Skip to content

Retrieval And Graph Queries

Jumpspace has two retrieval modes: evidence retrieval and graph queries.

Use retrieval when you need likely relevant tasks. Use graph queries when you need exact structural constraints.

Terminal window
npx jumpspace find approval
npx jumpspace find approval review --mode any
npx jumpspace find approval --module project-management --json

find defaults to --mode all so scripts keep strict matching. Use --mode any when recall matters more than precision.

Terminal window
npx jumpspace ask "How does approval work?"
npx jumpspace ask "How does approval work?" --json

ask returns an evidence summary, not an authoritative answer. It should include task IDs, paths, retrieval sources, match reasons, matched terms, unanswered terms, coverage, graph expansion paths, connected tasks, and linked code/tests.

Build the local semantic index:

Terminal window
npx jumpspace semantic build
npx jumpspace semantic status --json
npx jumpspace semantic search approval flow --json
npx jumpspace semantic eval --json

The default backend is deterministic and local. Optional LanceDB and ONNX/Transformers support can be used when local dependencies and models are available.

The differentiating idea is task-vector retrieval plus graph expansion: match the task, then expand across dependencies, refs, modules, spaces, and supersession chains.

Terminal window
npx jumpspace query --depends-on-transitive PM-ROADMAP-001 --no-tests --json
npx jumpspace query --where module=project-management --where tests=none
npx jumpspace query --ref implements:JS-008 --json

Use graph queries for questions like:

  • Which tasks depend on an ADR and have no tests?
  • Which tasks in this module have gaps?
  • Which verified tasks reference a decision?
  • Which tasks are ready but blocked by missing dependencies?
Terminal window
npx jumpspace link suggest DOC-EXAMPLE-001 --since main --json
npx jumpspace link update DOC-EXAMPLE-001 --code src/foo.ts --test src/foo.test.ts --dry-run --json

Changed-file status is context, not enough evidence by itself. Link suggestions should rank candidates by task-intent terms found in paths, basenames, identifiers, phrases, and bounded file content.

Use compact JSON for cheap first-pass orientation:

Terminal window
npx jumpspace find approval --json --compact
npx jumpspace ask "approval flow" --json --compact
npx jumpspace related DOC-EXAMPLE-001 --json --compact