| name | import-conversations |
| description | Import Corinne's ChatGPT conversation history into Eoin's memory. Accepts pasted text, Google Docs links, MD files, or share links. Reflects back what was learned. |
| version | 1.0.0 |
| author | eoin |
| tags | ["import","chatgpt","conversations","history","context","memory","knowledge","transfer"] |
import-conversations — ChatGPT History Import
Purpose
Corinne has 3 weeks of ChatGPT conversations with ideas, use cases, and thinking already formed. These are NOT "files to process" — they're Eoin's head start on understanding her. She should never have to repeat something she already told ChatGPT.
When to Trigger
- During onboarding (Step 5 of the
onboard skill)
- Corinne says "I have some old conversations" or "here's what I talked to ChatGPT about"
- Corinne sends a file, link, or pasted text that looks like a conversation export
Input Formats (accept ALL of these)
1. Pasted Text (lowest friction)
She pastes conversation text directly into Telegram.
- Accept it as-is
- Parse the conversation turns (human vs AI)
- Extract her questions, ideas, and conclusions
2. Google Docs Link
She shares a Google Doc containing copied conversations.
- Note the link for processing when Google Workspace is wired
- If Google access is available: fetch the doc, parse content
- If not yet wired: acknowledge, store link, process when available
3. Markdown Files
She sends .md files (ChatGPT export format).
- Parse the markdown structure
- Identify conversation turns
- Extract key content
4. ChatGPT Share Link
She shares a ChatGPT conversation share URL.
- Note: these may not be directly fetchable from server
- Ask her to copy-paste the key parts if the link can't be read
- Offer alternatives: "Can you paste the important parts, or drop it in a Google Doc?"
5. Screenshots
She sends screenshots of conversations.
- If OCR is available, extract text
- If not, ask her to type or paste the key points
Processing Flow
Step 1: Accept Input
Acknowledge immediately: "Got it — let me read through this."
Step 2: Parse Content
Extract from the conversation:
- Her questions — what she wanted to know
- Her ideas — things she proposed or brainstormed
- Her conclusions — decisions she reached
- Her frustrations — what wasn't working
- Key topics — recurring themes across conversations
- Action items — things she wanted to do but didn't yet
Step 3: Reflect Back
Show her what you learned. This is the trust-building moment.
Template:
I've been reading through your conversations. You've been thinking a lot about:
- [Topic] — [what she explored, in her words]
- [Topic] — [her conclusion or open question]
- [Topic] — [her idea or frustration]
Did I get the picture right? Anything I missed?
Buttons: [Yes, that's it] [You missed something] [Let me add more]
Step 4: Store in Memory
Save extracted knowledge to memory/corinne-prefs.md:
- Update
## Known Interests / Likely First Asks with discovered topics
- Add new expertise areas if revealed
- Note her thinking patterns and communication style
- Store key ideas she wants to pursue
Create a summary in Chartroom if the content is substantial:
- Chart ID:
corinne-chatgpt-import-[date]
- Category:
reading
- Content: key themes, ideas, and conclusions extracted
Step 5: Connect to Action
After reflecting back, connect what you learned to what you can do:
Based on what you've been thinking about, I can start working on [specific thing] right now. Or if you want to keep building on [topic], I'm ready for that too.
Rules
- Use HER words when reflecting back, not system language
- Don't summarize too aggressively — she should recognize her own thinking
- If the import is large, process in chunks and check in between
- Never say "I processed your data" — say "I read through your conversations"
- If something is unclear, ask. Don't guess at her meaning.
- This is a CONVERSATION about her ideas, not a data pipeline
intent: Responsive [I01], Connected [I10]
Escalation
If the import reveals topics that need Robert's input (system capabilities, technical requirements):
- Note them separately
- Queue a bearings question for Robert if needed
- Don't block the import conversation on Robert's answer
Multiple Imports
She may do this more than once. Each time:
- Merge new knowledge with existing, don't overwrite
- Note what's new vs what confirms existing understanding
- Reflect back only NEW insights, not everything again
