from the ChatGPT web interface (assistant responses). As it ingests external content and has capabilities like file access and browser automation, it represents an indirect prompt injection surface. However, the risk is minimized by the skill's specific task orientation and instruction to use prompt-authoring discipline.

chatgpt-web