idea-workbench

Idea Workbench

Overview

Turn local evidence into editable research idea notes. Ideas and distilled conversation notes are knowledge assets, so writes go through Python tools rather than direct hand edits. For read/index work, consult docs/qmd-cli.md and runtime qmd --help.

Available Tools

Tool	Purpose
upsert_wiki_page	Write idea and conversation assets
check_concept_alias	Normalize concept references before writing

Generate Ideas

1. Read the user's research direction from the conversation and settings.json when available.
2. Use qmd query to find relevant papers and concepts.
3. Use qmd get to inspect the strongest local evidence.
4. Use QMD collections or paths to avoid duplicating existing idea or conversation assets.
5. Draft fewer, stronger ideas grounded in explicit local evidence.

Write Idea Notes

Use upsert_wiki_page(page_type="idea", target=<slug>, frontmatter=..., body=...) to create or update an idea note in insights/ideas/. See references/idea-note-template.md for the full frontmatter shape and the recommended body sections (Summary / Local Evidence / Hypothesis / Kill Criteria / Next Step).

Write Conversation Notes

Use upsert_wiki_page(page_type="conversation", target=<slug>, frontmatter=..., body=...) when the useful asset is a distilled conversation takeaway rather than a research idea. Do not store verbatim transcripts. See references/conversation-note-template.md for the frontmatter and body shape.

Curate Ideas

• Use qmd query before creating a new idea.
• If an idea is rejected, parked, or superseded, update idea_state, decision_reason, decision_at, and superseded_by in the note rather than deleting it.
• Keep negative knowledge in the idea note itself.
• Rank active ideas by local evidence, novelty, feasibility, and kill-criterion clarity.
• Offer targeted follow-ups: search related papers, create a core concept page, or capture a conversation insight.

See references/curation-rules.md for state transitions, ranking heuristics, and when to retire an idea permanently.

Rules

• Ground every idea in local paper or concept evidence.
• State assumptions and conflicts explicitly.
• Prefer specific testable ideas over broad speculation.
• Use Python tools for all insights/ideas/ and insights/conversations/ writes.
• Use QMD CLI directly for qmd query, qmd get, qmd status, qmd update, qmd embed -f, and collection/context health.
• Check docs/qmd-cli.md first for the Paper Distill mapping.
• If command details are uncertain, follow runtime qmd --help.

When Not To Write Yet

• If the evidence is thin, keep the output as candidate ideas in chat instead of writing a note.
• If the same idea already exists, update or supersede it rather than creating a near-duplicate file.
• If the user only wants exploration, do the retrieval and ranking work first, then ask whether any candidate should be saved.

Skill / Tool Contract

This skill is responsible for research judgment, evidence selection, concise idea framing, and curation decisions.

Python tools are responsible for deterministic writes, path safety, and structured errors.