Skip to main content
Version: 2.29

Resuming Discovery

Discovery is built to be put down and picked back up. Nothing is held in your agent session, so closing your agent, restarting your machine, or stepping away for a week costs you no progress.

Picking up where you left off

Run the same prompt again from the same folder:

/cycle:begin-discovery

The agent finds the existing Cycle project, recognizes that Discovery has already been started, and returns you to the input-source hub with each source's status intact. There is no separate resume command, and you are never asked to start over.

Answered questions stay answered. An ingest that has run stays done. Whatever you had not finished is still waiting.

Where Discovery keeps its state

Discovery stores its state in a .cycle folder inside your Cycle project:

FileWhat it holds
discovery.dbYour questionnaire answers, including who supplied each one.
discovery_metadata.jsonWhich input sources have been worked, and how far.

The two are kept separate on purpose. Answers and the record of what work has been done are different concerns, so running document ingest can never disturb the answers you gave, and revisiting answers can never make Discovery forget that an ingest happened.

Sharing Discovery with your team

Commit the .cycle folder

Commit .cycle to source control along with the rest of your Cycle project. A teammate who pulls the project opens the hub at exactly the state you left it and can carry the work forward.

If you leave .cycle uncommitted, the Discovery work exists only on your machine. Anyone else who starts Discovery in the project begins from an empty questionnaire.

Discovery is often a team activity, because no single person usually knows every answer. Committing the state is what makes that division of labor practical.

Adding an input source later

A generated test plan is not the end of Discovery. Return to the hub whenever new material arrives or your understanding improves:

  • New documentation: Add it to Customer-Docs and run document ingest again.
  • Better answers: Revisit the questionnaire as you learn how the operation actually works. See Answering Questions.
  • A refreshed plan: Generate again after either of the above. See Making a Test Plan.