Handbook

Task `rag_query_plan` v1

Decide whether retrieval is needed for a question in context of facts, and emit concrete queries (and optional reasoning) for a downstream retriever—without performing retrieval itself.

Updated

Task rag_query_plan v1

Summary

Decide whether retrieval is needed for a question in context of facts, and emit concrete queries (and optional reasoning) for a downstream retriever—without performing retrieval itself.

Inputs

Field Type Required Notes
question string yes User or synthesized question.
facts object yes Known structured facts (may be {}).
temperature number no Default 0.05.

Output

JSON aligned with contract.json: includes retrieval intent, ranked queries, and confidence fields per sidecar.

Examples

{
  "question": "Do we need docs on operators?",
  "facts": { "repo": "forge-lcdl" }
}

Failure modes

  • SchemaFailure — missing required fields in or out.
  • ParseFailure — invalid assistant JSON.
  • GatewayFailure — model/provider errors.

Verification

Contract schema validation on input/output; optional deterministic query shape checks in task code.

Changelog

  • v1 — Initial RAG planning task.