SwarmCraft
Discovery and PlanningGuide

Docs

Use Deep Discovery

Understand the recommendation-led journey from local evidence and an early prototype to one reviewed Discovery Package and initial project board.

Use Deep Discovery

Deep Discovery is the evidence-led SwarmCraft route for an operation whose boundary cannot be established safely through a focused guided intake. It runs in your local VS Code repository before a project or Project Unlock exists.

Every discovery starts with four brief questions asked one at a time, followed by an invitation to add any useful source material. Your chosen local AI provider distributes each answer across the affected document drafts, researches the context you authorize, proposes a direction across nine discovery areas, and normally gives you an early disposable prototype to correct by exception. The nine Markdown documents remain the reviewed record, but they are a coverage map rather than nine consecutive interviews. SwarmCraft tooling then prepares one exact text package without calling local AI. You preview and authorize that package before anything is uploaded.

VS Code showing the expanded SwarmCraft Deep Discovery sidebar at the Gather sources stage and the local discovery guide

Choose Deep Discovery for the right job

Use Deep Discovery when:

  • important knowledge is distributed across people, documents, or repositories
  • actors, rules, exceptions, records, or controls need structured investigation
  • sources disagree and the owner needs to resolve the reviewed position
  • the operation may need to own bounded records and requires migration or continuity reasoning
  • the owner wants an exact review and consent boundary before server processing

If a focused guided intake can establish the workflow and the existing system of record will remain, use Fast Start. The routes use the same Project Unlock and create the same project deliverable.

The Deep Discovery journey

  1. Start or resume Deep Discovery in one trusted Git workspace.
  2. Answer the four progressive opening questions, then add and select any supporting sources.
  3. Review the proposed direction and early prototype, correct what is wrong by exception, then review the consolidated discovery documents.
  4. Prepare and authorize the Discovery Package.
  5. Continue through Project Unlock and open the resulting project.

Use Deep Discovery troubleshooting if eligibility, provider, source, package, upload, or generation checks stop the journey.

What stays local and what is uploaded

Local AI works only with the material you make available in your repository and provider session. Package preparation is deterministic and does not ask Codex or Copilot to generate tasks, identifiers, assertions, or a hidden Blueprint.

The upload contains:

  • all nine curated documents, each reviewed or explicitly marked NOT_APPLICABLE
  • only the selected .md or .txt supporting sources and valid Markdown sidecars
  • package metadata, review state, sizes, digests, and consent bound to the immutable package digest

Raw binary files, unselected repository files, unrelated working-tree content, and local paths are not uploaded. Review the Privacy Policy for the server-processing and retention contract.

What SwarmCraft creates

After the admitted package continues through Project Unlock, SwarmCraft privately interprets the reviewed content on the server and creates one project with its initial board. The immutable package is retained as project-level provenance. Deep Discovery does not promise task-level source citations, recurring monitoring, CI ingestion, drift alerts, or a paid regeneration loop.