Tasks And The Graph

Jumpspace’s core object is a task. A task is a source-backed unit of implementation memory parsed from Markdown.

What A Task Connects

A task can connect:

• intent from a product doc, spec, ADR, or runbook
• code files
• tests
• dependencies on other tasks
• references to related tasks
• external source links
• acceptance criteria
• durable plan state
• verification records
• known gaps

This lets an agent move from “what is this feature?” to “what files and checks matter?” without guessing from search results alone.

Source Of Truth

Markdown task blocks are source of truth. .jumpspace/index.json is generated by:

npx jumpspace scan

Commit the source docs. Treat generated indexes according to your team’s preference. If you use semantic retrieval, .jumpspace/semantic-index.json is generated and can be rebuilt.

Relationships

Use depends_on when one task cannot be safely executed until another is complete. Use refs for looser typed relationships such as related_to, implements, supersedes, or blocked_by.

depends_on:
  - PM-ROADMAP-001
refs:
  - type: related_to
    id: PM-RUNBOOK-004
    note: Project runbook describes manual validation steps.

Graph relationships make deterministic questions possible:

npx jumpspace query --depends-on-transitive PM-ROADMAP-001 --no-tests --json
npx jumpspace related DOC-EXAMPLE-001 --json

Health Checks

Use audit after metadata edits:

npx jumpspace audit --json

Audit catches broken references, missing linked files, invalid plans, dependency cycles, and completed plan steps without evidence.

Use doctor after mutations:

npx jumpspace doctor --json

Doctor focuses on post-change health and repair suggestions.