vac-knowledge-baselisted

# VAC Knowledge Base Draw the prompt-vs-KB boundary, then author the KB. Output: `knowledge-base.md` + a tier-1 spec handed to `vac-prompt`. **Position in the pipeline:** runs after `vac-design` (which flags whether a KB is needed) and **before** `vac-prompt`. This way the prompt writer puts only tier-1 substance into the Answer Bank and adds a "knowledge base available" hook; tier-2 is never duplicated in the prompt. **This skill does NOT deploy.** Upload + RAG-index + attach belong to `vac-deploy` (scripts in `vac-elevenlabs-api`). ## When a KB is warranted (the gate) A KB is for the **long tail of substance** an engaged decision maker might ask for. Build one only when BOTH hold: - The agent reaches an **engaged decision maker** with deep/varied questions (inbound-qualify, consultative outbound, any agent selling its OWN product). Same condition as the Answer Bank. - There is **substantial substance beyond tier-1**: several concrete examples, success stories/numbers, detailed how-it-works, company background, edge-case objections, secondary use-cases. **Skip the KB** for pure transactional/capture agents, gatekeeper-only probes, or agents whose whole substance fits comfortably in the prompt. Answer Bank without a KB is fine — tier-1 alone can be enough. The KB is the *subset* where the long tail justifies RAG. ## Source material Company-info KBs source their tier-2 detail from **`vac-company-research` (Step 4b: KB-Detail)** — it gathers the prospect-level detail (