Design Orchestrator

Product Design & Prototype Orchestrator

Core Principles

1. Design is trade-off, not accumulation -- Core paths must be extremely smooth; non-core paths can be compromised
2. Upstream quality determines downstream efficiency -- PRD quality gates cannot be bypassed; garbage in, garbage out
3. Design consistency is a system property -- From tokens to components to interaction specs must be coherent throughout; breaks equal debt
4. Bidirectional feedback loop -- PM->UI is forward constraint, UI->PM is reverse feedback; both together ensure design quality

Exception Handling

| Exception Type | Handling Strategy | |----------|----------| | Sub-Skill output file missing | Block current stage, output missing items list, prompt human to supplement upstream input | | Sub-Skill quality check not passed | Block entry to next stage, output failed items details, prompt human to confirm whether to fix or accept risk | | Context approaching limit | Prioritize current stage content, summarize completed stage outputs as key conclusions written to file | | Human decision timeout without response | Pause orchestration flow, preserve current state, wait for human decision before continuing | | Upstream input data format error | Attempt compatible parsing; if parsing fails, degrade to user-provided description, annotate "data format error" | | Stage summary generation failure | Generate partial summary based on completed sub-Skill output, mark missing items as "data missing", do not block orchestration completion |

Standalone Usage Input Acquisition Strategy

Standalone Trigger Scenario Identification

When this orchestrator is invoked directly (not through a parent orchestrator), it is considered a standalone trigger scenario. Typical trigger methods:

• User directly requests capabilities within this orchestrator's domain
• Triggered as an independent skill by external systems
• Parent orchestrator not executed, but user only needs this orchestrator's capability

Required Input Acquisition Strategy

| Required Input | Priority: Read from output/ | Fallback: Get from user conversation | Last Resort: AI knowledge base inference | |---------------|---------------------------|-------------------------------------|---------------------------------------| | PRD (prd.md) | Read output/pm-design/design-prd/prd.md | Ask user for PRD document or verbal requirements | Infer requirements from user description (low confidence, mark "PRD is AI-inferred") | | project_dir | — | Ask user for project directory path | Cannot infer, user must provide |

Upstream Orchestrator Auto-Backtracking

When critical required inputs are missing, suggest user execute upstream orchestrators in the following priority:

| Missing Input | Suggested Upstream Orchestrator | Description | |--------------|-------------------------------|-------------| | PRD | pm-design related orchestrator | PRD is the business basis for design-orchestrator, missing will result in execution without business foundation | | Sub-skill outputs | Sub-skill execution | Sub-skills (design-userflow, design-ia, change-impact-analysis...) produce domain-specific outputs |

Backtracking suggestion output format:

Critical input missing detected, suggest executing upstream orchestrator first:
1. [Priority] pm-design related orchestrator -> Produces PRD
Continue with AI-inferred values? (Inferred values confidence =0.5 | When confidence  project_dir validity -> Input confidence assessment

## Orchestration Protocol

The orchestration protocol follows the unified standard in [orchestrator-protocol.md](../../../codex-templates/orchestrator-protocol.md).

## Pipeline

```yaml
pipeline: design-orchestrator
version: 10.1
post_pipeline:
  - action: stage-summary
    output: output/phase-reports/pm-design/design-orchestrator.md

stages:
  - id: phase-0
    name: "UI Feedback Processing"
    depends_on: []
    skills: []
    trigger: output/pm-design/design-feedback/design_feedback.json exists
    gate:
      condition: "Feedback suggestions evaluated, accept/reject decided"
      fail_action: "Annotate unprocessed feedback, do not block main flow"

  - id: phase-1
    name: "Product Requirements Document"
    depends_on: [phase-0]
    skills: [design-prd]
    gate:
      condition: "PRD 4 quality gates all passed"
      fail_action: "Gate 1 or 2 failure blocks flow, output missing items list"

  - id: phase-2
    name: "Information Architecture Design"
    depends_on: [phase-1]
    parallel_with: [phase-3]
    skills: [design-ia]
    gate:
      condition: "IA plan human-confirmed"
      fail_action: "Generate 2-3 candidate plans for human selection"

  - id: phase-3
    name: "User Flow Design"
    depends_on: [phase-1]
    parallel_with: [phase-2]
    skills: [design-userflow]
    gate:
      condition: "User flow dead ends = 0"
      fail_action: "Dead ends must be fixed before entering prototype stage"

  - id: phase-4
    name: "Prototype Design"
    depends_on: [phase-2, phase-3]
    skills: [design-prototype]
    gate:
      condition: "Prototype design spec consistency >= 85%"
      fail_action: "Consistency PM feedback suggestions
Input:
  design_feedback: output/pm-design/design-feedback/design_feedback.json
Processing flow:
  1. Read design_feedback.json
  2. Group suggestions by target_artifact
  3. Evaluate each suggestion:
     - Accept: Mark as accepted, include in subsequent stage modification scope
     - Reject: Mark as rejected, record rejection reason
  4. [GATE] Human confirmation required for feedback processing results
  5. For accepted suggestions:
     - If target_artifact is prd.json: Include in phase-1 modification scope
     - If target_artifact is ia_proposals.json: Include in phase-2 modification scope
     - If target_artifact is userflow.json: Include in phase-3 modification scope
     - If target_artifact is component_catalog.json: Include in phase-4 modification scope
     - If target_artifact is interaction-spec.json: Include in phase-5 modification scope
  6. After processing, delete design_feedback.json to avoid duplicate consumption
Output: Feedback processing results (accepted/rejected list)
Validation: Feedback suggestions evaluated item by item, processing results human-confirmed
Mode: AI->Human

Invoke design-prd

Invoke: ${design-prd}
Input:
  ideation_workshop: output/pm-design/ideation-workshop/ideation-workshop.json
  strategic_output: User provided
  requirement_context: User provided (product_name required)
Output: output/pm-design/design-prd/
Validation: PRD 4 quality gates all passed
Mode: AI->Human

Invoke design-ia

Invoke: ${design-ia}
Input:
  prd: output/pm-design/design-prd/prd.md
  existing_ia: Optional
  user_research: Optional
Output: output/pm-design/design-ia/ia_proposals.json
Validation: IA plan human-confirmed
Mode: AI->Human

Invoke design-userflow

Invoke: ${design-userflow}
Input:
  prd: output/pm-design/design-prd/prd.md
  ia_proposals: output/pm-design/design-ia/ia_proposals.json
  user_research: Optional
Output: output/pm-design/design-userflow/userflow.json
Validation: User flow dead ends = 0
Mode: AI->Human

Invoke design-prototype

Invoke: ${design-prototype}
Input:
  ia_proposals: output/pm-design/design-ia/ia_proposals.json
  userflow: output/pm-design/design-userflow/userflow.json
  design_system: Optional
  design_tokens: Optional
Output: output/pm-design/design-prototype/prototype_spec.json
Validation: Prototype design spec consistency >= 85%
Mode: AI->Human

Invoke interaction-spec

Invoke: ${interaction-spec}
Input:
  userflow: output/pm-design/design-userflow/userflow.json
  prototype_spec: output/pm-design/design-prototype/prototype_spec.json
  handoff_doc: Optional
  brand_guidelines: Optional
Output: output/pm-design/interaction-spec/
Validation: Interaction state machine covers all 8 basic states
Mode: AI->Human

Invoke design-handoff-spec

Invoke: ${design-handoff-spec}
Input:
  prototype_spec: output/pm-design/design-prototype/prototype_spec.json
  design_tokens: Optional
  ia_proposals: output/pm-design/design-ia/ia_proposals.json
  userflow: output/pm-design/design-userflow/userflow.json
  prd: output/pm-design/design-prd/prd.md
  component_library: Optional
Output: output/pm-design/design-handoff-spec/
Validation: Handoff document pending confirmation items = 0
Mode: AI->Human

Invoke change-impact-analysis

Invoke: ${change-impact-analysis}
Input:
  prd_change: User provided (PRD change content)
  current_ia: output/pm-design/design-ia/ia_proposals.json
  current_userflow: output/pm-design/design-userflow/userflow.json
  current_prototype: output/pm-design/design-prototype/prototype_spec.json
Output: output/pm-design/change-impact-analysis/
Validation: Impact matrix covers all downstream design outputs; redo list is actionable
Mode: AI->Human

Stage Summary (post_pipeline)

Follows the stage summary protocol in [orchestrator-protocol.md](../../../codex-templates/orchestrator-protocol.md).

| Parameter | Value | |------|-----| | Sub-Skill output path | output/pm-design/ | | Summary output path | output/phase-reports/pm-design/design-orchestrator.md |

Next Steps: primary: metrics-orchestrator (product design complete, design metric system and instrumentation plan for PRD feature points) alternatives:

• target: validation-orchestrator

reason: High-risk assumptions in PRD need validation condition: When PRD features marked as high-risk exceed 30%

• target: api-design-orchestrator

reason: After PRD confirmation, parallel-start backend API design (cross-module: Backend) condition: In 0-to-1 product flow, after PRD confirmation need to parallel-start Backend development special_cases:

• target: design-handoff-spec

reason: Only need to generate handoff document for dev team condition: When design plan is confirmed and only development handoff summary is needed

Stage Gates

| Gate | Condition | Failed Handling | |------|------|------------| | PRD generation complete | design-prd output file generated and non-empty | Gate 1 or 2 failure blocks flow, output missing items list | | IA design complete | IA plan human-confirmed | Generate 2-3 candidate plans for human selection | | User flow complete | design-userflow output file generated and non-empty | Dead ends must be fixed before entering prototype stage | | Prototype complete | design-prototype output file generated and non-empty | Consistency < 85% requires human confirmation of violations | | Interaction spec complete | interaction-spec output file generated and non-empty | Supplement missing state definitions | | Design handoff complete | design-handoff-spec output file generated and non-empty | Pending items must be confirmed one by one or annotated as accepted risk | | Change impact analysis complete | change-impact-analysis output file generated and non-empty | Supplement missing downstream impact items | | Stage summary generated | output/phase-reports/pm-design/design-orchestrator.md generated and all 6 structure items non-empty | Supplement missing structure items and regenerate |

Human Decision Points

| Decision Point | Trigger Condition | Decision Content | |--------|----------|----------| | UI feedback processing confirmation | phase-0, when design_feedback.json exists | Confirm accept/reject of UI-side feedback suggestions | | PRD tier confirmation | AI auto-tiering confidence < 0.7 | Confirm PRD tier (L/S/X) | | IA plan selection | IA generates 2-3 candidate plans | Select final IA plan | | Design spec violation confirmation | Design spec consistency < 85% | Judge whether to accept violations | | Interaction spec confirmation | Interaction design spec generation complete | Confirm state machine, animation, and gesture specs |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: LuckyOneTwoThree
• Source: LuckyOneTwoThree/vibe-skill
• License: MIT
• Homepage: https://luckyonetwothree.github.io/all-skill-html/

Install and usage instructions live in the source repository linked above.