---
id: "2fd2f809-00c3-4534-9e38-2add0fe89aff"
name: "ChatGPT Clean Transcript"
micro_summary: "Fetch the conversation API, walk current_node ancestry, remove hidden/tool/internal messages, and report the result per the Clean Transcript Query output contract."
status: "resolved"
see_also: []
scope: "Extract clean active-lineage user/assistant Markdown or JSON from an authenticated ChatGPT conversation through browser-harness."
warnings: "ChatGPT’s conversation endpoint is private and may change. Never print or persist its bearer token or put conversation IDs, titles, or transcript contents in the shared skill tree."
---
## ChatGPT Clean Transcript

- Fetch `/api/auth/session`, then `/backend-api/conversation/<id>`; keep the bearer token inside page JavaScript.
- Walk `current_node → parent`, then reverse; do not timestamp-sort the entire mapping because it may contain abandoned branches.
- Keep visible user/assistant text, excluding hidden messages and assistant messages addressed to tools.
- Preserve timestamp, message ID, parent ID, role, and transcript text.
- Emit Markdown/JSON; report the result per the output contract in [[Clean Transcript Query|38ff3f8b-7215-4cfe-940f-10b689d4c824@2ba39fc8-0d4b-4f63-8a35-f198552fbb71]] — that node owns the counting method and field order.
- Verified result: 108 messages, 53,433 tokens. Conversation identity and contents omitted.