goop-discuss

/goop-discuss

Start the discovery interview. Capture vision, must-haves, constraints, out-of-scope items, assumptions, and risks before planning.

This command ALWAYS starts a brand-new workflow. Never reuse or append to the current or last active workflow. Every /goop-discuss invocation creates a fresh workflowId, a new git branch, and ultimately a new PR.

Immediate action

Load the interview protocol first:

goop_reference({ name: "discovery-interview" })

Do not load any documents (via goop_boot, goop_read_db, or any other tool) at the start of this command — no spec, blueprint, chronicle, or requirements from any workflow. This is a state-only boot; a brand-new discovery has no use for prior planning content. The reference load above is fine — only document loads are forbidden.

Steps

1. Create and activate a new workflow in a single call — do not reuse the active one, and do not split this into a separate create-workflow + set-active-workflow pair. Infer a kebab-case workflowId from the user's prompt (or use the supplied workflow-id argument), then:
   • goop_state({ action: "create-workflow", workflowId: "<new-id>", activate: true })
   • Trust a successful response and move to step 2 immediately. Do not regenerate the ID or repeat this call — see references/core-protocol.md §Trust Successful Tool Results for why repeating an already-succeeded state call is a bug, not a safety measure.
2. Checkout a new git branch named after the workflowId before any file writes:
   • git checkout -b <workflowId>
   • All implementation work for this workflow happens on this branch. Agents must never switch branches mid-workflow or work across multiple branches simultaneously.
3. Ask all seven discovery categories (vision, must-haves, constraints, out-of-scope, assumptions, risks, and atomic PR preference) directly. Use the question tool for structured answers; mark exactly one option (Recommended).
4. Probe until each category has specifics. Empty must-haves, out-of-scope, or risks are invalid.
5. Summarize and confirm with the user.
6. Write REQUIREMENTS.md via goop_write_db({ doc_type: "requirements", content: "..." }) and call goop_state({ action: "complete-interview" }). The tool renders the markdown sidecar automatically.

   REQUIREMENTS.md must include a ## Atomic PR Strategy section.

7. Remind the user that with Atomic PRs, one PR is opened per wave during /goop-execute, and /goop-accept will offer to merge the full stack in order.
8. Suggest /goop-plan. Note: /goop-plan now begins with a research-first gate — the orchestrator auto-dispatches goop-researcher (and optionally goop-explorer) before delegating to the planner. Trivial workflows (≤ 2 files, no domain unknowns) may skip the research pass, with the decision logged to ADL.