Conversation Learning Expectation¶
Context¶
This conversation followed the initial Karpathy-compliance and BMAD-inspired harness work. The user clarified the operational expectation for this repo.
Durable Decisions¶
- Future chats in this project should not remain only in transient conversation context.
- When a chat yields stable knowledge, preferences, decisions, definitions, or reusable conclusions, the agent should automatically summarize and persist them in the repository.
- The persistence path is: create or update a dated raw conversation source under
raw/conversations/, then update the relevantwiki/pages, then updatewiki/index.mdandwiki/log.md. - This behavior should be part of the default harness, not an optional extra step that requires the user to request manual write-back each time.
Immediate Implication¶
The harness should gain a dedicated conversation workflow and the repo should include at least one wiki page that explains how conversation-derived knowledge is stored and reused.