Discovery Sprint

Conductor of Discovery Track defined in docs/discovery-track.md. Job: sequence phases + gate between them — never do specialist work yourself. Each phase runs through facilitator subagent (.claude/agents/facilitator.md) which sequences consulted specialists.

Shared rules (gating, escalation, constraints common to all conductors): ../_shared/conductor-pattern.md.

Pre-workflow entry. Output = chosen-brief.md feeding /spec:idea. Skip sprint if user has brief — recommend /spec:start + /spec:idea instead.

Read first

Start by reading these (contract you enforce):

• docs/discovery-track.md — 5-phase definition + method library.
• docs/adr/0005-add-discovery-track-before-stage-1.md — why.
• memory/constitution.md — Articles II, III, VI, VII apply.
• Active discovery/<sprint>/discovery-state.md (if resuming).

The flow you're driving

Outcomes: go → handoff produce ≥ 1 brief → /spec:start + /spec:idea. no-go → sprint close, no brief. pivot → re-frame or restart.

What you do, step by step

Step 1 — Detect resume vs. fresh start

ls discovery/ 2>/dev/null

For each discovery/<slug>/discovery-state.md with status active, paused, or blocked, list slug | status | current_phase | last_updated. Then batch one AskUserQuestion asking user to pick:

• Resume listed sprint (recommended-first by last_updated).
• Start new sprint.
• Skip sprint, go straight to /orchestrate (when user has brief).

No resumable sprints → skip to Step 2.

Step 2 — Confirm fit (single AskUserQuestion, ≤ 3 questions)

Fresh start → batch into one call:

1. Brief or blank page? — Blank page (run sprint) (Recommended) / Have a brief — go to /orchestrate. Second → exit skill, recommend /orchestrate.
2. Sprint slug — kebab-case, ≤ 6 words. Push back on solution-named slugs ("loyalty-program"), propose outcome-named ones ("q2-retention-discovery").
3. Compression — Standard (5 phases) (Recommended) / Compressed (Frame+Diverge in one sit, Converge+Prototype same day) / Lightning (1 day, Frame and Diverge collapsed). Document choice in discovery-state.md ## Skips if compressed.

Step 3 — Bootstrap (fresh start only)

Invoke /discovery:start <slug>. Creates discovery/<slug>/ + discovery-state.md with all artifacts pending. Don't edit those files yourself.

Step 4 — Confirm specialists in room

Ask user via AskUserQuestion: for each of 6 specialist roles (product-strategist, user-researcher, game-designer, divergent-thinker, critic, prototyper), is there human specialist participating, or should AI agent carry role? Write answer into discovery-state.md's ## Specialists table.

Human specialist participating → AI agent = note-taker + cross-check, not substitute.

Step 5 — Run phases sequentially

For each phase in order:

1. Pre-flight — read discovery-state.md, confirm prior phase complete (or skipped with documented compression).
2. Dispatch slash command (/discovery:frame, etc.). Slash command spawns facilitator, which sequences consulted specialists.
3. Wait for slash command complete + artifact exist.
4. Gate with user via single AskUserQuestion:
   • Continue to <next phase> (Recommended)
   • Pause here — set status: paused in discovery-state.md, exit; resume by re-invoking /discovery-sprint.
   • Re-run <this phase> with feedback (free-text in "Other").
   • Phase 5 only: Hand off (verdict: go) / Close sprint (verdict: no-go) / Re-frame (verdict: pivot).

Step 6 — Handoff

After Phase 5 with verdict go, dispatch /discovery:handoff. Facilitator writes one chosen-brief.md per surviving concept. Then:

• Sprint selected new product or materially changed product positioning → recommend /product:page so public page created/refreshed from chosen brief. Do before or alongside /orchestrate; page = product-facing surface, not implementation stage artifact.
• For each brief, recommend /spec:start <recommended_feature_slug> [<AREA>] then /spec:idea.
• Confirm with user: chain into /orchestrate now or pause.
• Set discovery-state.md status: complete.

Constraints (discovery-sprint-specific)

Generic conductor constraints + escalation pattern: ../_shared/conductor-pattern.md. Specifics for this skill:

• Never open specs/<feature>/ from inside this skill. That happens after handoff via /spec:start.
• Never recommend a sprint when the user already has a brief — send them to /orchestrate instead.

References

• docs/discovery-track.md — full methodology.
• docs/adr/0005-add-discovery-track-before-stage-1.md.
• docs/sink.md — discovery/ sink layout.
• .claude/agents/facilitator.md — agent this skill dispatches.