Chatgpt Api

Overview

Use this skill to operate against the authenticated ChatGPT web app backend, not the public OpenAI API. Prefer this skill when a task specifically depends on chatgpt.com web behavior, existing ChatGPT browser auth, hidden upload controls, or the exact backend-api/f/conversation payload shape.

The safest execution mode is browser-first: run requests from an already logged-in ChatGPT tab with agent-browser or equivalent browser automation. Direct HTTP replay is possible only when you already have the browser-derived auth/session material and you accept that the web app's auth layer may change.

Quick Start

1. Reuse an authenticated ChatGPT browser session.
2. For uploads, use the hidden file input #upload-photos for images or #upload-files for general files.
3. Build the f/conversation payload with scripts/build_conversation_payload.py when you need a raw JSON body.
4. Send the request to https://chatgpt.com/backend-api/f/conversation.
5. Poll conversation/{chat_id}/stream_status for normal text/reference-image generation flows.
6. Poll conversation/{chat_id}/async-status for some longer-running image edit flows.
7. Fetch rendered files with /backend-api/files/download/{file_id} or signed /backend-api/estuary/content?....

Browser-First Workflow