The onboarding conversation completes and saves to onboarding.yaml, but the world model extraction is never triggered.
What needs to happen:
- Call
world_model_manager.build_from_onboarding() at the end of onboarding
- Implement actual deviation extraction using Claude (currently a stub)
Expected output:
~/.chevin/world_model.json with structured deviations (epistemic, motivational, relational, etc.)
~/.chevin/sources/ with source documents