recipe-agent-team-design-interview

name	recipe-agent-team-design-interview
description	Deep design interview that produces a design brief. Socratic 4-line probes lock the design's output description, tensions, assumptions, priorities, and failure modes. Output: design-brief.md. Skip for backend architecture, terminology cleanup, fuzzy planning, implementation, review, post-run learning, or rendered assets. Hand off to recipe-agent-team-design-spec for artifact production.
metadata	{"version":"2.0.0","openclaw":{"category":"recipe","domain":"agent-orchestration"},"requires":{"bins":["agent-team"],"skills":["agent-team-shared","agent-team-run","agent-team-task","recipe-agent-team-worker-checkpoint"]}}

Agent Team Design Interview Recipe

PREREQUISITE: Load agent-team-shared before any CLI command. agent-team-run and agent-team-task are required inside an agent-team run.

Worker integration: Inside an agent-team task, the worker must follow recipe-agent-team-worker-checkpoint for sync check, inbox handling, task start, and completion. This recipe defines brief generation only.

Output Description

This recipe does not enumerate design types. The brief's Output field is plain-text and authoritative. Examples:

• "Solo-fitness app main dashboard screen — onboarded-user view, mobile-first, light/dark themes."
• "Heritage-tone brand identity — logotype, symbol, palette, usage rules for digital + print."
• "Reusable token catalog (DESIGN.md alpha) shared across product UI and marketing surfaces."
• "Playable NPC character — combat role, skills, animation states, and visual style guide."
• "Level 1 zone layout — set dressing, lighting beats, traversal flow."

Long, concrete descriptions beat one-word categories.

recipe-agent-team-design-spec/references/ ships non-binding patterns (ui.md, logo-branding.md, design-system.md, character.md, environment.md, icon-illustration.md). Cite relevant ones in the brief's Pattern Hints field; the spec recipe may compose, adapt, or ignore them.

Artifact Target

With RUN_ID, write the brief to _workspace/{run_id}/design/{slug}/design-brief.md. {slug} is a short kebab-case identifier captured during Output Capture (e.g., dashboard, heritage-brand, token-catalog, npc-roster-v1, zone-1).

Without RUN_ID, return the same brief structure inline and state durable files were skipped.

Pipeline

Uncertainty Scan → Discovery Phases → Synthesis Gate → Brief.

Discovery

Use the smallest depth that still produces an actionable brief.

• Fast path: the brief's minimum viable fields exist upstream: Output, Core Story, Priority Ranking, Constraints, and Success Criteria. Score covered phases 0 and probe only remaining applicable gaps.
• Full path: request is ambiguous, multiple workers need a durable source of truth, or decisions have meaningful tradeoffs.

Every user-facing probe is a single 4-line block. Korean labels exact; values in the requester's language.

현재 이해: <one-sentence summary of what is decided so far>
막힌 결정: <single biggest uncertainty right now>
추천 답안: <concrete starter answer the requester can react to>
질문: <one concrete question>

First probe convention: when no decisions exist yet, write 현재 이해: 인터뷰 시작 (아직 결정 없음).

Rules: one question per turn; no compound questions; 추천 답안 is required; restate 현재 이해 after each answer; do not ask what local context already answers; log every executed probe in the brief. See references/probe-toolkit.md for technique-to-phase mapping.

Uncertainty Scan

Inspect task body, planning-grill.md or plan.json, terminology.md, prior design artifacts, and the opening request.

Score phases as 2 no capture, 1 partial signal, 0 fully covered upstream. Emit the scan as a Probe Format block with 추천 답안 set to the ordered phase list. Confirm before starting unless the requester explicitly asked for a quick standalone draft. Record any later phase reordering or reopened gap as an Interview Log re-scan trigger; otherwise record none.

The first phase is Output Capture unless the upstream task body already contains an unambiguous one-paragraph output description (score 0 and cite the source).

Discovery Phases

• Output Capture — Output description, {slug}, Pattern Hints, brief Subject summary. Use literal none when no Pattern Hint applies. If output stays vague after two probes, hand off to recipe-agent-team-planning-grill.
• Context & Story — Core Story, audience, problem-feeling, emulate/avoid references. If no concrete user or moment can be named, run planning grill.
• Specificity Drill — concrete signals, earn-it examples, fake-it examples, rejected adjectives.
• Trade-off Surface — tensions, chosen side, accepted cost, hard constraints.
• Assumption Probe — at least three assumptions with evidence, breakage impact, risk flag, upstream input citations.
• Forced Choice & Priority — priority ranking, First 5 Seconds, success criteria.
• Failure Rehearsal — three failure modes with root cause, earliest signal, mitigation seed.

Probes adapt to the captured Output. Pattern Hints name likely artifact patterns only; do not load pattern files during interview, and let the Output description override hint choice.

Synthesis Gate

After phases, deliver one final Probe Format block:

• 현재 이해: 3-5 sentences covering output, story, priority, tension, risk
• 막힌 결정: "brief 잠그기 전 빠진 부분?"
• 추천 답안: "없음 - 진행"
• 질문: "이대로 brief 잠그고 design-spec으로 넘어갈까?"

Only proceed after explicit confirmation, unless the requester asked for a quick standalone draft. Record confirmation as either the quote or quick-draft requested.

Brief Contract

Use references/brief-template.md. Empty sections are forbidden; record a refusal, risk, or upstream citation instead.

Acceptance

1. Brief exists at _workspace/{run_id}/design/{slug}/design-brief.md, or the final response contains the same structure with no run.
2. Brief's Output section contains a one-paragraph description plus {slug} and Output Path.
3. Every applicable phase has a capture, refusal, upstream citation, or skip.
4. Interview Log records gap scores for applicable phases plus skip reasons.
5. Synthesis Gate confirmation is recorded as a requester quote or quick-draft requested.
6. When running as an agent-team task, call agent-team task complete with --artifact set to the brief path and --evidence citing the brief path plus confirmation state.

A brief with only refusals does not pass. Minimum: explicit Output, constraints, success criteria — captured or upstream-cited.

Do Not

• Load pattern reference files; cite their paths only.
• Produce artifacts (this recipe ends at the brief).
• Bypass Output Capture without a 0 score and upstream citation.
• Batch multiple user questions per turn.
• Force the Output into a single pattern label when the description spans multiple shapes; record all relevant hints instead.

Hand Off

Situation	Hand off
Brief locked, artifacts needed and RUN_ID exists	recipe-agent-team-design-spec with brief path
Brief locked, artifacts needed but no RUN_ID	Ask orchestrator to create a run and materialize the inline brief before spec
Output stays vague after two probes	recipe-agent-team-planning-grill
Term ambiguity blocks Output Capture	recipe-agent-team-terminology-context
Backend gap surfaces	recipe-agent-team-architecture-design
Multi-output effort (e.g., shared tokens + consuming surface)	Re-run this recipe per Output; cite earlier briefs as Upstream Inputs