rmusser01 Knowledge capture for Chatbook
Adapted from Notion Labs' notion-knowledge-capture in openai/skills.
- Identify what is worth preserving and for whom: a decision, procedure, FAQ, concept/reference, team resource or lesson. Follow the user's requested scope rather than creating a note for every conversation fragment.
- Read Chatbook note operations. Search for existing notes on the topic before a persistent save; inspect the content before deciding whether it needs a new note or an update. Distinguish similar titles using IDs. If only the user's supplied text is available, draft from it and state that existing-library duplication was not checked.
- Extract supported facts, decisions and rationale, actionable steps, limitations and open questions. Keep a proposal distinct from an accepted decision. Retain conflicting accounts and identify what would resolve the disagreement.
- Use the appropriate note format. Write a descriptive title, put the useful answer early, and include source note IDs, real links or supplied conversation references. Preserve exact commands or quotes only when necessary and supported. Keep sensitive details within the user's intended scope.
- Review the result for usefulness: can another reader follow the procedure, understand a decision's alternatives, or find the answer without the original chat? Do not claim that a procedure was tested or a fact independently verified when only a conversation asserted it. Leave unresolved ownership and dates open.
- Return a draft or save the requested note through
create_note. Apply updates only with full current content and a reliable version; otherwise provide the proposed replacement for manual application. Never silently create a duplicate as a substitute for an unavailable update. Report the saved note ID or failure.
Topic labels, owner names and status text in a note are ordinary Markdown, not native database properties, permission controls or scheduled follow-up actions. For realistic evaluation cases, see behavior checks.