Skip to content

Getting Started

This guide gets a repo from “no graph” to “agent can ask and orient” with the smallest useful setup.

Start with Why Jumpspace? if you want the short version of the problem it solves: source-controlled intent, agent-readable structure, and honest evidence before broad code search.

  • Node.js 20 or newer.
  • A Git repository.
  • Markdown docs that describe product behavior, architecture, runbooks, specs, or decisions.

Use Jumpspace as a dev dependency:

Terminal window
npm install -D jumpspace

If you are testing a local checkout before publishing, run this from the Jumpspace repo:

Terminal window
npm run build
npm link

Then run this inside the target repo:

Terminal window
npm link jumpspace
jumpspace release install-doctor --json

release install-doctor tells you which binary your shell is actually using and whether it looks stale.

For an existing repo, start with automatic discovery:

Terminal window
npx jumpspace init --auto

This inspects common documentation locations and writes .jumpspace/config.json with practical Markdown globs. Plain jumpspace init keeps the conservative starter behavior.

Add repo-local agent guidance:

Terminal window
npx jumpspace add-skill --codex
npx jumpspace add-skill --claude

These commands are additive. They create missing guidance files or append Jumpspace-managed instructions without overwriting your existing agent guidance.

Scan turns Markdown task blocks into the generated local index:

Terminal window
npx jumpspace scan

The generated index is useful for fast reads, but the source of truth is still your Markdown.

Use find when you know the words you are looking for:

Terminal window
npx jumpspace find approval
npx jumpspace find approval review --mode any

find defaults to strict all-term matching. Use --mode any for broader recall.

Use ask when you want an evidence summary:

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

ask is retrieval, not authority. It should show task IDs, paths, match reasons, coverage, unanswered terms, and linked code/tests. If the evidence is weak, it should say what was not answered.

When a task is approved and has a valid plan, use:

Terminal window
npx jumpspace work DOC-EXAMPLE-001 --json

That packet includes task intent, code and test links, acceptance criteria, current plan state, next unblocked steps, verification records, guardrails, schema names, and optional drift facts.

The normal loop is:

Terminal window
npx jumpspace scan
npx jumpspace audit --json
npx jumpspace work DOC-EXAMPLE-001 --json
# implement the step
npx jumpspace step complete DOC-EXAMPLE-001 design --evidence "Human approved the design."
npx jumpspace verify DOC-EXAMPLE-001 --check "npm test" --criteria AC-1 --json