Wb Query

2. Temporal Query (Resolve)

For questions about state at a specific time.

Can also combine with structured query: $WB query --type person --at "SA:200" resolves temporal state for all results.

3. Graph Query (Relationships)

For questions about connections between entities.

4. Spatial Query

For questions about geographic containment and hierarchy.

5. Semantic Query (Vector)

For fuzzy/conceptual questions where exact terms are unknown.

Steps

1. Parse the Question

Read $ARGUMENTS or the user's natural language question. Identify:

• What type of information they want (entities, relationships, state-at-time, spatial)
• Any filters (type, tags, name, time, location)
• Whether this needs one query or multiple combined queries

2. Execute Queries

Run the appropriate $WB commands via Bash. Always use --vault <vault_root> and --pretty for readability.

For complex questions, chain multiple queries:

• "Who in the Iron Covenant was alive during the Great War?" -> query faction members, then resolve each at the war's time period.

3. Enrich Results

For each result, consider reading the full entity file with the Read tool if the user needs detail beyond what the query returns (which is metadata only).

4. Format and Present

Present results in a clear, readable format:

• For entity lists: table or bullet list with name, type, key attributes
• For temporal queries: timeline format with periods and events
• For graph queries: relationship map showing connections
• For spatial queries: hierarchy tree

5. Suggest Follow-ups

After presenting results, suggest 2-3 follow-up queries the user might find useful. Examples:

• "You could explore [Entity]'s relationships with /wb-query related to [Entity]"
• "To see how this changes over time, try /wb-query [Entity] at SA:300"
• "For a deeper network view, try /wb-query connections of [Entity] within 3 hops"

Query Strategy

Each flag maps to a specific store:

• --name, --type, --tags, --within -> SQLite. Fastest path to known entities and structured filters.
• --related-to, --hops -> Graph. Path-finding and neighbor traversal.
• --semantic -> Vector (OpenAI text-embedding-3-large with hierarchical chunking). Best for thematic, creative, cross-type, and section-precision queries.
• resolve --at -> SQLite. Timeline data lives in frontmatter, indexed in SQLite.
• spatial -> Graph. Parent/child containment is a graph operation.
• For best results on open-ended questions, use --semantic first, then --related-to on the top results.
• Complex questions: Chain stores. Vector for discovery, SQLite to filter, graph for connections, file read for detail.

Tips

• If a query returns no results, try broader terms or semantic search as fallback.
• Combine query types for complex questions (structured to find candidates, then resolve for temporal detail).
• The --text flag does full-text search across all content, useful when you are not sure which field contains the answer.

Content Integrity

• Report only what exists. If a query returns no results, say so clearly. Never fill gaps with plausible-sounding lore or "educated guesses" about what the world might contain.
• Distinguish data from inference. When presenting results, the entity data is canon. Any commentary you add (e.g., "this suggests X might be related to Y") must be clearly framed as your inference, not established fact.
• This skill is read-only. Do not create, update, or delete entities. If the user's question reveals a gap in the world, suggest using /wb-create or /wb-suggest — do not fill the gap yourself.