memory-setuplisted

# memory-setup — gather info + wire everything up You are the setup wizard. The user just cloned the kit and wants it working with their data. Your job: ask the right questions, gather the right info, write the right config files, generate the right scheduled tasks, run the first cycle. **Everything subsequent runs need has to be collected during this skill** — don't assume they'll come back to fill in gaps. ## The bootstrap checklist (track + tick every item) ``` [ ] 1. Python 3.10+ available [ ] 2. Kit cloned + importable [ ] 3. Tier picked (Lean / Full) [ ] 4. Org name + vault-owner-name gathered [ ] 5. Vault scaffolded (memories/, entities/, .mvkit/, profile.json) [ ] 6. Vault-owner entity created (vault_owner: true) [ ] 7. PROBE: which MCPs do they have installed? [ ] 8. ASK: which sources do they want included [ ] 9. ASK per-source config (or auto-detect — recommended) [ ] 10. Write `.mvkit/connected_sources.json` [ ] 11. EVAL-FIRST: generate ~30 questions from the current Claude session's understanding of the user + org + sources. Write to <vault>/evals/retrieval/questions.jsonl. This is the acceptance criterion the ingest is trying to satisfy. [ ] 12. INGEST LOOP: spawn parallel sub-agents per source with a real backfill window (default 60 days). After each batch lands, re-run the eval. Continue until soft coverage >= 0.6 OR sources exhausted OR hard cap hit. This is the BIGGEST ingest the kit will ever do — not 5-10