Starting Discovery

Starting Discovery takes one prompt. The agent handles the rest of the setup: confirming the system under test, determining which Cycle project to work in, and preparing the questionnaire. This page covers that opening sequence and how to read the hub you land on.

Before you begin, confirm your agent is connected to the Cycle MCP Server and that you are signed in to Cycle. See the Discovery prerequisites.

Starting the flow

Run the begin-discovery prompt from your agent:

/cycle:begin-discovery

The prompt takes one optional value: the system under test. Provide it up front to skip a question:

/cycle:begin-discovery Blue Yonder

Agent hosts surface prompts differently. Claude Code and Cursor list them as slash commands, and other hosts use a prompt picker. See Prompts for how your host presents them.

Confirming the system under test

If you did not name a system, the agent asks before doing anything else.

Blue Yonder proceeds into the flow. Any other answer ends the flow with a message that the system is not yet supported. The agent does not fall back to a general-purpose alternative, and no Discovery work begins.

Resolving the project

Discovery always works inside a Cycle project, because that is where its questions, answers, and output live.

The agent looks in your working folder and its immediate subfolders for a folder containing a .cycproj file. What happens next depends on what it finds:

What the agent finds	What happens
Exactly one project	That project is used. If no user has been recorded for it yet, the agent asks for your username first, then claims the project for you.
Several projects	The agent presents the projects it found and asks you to pick one.
No project	The agent asks for a project name and your username, then creates a project for you.

When it asks for a project name, the agent offers a few suggestions drawn from the system you named and from any naming pattern already used in the folder. Take one of those or supply your own.

The first time Discovery runs in a project, the agent prepares the questionnaire for the system type. Every question starts unanswered.

Reading the input-source hub

Once the project is resolved, the agent shows you the input-source hub. Return here after every piece of work, so you always know what is done and what remains.

The hub reports two things. First, it reports each input source and its status:

Status	Meaning
not started	No work has been done with this source yet.
in progress	Work has begun but is not finished.
done	The source has been worked. Applies to the config scan and to document ingest.
complete	Every applicable question has been answered. Applies to the questionnaire.

Every source starts at not started, including the config scan. Creating the project does not scan anything. The scan runs when you choose it, and not before.

Second, the hub reports whether a test plan can be generated. This stays no until every applicable question has been answered. Even after it changes to yes, nothing is generated until you ask. See Making a Test Plan.

Choosing an input source

There is no required order at the hub, and you can return to any source as often as you like. Ask the agent for the source you want, and it takes you into that work and returns you to the hub afterward.

One ordering is still worth following. Run the config scan before you answer questions by hand. The configuration export is authoritative, so a scan run later replaces answers you have already given. See Scanning a Configuration Export.

Document ingest is independent of the other two sources. It produces test specifications and never changes an answer, so it can run at any point.

Picking from the agent's choices

Discovery asks you to choose from a set of options at three points: picking among the Cycle projects it found, picking an input source at the hub, and answering the questionnaire.

In Claude Code, each of these arrives as a selectable question rather than as a list you reply to in prose. Choose an option and the agent continues with it.

Other agent hosts receive the same options and present them as text. Name the project, source, or answer you want, and the flow proceeds the same way.

The presentation is a request, not a guarantee

Cycle asks the agent to use its question interface where one is available. An agent that asks in prose instead is still driving the same flow, and your answer counts the same.

Example

You say:

/cycle:begin-discovery Blue Yonder

The agent:

1. Confirms that Blue Yonder is supported and continues.

2. Searches the working folder for a Cycle project, finds none, and asks for a project name and your username.

3. Creates the project after you answer, then prepares the questionnaire with every question unanswered.

4. Shows you the hub:

   Input source     Status
   ---------------  -----------
   config scan      not started
   document ingest  not started
   questions        not started
   
   Ready to generate a test plan: no

5. Asks which input source you want to work on.

From here you choose a source and begin.

Working your first input source

• Scanning a Configuration Export: Let the system's own configuration answer what it can, before you answer anything by hand.
• Answering Questions: The questionnaire, and the source a test plan is built from.
• Ingesting Documents: Turn test documentation you already have into consistent test specifications.