Idea mining skill

Goal

• •Extract reusable, ATOMIC ideas from an owned transcript into ideas/*.md.

Inputs

• •Document id doc_YYYYMMDD_xxxxxxxx with vault/owned/<doc_id>/transcript.md.
• •Optional max idea count.

Outputs

• •New ideas/idea_YYYYMMDD_xxxxxxxx.md files with:
  • •sources pointing to the doc id and a locator
  • •short statements and tags
• •Updated catalog/* and exported site pages.

Procedure

1. •

   Run the miner:

   • •python tools/townctl.py mine-ideas --doc-id <doc_id> --max-ideas 12 This uses the configured LLM backend (OpenAI/Anthropic) via environment variables:
   • •TOWN_LLM_BACKEND=openai|anthropic
   • •OPENAI_API_KEY or ANTHROPIC_API_KEY
   • •optional: TOWN_LLM_MODEL
2. •

   Regenerate indexes:

   • •python tools/townctl.py index
   • •python tools/townctl.py export-site
3. •

   Validate:

   • •python tools/townctl.py lint

Guardrails

• •Do not invent claims; every idea must be grounded in the transcript.
• •Keep each idea atomic (one claim/definition/theorem).
• •For reference-only sources, do not include long quotes.