query-vault

Query Vault — Scoped Graph Retrieval

You are answering a question by navigating the Infinite Brain knowledge graph.

Steps

1. Ask: "What would you like to know?"
2. Read _system/INDEX.md — get the full map of existing nodes.
3. Determine scope:
   • Named namespace → prefer nodes from that namespace
   • No namespace → use public nodes; use namespace nodes only if summary clearly matches
   • Explicit private request → use private nodes
   • Never present system nodes as answer content
4. Select likely node types based on question shape:
   • "why" → pillar, decision
   • "how" → playbook, pattern
   • "what if" → hypothesis
   • "what is" → concept, fact
   • "when/where" → event, note
   • "who" → contact
   • open unknowns → question
5. Navigate via edges — do NOT read every node:
   • supports / contradicts → find polarized positions
   • derived_from → trace conclusions to evidence
   • depends_on → find prerequisites
6. Read only nodes whose summary, applicable_when, namespace, and visibility match.
7. Synthesize and output in this format. Write the answer prose, source descriptions, and any saved synthesis node in the vault's document language (see _system/AGENTS.md → Document Language); keep [[node-id]] references, type names, and other structural vocabulary in canonical English:

---

Answer

<direct answer, 1-3 paragraphs>

Sources

• [[node-id]] —

Confidence

<average of source nodes' confidence values, 0.0–1.0>

Related Nodes to Explore

• [[node-id]] — <why it's relevant>

---

Ask: "Would you like to save this as a synthesis node?"

8. Write a log node to logs/log-query-vault-YYYYMMDD-HHmmss.md using the log schema from _system/FRONTMATTER-SCHEMA.md:
   • operation: query-vault
   • affected_nodes: node IDs read during traversal
   • summary: one sentence — the question asked (truncated to 100 chars if needed)
   • Body (30–80 words): question asked, nodes traversed, confidence returned, whether a synthesis node was saved.