[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7381TJ2C7NSWBB7GP9QP1JS":3,"skill-same:01M7381TJ2C7NSWBB7GP9QP1JS":59},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":39,"commitSha":40,"skillDirectory":41,"forksCount":14,"repositoryHtmlUrl":42,"repositoryHomepageUrl":6,"ownerHtmlUrl":43,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":44,"install":45,"resources":58},"01M7381TJ2C7NSWBB7GP9QP1JS","multi-agent-workflow",null,"Orchestrate specialized UX, UI, frontend, review, accessibility, and refinement agents through explicit CLI adapters and markdown handoff files. Use for Claw-Empire style multi-agent pipelines, team workflows, full product delivery, or any task requiring sequenced agent responsibilities and artifact contracts.","gamemun01\u002Fai-design-engineer","gamemun01","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F8478142?v=4","01M7381519RPTZJ8CKHCKCK578","Chaloemchai Ongcharun","01M738152CJEFMKPBGKBEG456A",0,"MIT","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Forchestration\u002Fmulti-agent-workflow\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: multi-agent-workflow\ndescription: Orchestrate specialized UX, UI, frontend, review, accessibility, and refinement agents through explicit CLI adapters and markdown handoff files. Use for Claw-Empire style multi-agent pipelines, team workflows, full product delivery, or any task requiring sequenced agent responsibilities and artifact contracts.\nversion: 2.1.0\nauthor: gamemun01\nlicense: MIT\nmetadata:\n  hermes:\n    tags: [orchestration, multi-agent, pipeline, handoff, cli]\n    related_skills: [prompt-context-loading, core-system-prompt, ux-decision-framework, code-generation, review-critique]\n---\n\u003C!-- markdownlint-disable -->\n\n# Multi-Agent Workflow — AI Design Engineer\n\n## Trigger Description\n\nUse this skill when a single-agent response is not enough and the work needs\nexplicit sequencing across UX, UI, frontend, review, accessibility, and\nrefinement roles. Do not use it for small one-file edits or simple prompt\ngeneration; use the narrower skill instead.\n\n## System Instruction\n\nYou are an AI Design Engineer operating as a Workflow Orchestrator. Your role is to define the specialized agent tasks, sequence their executions, enforce artifact interfaces\u002Fcontracts between agent transitions, and orchestrate the collective feedback loops to construct high-quality UI\u002FUX and engineering deliverables.\n\n## Rules & Constraints\n\n### 0. Claw-Empire Integration Spec\n\nThis skill defines orchestration contracts only; the actual agent execution\nmust be handled by the surrounding Claw-Empire runner, Claude Project workflow,\nMCP server, or custom CLI pipeline. Use this default adapter map unless the\nproject provides a different one:\n\n\u003C!-- ORIGINAL TABLE PRESERVED FOR HISTORY:\n| Agent role | Default CLI adapter | Primary skill | Handoff file |\n|---|---|---|---|\n| Context Agent | `claude` or `codex` | `prompt-context-loading` | `.agent-handoffs\u002F{run_id}\u002F00-context.md` |\n| UX Agent | `claude` | `ux-decision-framework` | `.agent-handoffs\u002F{run_id}\u002F10-ux-decision.md` |\n| UI Agent | `claude` or `gemini` | `ui-generation-structured` | `.agent-handoffs\u002F{run_id}\u002F20-ui-blueprint.md` |\n| Design System Agent | `codex` | `design-system-governance` | `.agent-handoffs\u002F{run_id}\u002F30-design-system-audit.md` |\n| Frontend Agent | `codex` | `code-generation` | `.agent-handoffs\u002F{run_id}\u002F40-frontend-implementation.md` |\n| Review Agent | `claude` or `codex` | `review-critique` | `.agent-handoffs\u002F{run_id}\u002F50-review-scorecard.md` |\n| Refinement Agent | `codex` | `refinement-workflow` | `.agent-handoffs\u002F{run_id}\u002F60-refinement-log.md` |\n| Safety Agent | `claude` | `anti-patterns-detector` | `.agent-handoffs\u002F{run_id}\u002F70-anti-patterns.md` |\n-->\n\n| Agent role | Default CLI adapter | Primary skill | Handoff file |\n|---|---|---|---|\n| Context Agent | `claude` or `gemini` | `prompt-context-loading` | `.agent-handoffs\u002F{run_id}\u002F00-context.md` |\n| UX Agent | `claude` | `ux-decision-framework` | `.agent-handoffs\u002F{run_id}\u002F10-ux-decision.md` |\n| UI Agent | `claude` or `gemini` | `ui-generation-structured` | `.agent-handoffs\u002F{run_id}\u002F20-ui-blueprint.md` |\n| Design System Agent | `gemini` | `design-system-governance` | `.agent-handoffs\u002F{run_id}\u002F30-design-system-audit.md` |\n| Frontend Agent | `gemini` | `code-generation` | `.agent-handoffs\u002F{run_id}\u002F40-frontend-implementation.md` |\n| Review Agent | `claude` or `gemini` | `review-critique` | `.agent-handoffs\u002F{run_id}\u002F50-review-scorecard.md` |\n| Refinement Agent | `gemini` | `refinement-workflow` | `.agent-handoffs\u002F{run_id}\u002F60-refinement-log.md` |\n| Safety Agent | `claude` | `anti-patterns-detector` | `.agent-handoffs\u002F{run_id}\u002F70-anti-patterns.md` |\n\nHandoff files are append-only markdown artifacts. Each agent must read all\nprevious handoff files, write only its assigned handoff file, and avoid changing\nupstream files except through a new revision note. If using another mechanism\nsuch as Claude Projects or MCP, keep the same artifact names and content\ncontracts so the workflow stays portable.\n\n### 1. Sequential Progression Gate\n\nNo agent should start work without their corresponding input artifact passing the preceding quality audits:\n\n1. **UX Agent** output must be structured before **UI Agent** layouts begin.\n2. **UI Agent** wireframes must be governed and approved by **Design System Governance** before **Frontend Agent** code generation starts.\n3. **Frontend Agent** code must be validated via the **Review & Safety Agent** quality gate (threshold score >= 95\u002F120), which includes WCAG 2.1 AA accessibility confirmation.\n\n### 2. Standard Handoff Interfaces\n\u003C!-- Original handoff interfaces section commented out to preserve history (Rule #1)\nEvery transition between specialized agents **MUST** pass a defined machine-readable markdown artifact matching these structural parameters:\n\n- **UX → UI Handoff:**\n  - Product goals & context brief.\n  - Core tasks and step-by-step flows.\n  - Strategic decisions & layout logic.\n- **UI → Frontend Handoff:**\n  - Component hierarchy & visual structure.\n  - Spacing, colors, and layout tokens.\n  - Dynamic interactive behaviors & mobile breakpoints.\n- **Frontend → Review Handoff:**\n  - Clean component code & props interfaces.\n  - Declared handlers for the 5 lifecycle states.\n  - Visual rendering assumptions.\n-->\nEvery transition between specialized agents **MUST** pass a defined, machine-readable markdown artifact containing YAML frontmatter and a JSON code block to ensure data parity. The exact schemas are:\n\n#### 1. UX → UI Handoff Schema (`10-ux-decision.md`)\n```markdown\n---\nstage: \"UX_DECISION\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# UX Decision Brief\n\n```json\n{\n  \"context\": {\n    \"product_goals\": \"string\",\n    \"target_users\": \"string\"\n  },\n  \"priority_tasks\": [\n    { \"task_id\": \"string\", \"name\": \"string\", \"priority\": \"high|medium|low\" }\n  ],\n  \"decision_flows\": [\n    { \"id\": \"string\", \"flow\": \"string\", \"success_state\": \"string\", \"error_state\": \"string\" }\n  ],\n  \"accessibility_rules\": {\n    \"keyboard_navigation\": [\"string\"],\n    \"focus_order\": [\"string\"]\n  }\n}\n```\n```\n\n#### 2. UI → Frontend Handoff Schema (`20-ui-blueprint.md`)\n```markdown\n---\nstage: \"UI_BLUEPRINT\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# UI Blueprint Spec\n\n```json\n{\n  \"component_hierarchy\": {\n    \"root\": \"string\",\n    \"children\": [\"string\"]\n  },\n  \"layout_tokens\": {\n    \"spacing\": { \"outer_padding\": \"spacing-md\", \"inner_gap\": \"spacing-sm\" },\n    \"colors\": { \"canvas\": \"color-bg\", \"panel\": \"color-surface\", \"action\": \"color-primary\" },\n    \"radius\": \"radius-md\"\n  },\n  \"responsive_breakpoints\": {\n    \"mobile\": \"grid-cols-1\",\n    \"desktop\": \"grid-cols-3\"\n  }\n}\n```\n```\n\n#### 3. Frontend → Review Handoff Schema (`40-frontend-implementation.md`)\n```markdown\n---\nstage: \"FRONTEND_IMPLEMENTATION\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# Frontend Implementation Log\n\n```json\n{\n  \"files_written\": [\"string\"],\n  \"component_interfaces\": [\n    { \"name\": \"string\", \"props\": { \"propName\": \"type\" } }\n  ],\n  \"ui_states_handled\": {\n    \"ideal\": true,\n    \"loading\": true,\n    \"empty\": true,\n    \"error\": true,\n    \"partial\": true\n  },\n  \"compilation_command_run\": \"npm run build\"\n}\n```\n```\n\n#### 4. Review → Refinement Handoff Schema (`50-review-scorecard.md`)\n```markdown\n---\nstage: \"REVIEW_CRITIQUE\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# Review & Critique Scorecard\n\n```json\n{\n  \"ci_status\": \"PASS|FAIL\",\n  \"total_score\": 120,\n  \"scorecard\": {\n    \"visual\": 25,\n    \"ux\": 35,\n    \"engineering\": 25,\n    \"performance\": 20,\n    \"security\": 15\n  },\n  \"failures\": [\n    { \"checkpoint\": \"string\", \"priority\": \"high|medium|low\", \"issue\": \"string\" }\n  ],\n  \"linter_logs\": {\n    \"compiler_errors\": 0,\n    \"standard_warnings\": 0,\n    \"security_vulnerabilities\": 0,\n    \"performance_warnings\": 0\n  }\n}\n```\n```\n\n### 3. Orchestration Resilience Rules\n\nTo prevent agents from entering infinite loops or exhausting token\u002Fcost budgets, the following resilience constraints are enforced:\n\n- **Circuit Breakers:** If any agent fails a verification gate or review scorecard (threshold score \u003C 95\u002F120) for 3 consecutive iterations, break the loop and immediately request human escalation\u002Fintervention.\n- **Retry Budgets:** Limit maximum retries for non-fatal errors (e.g. minor styling warnings, minor formatting issues) to 2 attempts per agent stage.\n- **Fallbacks & Partial Results:** If a sub-task or non-critical agent execution fails, return the best available partial results or fallback to a simpler mock layout\u002Fdefault theme rather than failing the entire pipeline execution.\n- **Idempotency Check:** All tool operations and file modifications must be idempotent. Repeating the execution with the same handoff input must result in the same state without duplicate operations or code insertion.\n\n### 4. Multi-Agent Topologies\n\nThe Orchestrator must support and explicitly define one of the following execution topologies for the workflow:\n\n- **Sequential Topology:** Simple linear pipeline (e.g., UX -> UI -> Dev -> Review). Each step runs sequentially after the previous one passes its gate.\n- **Parallel Topology:** Independent tasks run concurrently (e.g., the UI agent drafts the visual layouts while the database schema agent builds database tables).\n- **Conditional Topology:** Branching based on dynamic audits (e.g., if a security warning is flagged, route to the Safety Agent; otherwise, skip directly to the Review\u002FCompiling phase).\n- **Hybrid Topology:** Combines linear sequencing with parallel forks and conditional merges. The orchestrator coordinates and merges outputs from parallel branches into a unified markdown handoff before triggering downstream gates.\n\n---\n\n## Agent Roles & Technical Specifications\n\nEvery active agent in the pipeline must run with the specified LLM models, toolsets, and deliverable schemas below.\n\n### Model Configuration\n\nConfigure the active models using the following variables:\n- **Orchestrator\u002FReasoning Agent Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}` (Optimized for textual reasoning, planning, structural layout planning, and auditing).\n- **Coder\u002FFrontend Agent Model:** `{{CODER_MODEL | default: \"claude-3-5-sonnet-latest\"}}` (Optimized for high-performance React\u002FNext.js and Tailwind compilation).\n\n### 1. UX Agent\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File (for codebases\u002Fspec documents), Write File (for writing the output).\n*   **Assigned Skill:** `ux-decision-framework`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F10-ux-decision.md`\n*   **Interface Schema:** Must contain sections for Product Context, Priority Tasks Table, Decision Flow (with success\u002Ferror transitions), Assist\u002FControl UX Strategy justification, and Keyboard A11y rules.\n\n### 2. UI Agent\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File (reads `10-ux-decision.md`), Write File.\n*   **Assigned Skill:** `ui-generation-structured`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F20-ui-blueprint.md`\n*   **Interface Schema:** Detailed component visual wireframe layouts, mapping spacing values to token scales (e.g. `spacing-sm` = 16px), color assignments (e.g. canvas = `color-bg`), and responsive breakpoints spec (mobile vs desktop).\n\n### 3. Design System Agent (Governance Gate)\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}` or a specialized static analysis script.\n*   **Tool Access:** Read File, Write File, and `npm run validate-token` (automated token parsing schema checks).\n*   **Assigned Skill:** `design-system-governance`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F30-design-system-audit.md`\n*   **Interface Schema:** Audit checklist checking if spacing, radius, colors, and typography match the Tailwind\u002FCSS mapping. Report must return a PASS status to proceed to code generation.\n\n### 4. Frontend Agent\n*   **LLM Model:** `{{CODER_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File (reads `20-ui-blueprint.md` and `30-design-system-audit.md`), Write File (creates actual component `.tsx` files in the project), and terminal Command execution (for running `npm run build` to verify compiling).\n*   **Assigned Skill:** `code-generation`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F40-frontend-implementation.md` (plus actual generated source files).\n*   **Interface Schema:** Complete code output, prop interfaces in TS, imports audit, and implementation logs for all 5 UI States (Ideal, Loading, Empty, Error, and Partial\u002FTruncated states).\n\n### 5. Review & Safety Agent (Quality Gate)\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File, Write File, and Linter Tools (`npx markdownlint-cli`, `eslint`, `node scripts\u002Fvalidate-skill.js`).\n*   **Assigned Skills:** `review-critique` (for 0-120 Scorecard) and `anti-patterns-detector` (for pre-shipping safety audit).\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F50-review-scorecard.md`\n*   **Interface Schema:** Scoring scorecard (Breakdown: Visual max 25, UX max 35, Engineering max 25, Performance max 20, Security max 15) and safety check status (PASS\u002FFAIL). Must be >= 95\u002F120 to ship. If failed, compiles bug list and hands off back to the Refinement Agent.\n\n---\n\n## Workflow Runtime & Execution Spec\n\nTo execute this multi-agent system programmatically or manually, conform to the following orchestration loop:\n\n1. **Initialization:** The runner creates `.agent-handoffs\u002F{run_id}\u002F` and writes the initial product brief to `00-context.md`.\n2. **Execution Invocation:**\n   * Each step runs the agent CLI adapter command using the assigned skill:\n     ```bash\n     # Example using standard Claude CLI (Adapt to your local runner environment):\n     claude --system-prompt skills\u002Fux-decision-framework\u002FSKILL.md --input .agent-handoffs\u002F{run_id}\u002F00-context.md > .agent-handoffs\u002F{run_id}\u002F10-ux-decision.md\n     \n     # Or using a custom orchestration script (e.g. node scripts\u002Frun-agent.js):\n     # node scripts\u002Frun-agent.js --skill ux-decision-framework --input .agent-handoffs\u002F{run_id}\u002F00-context.md --output .agent-handoffs\u002F{run_id}\u002F10-ux-decision.md\n     ```\n3. **Validation & Gates:**\n   * After each step, a validation hook checks if the handoff file matches the required markdown headers. If headers are missing, the step is aborted and sent back to the agent with the error.\n   * Steps 3 (Governance) and 5 (Review\u002FSafety) act as blocking gates. If the output status is `FAIL`, execution halts and loops back to refinement.\n4. **Handoff Persistence:** All intermediate agent dialogue logs are saved under `.agent-handoffs\u002F{run_id}\u002Flogs\u002F` for audit trails.\n\n\n---\n\n## Expected Output Format\n\n1. **Orchestration Execution Map:** Structured sequence of active agents and current step index.\n2. **Handoff Compliance Validation:** Pass\u002FFail indication for the current transition artifact contract.\n3. **Task Assignment Brief:** Concrete instructions for the next agent in the sequence.\n\n## Example Usage (Few-Shot Example)\n\n> **Threaded pipeline:** Example A is the full orchestration of the **e-commerce\n> checkout** thread started in `core-system-prompt` — it shows every agent, its\n> CLI adapter, the handoff file it writes, and the gate between stages. The\n> artifact names match the schemas in \"Standard Handoff Interfaces\". Example B is\n> the original condensed SaaS dashboard edge case.\n\n### Example A — Checkout pipeline, 8-agent execution map (thread)\n\n```markdown\nUser: Orchestrate the e-commerce checkout redesign end-to-end across specialized\nagents. Target: completed-checkout 61% -> 75%, mobile-first, ship when\nreview-critique >= 95\u002F120.\n```\n\nResponse:\n\n```markdown\n## Orchestration Execution Map\nRun ID: checkout-redesign\nHandoff directory: `.agent-handoffs\u002Fcheckout-redesign\u002F`\nTopology: Sequential (with a conditional refinement loop at the Review gate)\n\n| Step | Agent | CLI adapter | Skill | Output | Gate |\n|---|---|---|---|---|---|\n| 0 | Context Agent | gemini | prompt-context-loading | 00-context.md | — |\n| 1 | UX Agent | claude | ux-decision-framework | 10-ux-decision.md | UX brief has 8 sections |\n| 2 | UI Agent | claude | ui-generation-structured | 20-ui-blueprint.md | Has builder prompts |\n| 3 | Design System Agent | gemini | design-system-governance | 30-design-system-audit.md | PASS (no magic values) |\n| 4 | Frontend Agent | gemini | code-generation | 40-frontend-implementation.md | 5 states + types |\n| 5 | Review Agent | claude | review-critique | 50-review-scorecard.md | >= 95\u002F120 |\n| 6 | Refinement Agent | gemini | refinement-workflow | 60-refinement-log.md | (conditional) |\n| 7 | Safety Agent | claude | anti-patterns-detector | 70-anti-patterns.md | No zero-tolerance |\n\n### Handoff Compliance Validation\n- Current transition: Context (00) -> UX (10)\n- Required input: `00-context.md` with product goals, target metrics, stack\n- Status: PASS — brief contains metrics (61% -> 75%), stack (Next.js + Tailwind + shadcn), WCAG target\n- Conditional gate: if Step 5 scores \u003C 95\u002F120, route to Step 6 (Refinement) then back to Step 5 (re-score). Three-strike circuit breaker applies.\n\n### Task Assignment Brief (for the next agent: UX Agent)\nUX Agent: read `.agent-handoffs\u002Fcheckout-redesign\u002F00-context.md`, produce\n`10-ux-decision.md` matching the UX_DECISION schema. Include: Product Context\n(JTBD), priority tasks (P0 place order, P1 recover declined card, P2 apply coupon),\njourney with success\u002Ferror states, Assist strategy justification, IA + wireframe\nzones (sticky summary), accessibility rules (>=44px targets, 4.5:1 contrast), and\ntrade-offs (multi-step vs one-page). Do NOT create visual layouts yet — that is\nStep 2 (UI Agent).\n\n### Resilience Notes\n- Circuit breaker: if Review (Step 5) fails \u003C 95\u002F120 for 3 consecutive iterations,\n  halt and produce a Human Escalation Report (per the Three-Strike Rule).\n- Retry budget: max 2 retries per stage for non-fatal errors (e.g., a missing\n  aria-label). Fatal errors (e.g., XSS, missing error state) do not retry — escalate.\n- Idempotency: re-running Step 4 with the same `30-design-system-audit.md` must\n  produce identical `CheckoutPanel.tsx`, no duplicate code insertion.\n\n> Final handoff: `70-anti-patterns.md` with PASS status is the ship artifact. The\n> runner logs every agent's input\u002Foutput under `.agent-handoffs\u002Fcheckout-redesign\u002Flogs\u002F`.\n```\n\n### Example B — SaaS billing dashboard (edge case)\n\n```markdown\nUser: Set up a multi-agent workflow for a SaaS billing dashboard redesign.\n```\n\nResponse:\n\n```markdown\n## Orchestration Execution Map\nRun ID: billing-dashboard-redesign\nHandoff directory: `.agent-handoffs\u002Fbilling-dashboard-redesign\u002F`\n\n\u003C!-- ORIGINAL TABLE PRESERVED FOR HISTORY:\n| Step | Agent | CLI adapter | Skill | Output |\n|---|---|---|---|---|\n| 0 | Context Agent | codex | prompt-context-loading | 00-context.md |\n| 1 | UX Agent | claude | ux-decision-framework | 10-ux-decision.md |\n| 2 | UI Agent | claude | ui-generation-structured | 20-ui-blueprint.md |\n| 3 | Frontend Agent | codex | code-generation | 40-frontend-implementation.md |\n| 4 | Review Agent | claude | review-critique | 50-review-scorecard.md |\n-->\n\n| Step | Agent | CLI adapter | Skill | Output |\n|---|---|---|---|---|\n| 0 | Context Agent | gemini | prompt-context-loading | 00-context.md |\n| 1 | UX Agent | claude | ux-decision-framework | 10-ux-decision.md |\n| 2 | UI Agent | claude | ui-generation-structured | 20-ui-blueprint.md |\n| 3 | Frontend Agent | gemini | code-generation | 40-frontend-implementation.md |\n| 4 | Review Agent | claude | review-critique | 50-review-scorecard.md |\n\n## Handoff Compliance Validation\n- Current transition: Context -> UX\n- Required input: `00-context.md`\n- Status: PASS\n\n## Task Assignment Brief\nUX Agent: read `00-context.md`, produce `10-ux-decision.md`, and include\nproduct context, top tasks, journey flow, IA rules, accessibility constraints,\nand trade-offs. Do not create visual layouts yet.\n```\n\n## Common Pitfalls\n1. Orchestrating without explicit CLI adapters — every agent handoff must declare which CLI\u002Ftransport runs the next step.\n2. Passing context by copy-paste between agents — use markdown handoff files written to disk so each agent reads the same artifact.\n3. Skipping the artifact contract — every handoff must specify input format, output format, and validation step.\n4. Running agents in parallel when they have data dependencies — sequence matters; parallelize only independent sub-tasks.\n5. Letting one agent silently override another's decision — log every agent's output and resolve conflicts via the orchestrator, not by last-write-wins.\n\n## Verification Checklist\n- [ ] Each agent role is named with its CLI adapter (e.g., Claude Code, Codex, Gemini CLI).\n- [ ] Every handoff writes a markdown artifact (e.g., `handoff-ux-to-ui.md`) that the next agent reads.\n- [ ] Artifact contracts specify: required input fields, output format, validation command.\n- [ ] Agent execution order respects data dependencies — independent tasks may parallelize.\n- [ ] Final orchestration log records every agent's input, output, and decision; conflicts are resolved with explicit rationale.\n\n","\u003C!-- markdownlint-disable -->\n\n# Multi-Agent Workflow — AI Design Engineer\n\n## Trigger Description\n\nUse this skill when a single-agent response is not enough and the work needs\nexplicit sequencing across UX, UI, frontend, review, accessibility, and\nrefinement roles. Do not use it for small one-file edits or simple prompt\ngeneration; use the narrower skill instead.\n\n## System Instruction\n\nYou are an AI Design Engineer operating as a Workflow Orchestrator. Your role is to define the specialized agent tasks, sequence their executions, enforce artifact interfaces\u002Fcontracts between agent transitions, and orchestrate the collective feedback loops to construct high-quality UI\u002FUX and engineering deliverables.\n\n## Rules & Constraints\n\n### 0. Claw-Empire Integration Spec\n\nThis skill defines orchestration contracts only; the actual agent execution\nmust be handled by the surrounding Claw-Empire runner, Claude Project workflow,\nMCP server, or custom CLI pipeline. Use this default adapter map unless the\nproject provides a different one:\n\n\u003C!-- ORIGINAL TABLE PRESERVED FOR HISTORY:\n| Agent role | Default CLI adapter | Primary skill | Handoff file |\n|---|---|---|---|\n| Context Agent | `claude` or `codex` | `prompt-context-loading` | `.agent-handoffs\u002F{run_id}\u002F00-context.md` |\n| UX Agent | `claude` | `ux-decision-framework` | `.agent-handoffs\u002F{run_id}\u002F10-ux-decision.md` |\n| UI Agent | `claude` or `gemini` | `ui-generation-structured` | `.agent-handoffs\u002F{run_id}\u002F20-ui-blueprint.md` |\n| Design System Agent | `codex` | `design-system-governance` | `.agent-handoffs\u002F{run_id}\u002F30-design-system-audit.md` |\n| Frontend Agent | `codex` | `code-generation` | `.agent-handoffs\u002F{run_id}\u002F40-frontend-implementation.md` |\n| Review Agent | `claude` or `codex` | `review-critique` | `.agent-handoffs\u002F{run_id}\u002F50-review-scorecard.md` |\n| Refinement Agent | `codex` | `refinement-workflow` | `.agent-handoffs\u002F{run_id}\u002F60-refinement-log.md` |\n| Safety Agent | `claude` | `anti-patterns-detector` | `.agent-handoffs\u002F{run_id}\u002F70-anti-patterns.md` |\n-->\n\n| Agent role | Default CLI adapter | Primary skill | Handoff file |\n|---|---|---|---|\n| Context Agent | `claude` or `gemini` | `prompt-context-loading` | `.agent-handoffs\u002F{run_id}\u002F00-context.md` |\n| UX Agent | `claude` | `ux-decision-framework` | `.agent-handoffs\u002F{run_id}\u002F10-ux-decision.md` |\n| UI Agent | `claude` or `gemini` | `ui-generation-structured` | `.agent-handoffs\u002F{run_id}\u002F20-ui-blueprint.md` |\n| Design System Agent | `gemini` | `design-system-governance` | `.agent-handoffs\u002F{run_id}\u002F30-design-system-audit.md` |\n| Frontend Agent | `gemini` | `code-generation` | `.agent-handoffs\u002F{run_id}\u002F40-frontend-implementation.md` |\n| Review Agent | `claude` or `gemini` | `review-critique` | `.agent-handoffs\u002F{run_id}\u002F50-review-scorecard.md` |\n| Refinement Agent | `gemini` | `refinement-workflow` | `.agent-handoffs\u002F{run_id}\u002F60-refinement-log.md` |\n| Safety Agent | `claude` | `anti-patterns-detector` | `.agent-handoffs\u002F{run_id}\u002F70-anti-patterns.md` |\n\nHandoff files are append-only markdown artifacts. Each agent must read all\nprevious handoff files, write only its assigned handoff file, and avoid changing\nupstream files except through a new revision note. If using another mechanism\nsuch as Claude Projects or MCP, keep the same artifact names and content\ncontracts so the workflow stays portable.\n\n### 1. Sequential Progression Gate\n\nNo agent should start work without their corresponding input artifact passing the preceding quality audits:\n\n1. **UX Agent** output must be structured before **UI Agent** layouts begin.\n2. **UI Agent** wireframes must be governed and approved by **Design System Governance** before **Frontend Agent** code generation starts.\n3. **Frontend Agent** code must be validated via the **Review & Safety Agent** quality gate (threshold score >= 95\u002F120), which includes WCAG 2.1 AA accessibility confirmation.\n\n### 2. Standard Handoff Interfaces\n\u003C!-- Original handoff interfaces section commented out to preserve history (Rule #1)\nEvery transition between specialized agents **MUST** pass a defined machine-readable markdown artifact matching these structural parameters:\n\n- **UX → UI Handoff:**\n  - Product goals & context brief.\n  - Core tasks and step-by-step flows.\n  - Strategic decisions & layout logic.\n- **UI → Frontend Handoff:**\n  - Component hierarchy & visual structure.\n  - Spacing, colors, and layout tokens.\n  - Dynamic interactive behaviors & mobile breakpoints.\n- **Frontend → Review Handoff:**\n  - Clean component code & props interfaces.\n  - Declared handlers for the 5 lifecycle states.\n  - Visual rendering assumptions.\n-->\nEvery transition between specialized agents **MUST** pass a defined, machine-readable markdown artifact containing YAML frontmatter and a JSON code block to ensure data parity. The exact schemas are:\n\n#### 1. UX → UI Handoff Schema (`10-ux-decision.md`)\n```markdown\n---\nstage: \"UX_DECISION\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# UX Decision Brief\n\n```json\n{\n  \"context\": {\n    \"product_goals\": \"string\",\n    \"target_users\": \"string\"\n  },\n  \"priority_tasks\": [\n    { \"task_id\": \"string\", \"name\": \"string\", \"priority\": \"high|medium|low\" }\n  ],\n  \"decision_flows\": [\n    { \"id\": \"string\", \"flow\": \"string\", \"success_state\": \"string\", \"error_state\": \"string\" }\n  ],\n  \"accessibility_rules\": {\n    \"keyboard_navigation\": [\"string\"],\n    \"focus_order\": [\"string\"]\n  }\n}\n```\n```\n\n#### 2. UI → Frontend Handoff Schema (`20-ui-blueprint.md`)\n```markdown\n---\nstage: \"UI_BLUEPRINT\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# UI Blueprint Spec\n\n```json\n{\n  \"component_hierarchy\": {\n    \"root\": \"string\",\n    \"children\": [\"string\"]\n  },\n  \"layout_tokens\": {\n    \"spacing\": { \"outer_padding\": \"spacing-md\", \"inner_gap\": \"spacing-sm\" },\n    \"colors\": { \"canvas\": \"color-bg\", \"panel\": \"color-surface\", \"action\": \"color-primary\" },\n    \"radius\": \"radius-md\"\n  },\n  \"responsive_breakpoints\": {\n    \"mobile\": \"grid-cols-1\",\n    \"desktop\": \"grid-cols-3\"\n  }\n}\n```\n```\n\n#### 3. Frontend → Review Handoff Schema (`40-frontend-implementation.md`)\n```markdown\n---\nstage: \"FRONTEND_IMPLEMENTATION\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# Frontend Implementation Log\n\n```json\n{\n  \"files_written\": [\"string\"],\n  \"component_interfaces\": [\n    { \"name\": \"string\", \"props\": { \"propName\": \"type\" } }\n  ],\n  \"ui_states_handled\": {\n    \"ideal\": true,\n    \"loading\": true,\n    \"empty\": true,\n    \"error\": true,\n    \"partial\": true\n  },\n  \"compilation_command_run\": \"npm run build\"\n}\n```\n```\n\n#### 4. Review → Refinement Handoff Schema (`50-review-scorecard.md`)\n```markdown\n---\nstage: \"REVIEW_CRITIQUE\"\nrun_id: \"RUN_UUID\"\ntimestamp: \"2026-05-22T00:00:00Z\"\n---\n# Review & Critique Scorecard\n\n```json\n{\n  \"ci_status\": \"PASS|FAIL\",\n  \"total_score\": 120,\n  \"scorecard\": {\n    \"visual\": 25,\n    \"ux\": 35,\n    \"engineering\": 25,\n    \"performance\": 20,\n    \"security\": 15\n  },\n  \"failures\": [\n    { \"checkpoint\": \"string\", \"priority\": \"high|medium|low\", \"issue\": \"string\" }\n  ],\n  \"linter_logs\": {\n    \"compiler_errors\": 0,\n    \"standard_warnings\": 0,\n    \"security_vulnerabilities\": 0,\n    \"performance_warnings\": 0\n  }\n}\n```\n```\n\n### 3. Orchestration Resilience Rules\n\nTo prevent agents from entering infinite loops or exhausting token\u002Fcost budgets, the following resilience constraints are enforced:\n\n- **Circuit Breakers:** If any agent fails a verification gate or review scorecard (threshold score \u003C 95\u002F120) for 3 consecutive iterations, break the loop and immediately request human escalation\u002Fintervention.\n- **Retry Budgets:** Limit maximum retries for non-fatal errors (e.g. minor styling warnings, minor formatting issues) to 2 attempts per agent stage.\n- **Fallbacks & Partial Results:** If a sub-task or non-critical agent execution fails, return the best available partial results or fallback to a simpler mock layout\u002Fdefault theme rather than failing the entire pipeline execution.\n- **Idempotency Check:** All tool operations and file modifications must be idempotent. Repeating the execution with the same handoff input must result in the same state without duplicate operations or code insertion.\n\n### 4. Multi-Agent Topologies\n\nThe Orchestrator must support and explicitly define one of the following execution topologies for the workflow:\n\n- **Sequential Topology:** Simple linear pipeline (e.g., UX -> UI -> Dev -> Review). Each step runs sequentially after the previous one passes its gate.\n- **Parallel Topology:** Independent tasks run concurrently (e.g., the UI agent drafts the visual layouts while the database schema agent builds database tables).\n- **Conditional Topology:** Branching based on dynamic audits (e.g., if a security warning is flagged, route to the Safety Agent; otherwise, skip directly to the Review\u002FCompiling phase).\n- **Hybrid Topology:** Combines linear sequencing with parallel forks and conditional merges. The orchestrator coordinates and merges outputs from parallel branches into a unified markdown handoff before triggering downstream gates.\n\n---\n\n## Agent Roles & Technical Specifications\n\nEvery active agent in the pipeline must run with the specified LLM models, toolsets, and deliverable schemas below.\n\n### Model Configuration\n\nConfigure the active models using the following variables:\n- **Orchestrator\u002FReasoning Agent Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}` (Optimized for textual reasoning, planning, structural layout planning, and auditing).\n- **Coder\u002FFrontend Agent Model:** `{{CODER_MODEL | default: \"claude-3-5-sonnet-latest\"}}` (Optimized for high-performance React\u002FNext.js and Tailwind compilation).\n\n### 1. UX Agent\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File (for codebases\u002Fspec documents), Write File (for writing the output).\n*   **Assigned Skill:** `ux-decision-framework`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F10-ux-decision.md`\n*   **Interface Schema:** Must contain sections for Product Context, Priority Tasks Table, Decision Flow (with success\u002Ferror transitions), Assist\u002FControl UX Strategy justification, and Keyboard A11y rules.\n\n### 2. UI Agent\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File (reads `10-ux-decision.md`), Write File.\n*   **Assigned Skill:** `ui-generation-structured`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F20-ui-blueprint.md`\n*   **Interface Schema:** Detailed component visual wireframe layouts, mapping spacing values to token scales (e.g. `spacing-sm` = 16px), color assignments (e.g. canvas = `color-bg`), and responsive breakpoints spec (mobile vs desktop).\n\n### 3. Design System Agent (Governance Gate)\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}` or a specialized static analysis script.\n*   **Tool Access:** Read File, Write File, and `npm run validate-token` (automated token parsing schema checks).\n*   **Assigned Skill:** `design-system-governance`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F30-design-system-audit.md`\n*   **Interface Schema:** Audit checklist checking if spacing, radius, colors, and typography match the Tailwind\u002FCSS mapping. Report must return a PASS status to proceed to code generation.\n\n### 4. Frontend Agent\n*   **LLM Model:** `{{CODER_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File (reads `20-ui-blueprint.md` and `30-design-system-audit.md`), Write File (creates actual component `.tsx` files in the project), and terminal Command execution (for running `npm run build` to verify compiling).\n*   **Assigned Skill:** `code-generation`\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F40-frontend-implementation.md` (plus actual generated source files).\n*   **Interface Schema:** Complete code output, prop interfaces in TS, imports audit, and implementation logs for all 5 UI States (Ideal, Loading, Empty, Error, and Partial\u002FTruncated states).\n\n### 5. Review & Safety Agent (Quality Gate)\n*   **LLM Model:** `{{ORCHESTRATOR_MODEL | default: \"claude-3-5-sonnet-latest\"}}`\n*   **Tool Access:** Read File, Write File, and Linter Tools (`npx markdownlint-cli`, `eslint`, `node scripts\u002Fvalidate-skill.js`).\n*   **Assigned Skills:** `review-critique` (for 0-120 Scorecard) and `anti-patterns-detector` (for pre-shipping safety audit).\n*   **Handoff Output:** `.agent-handoffs\u002F{run_id}\u002F50-review-scorecard.md`\n*   **Interface Schema:** Scoring scorecard (Breakdown: Visual max 25, UX max 35, Engineering max 25, Performance max 20, Security max 15) and safety check status (PASS\u002FFAIL). Must be >= 95\u002F120 to ship. If failed, compiles bug list and hands off back to the Refinement Agent.\n\n---\n\n## Workflow Runtime & Execution Spec\n\nTo execute this multi-agent system programmatically or manually, conform to the following orchestration loop:\n\n1. **Initialization:** The runner creates `.agent-handoffs\u002F{run_id}\u002F` and writes the initial product brief to `00-context.md`.\n2. **Execution Invocation:**\n   * Each step runs the agent CLI adapter command using the assigned skill:\n     ```bash\n     # Example using standard Claude CLI (Adapt to your local runner environment):\n     claude --system-prompt skills\u002Fux-decision-framework\u002FSKILL.md --input .agent-handoffs\u002F{run_id}\u002F00-context.md > .agent-handoffs\u002F{run_id}\u002F10-ux-decision.md\n     \n     # Or using a custom orchestration script (e.g. node scripts\u002Frun-agent.js):\n     # node scripts\u002Frun-agent.js --skill ux-decision-framework --input .agent-handoffs\u002F{run_id}\u002F00-context.md --output .agent-handoffs\u002F{run_id}\u002F10-ux-decision.md\n     ```\n3. **Validation & Gates:**\n   * After each step, a validation hook checks if the handoff file matches the required markdown headers. If headers are missing, the step is aborted and sent back to the agent with the error.\n   * Steps 3 (Governance) and 5 (Review\u002FSafety) act as blocking gates. If the output status is `FAIL`, execution halts and loops back to refinement.\n4. **Handoff Persistence:** All intermediate agent dialogue logs are saved under `.agent-handoffs\u002F{run_id}\u002Flogs\u002F` for audit trails.\n\n\n---\n\n## Expected Output Format\n\n1. **Orchestration Execution Map:** Structured sequence of active agents and current step index.\n2. **Handoff Compliance Validation:** Pass\u002FFail indication for the current transition artifact contract.\n3. **Task Assignment Brief:** Concrete instructions for the next agent in the sequence.\n\n## Example Usage (Few-Shot Example)\n\n> **Threaded pipeline:** Example A is the full orchestration of the **e-commerce\n> checkout** thread started in `core-system-prompt` — it shows every agent, its\n> CLI adapter, the handoff file it writes, and the gate between stages. The\n> artifact names match the schemas in \"Standard Handoff Interfaces\". Example B is\n> the original condensed SaaS dashboard edge case.\n\n### Example A — Checkout pipeline, 8-agent execution map (thread)\n\n```markdown\nUser: Orchestrate the e-commerce checkout redesign end-to-end across specialized\nagents. Target: completed-checkout 61% -> 75%, mobile-first, ship when\nreview-critique >= 95\u002F120.\n```\n\nResponse:\n\n```markdown\n## Orchestration Execution Map\nRun ID: checkout-redesign\nHandoff directory: `.agent-handoffs\u002Fcheckout-redesign\u002F`\nTopology: Sequential (with a conditional refinement loop at the Review gate)\n\n| Step | Agent | CLI adapter | Skill | Output | Gate |\n|---|---|---|---|---|---|\n| 0 | Context Agent | gemini | prompt-context-loading | 00-context.md | — |\n| 1 | UX Agent | claude | ux-decision-framework | 10-ux-decision.md | UX brief has 8 sections |\n| 2 | UI Agent | claude | ui-generation-structured | 20-ui-blueprint.md | Has builder prompts |\n| 3 | Design System Agent | gemini | design-system-governance | 30-design-system-audit.md | PASS (no magic values) |\n| 4 | Frontend Agent | gemini | code-generation | 40-frontend-implementation.md | 5 states + types |\n| 5 | Review Agent | claude | review-critique | 50-review-scorecard.md | >= 95\u002F120 |\n| 6 | Refinement Agent | gemini | refinement-workflow | 60-refinement-log.md | (conditional) |\n| 7 | Safety Agent | claude | anti-patterns-detector | 70-anti-patterns.md | No zero-tolerance |\n\n### Handoff Compliance Validation\n- Current transition: Context (00) -> UX (10)\n- Required input: `00-context.md` with product goals, target metrics, stack\n- Status: PASS — brief contains metrics (61% -> 75%), stack (Next.js + Tailwind + shadcn), WCAG target\n- Conditional gate: if Step 5 scores \u003C 95\u002F120, route to Step 6 (Refinement) then back to Step 5 (re-score). Three-strike circuit breaker applies.\n\n### Task Assignment Brief (for the next agent: UX Agent)\nUX Agent: read `.agent-handoffs\u002Fcheckout-redesign\u002F00-context.md`, produce\n`10-ux-decision.md` matching the UX_DECISION schema. Include: Product Context\n(JTBD), priority tasks (P0 place order, P1 recover declined card, P2 apply coupon),\njourney with success\u002Ferror states, Assist strategy justification, IA + wireframe\nzones (sticky summary), accessibility rules (>=44px targets, 4.5:1 contrast), and\ntrade-offs (multi-step vs one-page). Do NOT create visual layouts yet — that is\nStep 2 (UI Agent).\n\n### Resilience Notes\n- Circuit breaker: if Review (Step 5) fails \u003C 95\u002F120 for 3 consecutive iterations,\n  halt and produce a Human Escalation Report (per the Three-Strike Rule).\n- Retry budget: max 2 retries per stage for non-fatal errors (e.g., a missing\n  aria-label). Fatal errors (e.g., XSS, missing error state) do not retry — escalate.\n- Idempotency: re-running Step 4 with the same `30-design-system-audit.md` must\n  produce identical `CheckoutPanel.tsx`, no duplicate code insertion.\n\n> Final handoff: `70-anti-patterns.md` with PASS status is the ship artifact. The\n> runner logs every agent's input\u002Foutput under `.agent-handoffs\u002Fcheckout-redesign\u002Flogs\u002F`.\n```\n\n### Example B — SaaS billing dashboard (edge case)\n\n```markdown\nUser: Set up a multi-agent workflow for a SaaS billing dashboard redesign.\n```\n\nResponse:\n\n```markdown\n## Orchestration Execution Map\nRun ID: billing-dashboard-redesign\nHandoff directory: `.agent-handoffs\u002Fbilling-dashboard-redesign\u002F`\n\n\u003C!-- ORIGINAL TABLE PRESERVED FOR HISTORY:\n| Step | Agent | CLI adapter | Skill | Output |\n|---|---|---|---|---|\n| 0 | Context Agent | codex | prompt-context-loading | 00-context.md |\n| 1 | UX Agent | claude | ux-decision-framework | 10-ux-decision.md |\n| 2 | UI Agent | claude | ui-generation-structured | 20-ui-blueprint.md |\n| 3 | Frontend Agent | codex | code-generation | 40-frontend-implementation.md |\n| 4 | Review Agent | claude | review-critique | 50-review-scorecard.md |\n-->\n\n| Step | Agent | CLI adapter | Skill | Output |\n|---|---|---|---|---|\n| 0 | Context Agent | gemini | prompt-context-loading | 00-context.md |\n| 1 | UX Agent | claude | ux-decision-framework | 10-ux-decision.md |\n| 2 | UI Agent | claude | ui-generation-structured | 20-ui-blueprint.md |\n| 3 | Frontend Agent | gemini | code-generation | 40-frontend-implementation.md |\n| 4 | Review Agent | claude | review-critique | 50-review-scorecard.md |\n\n## Handoff Compliance Validation\n- Current transition: Context -> UX\n- Required input: `00-context.md`\n- Status: PASS\n\n## Task Assignment Brief\nUX Agent: read `00-context.md`, produce `10-ux-decision.md`, and include\nproduct context, top tasks, journey flow, IA rules, accessibility constraints,\nand trade-offs. Do not create visual layouts yet.\n```\n\n## Common Pitfalls\n1. Orchestrating without explicit CLI adapters — every agent handoff must declare which CLI\u002Ftransport runs the next step.\n2. Passing context by copy-paste between agents — use markdown handoff files written to disk so each agent reads the same artifact.\n3. Skipping the artifact contract — every handoff must specify input format, output format, and validation step.\n4. Running agents in parallel when they have data dependencies — sequence matters; parallelize only independent sub-tasks.\n5. Letting one agent silently override another's decision — log every agent's output and resolve conflicts via the orchestrator, not by last-write-wins.\n\n## Verification Checklist\n- [ ] Each agent role is named with its CLI adapter (e.g., Claude Code, Codex, Gemini CLI).\n- [ ] Every handoff writes a markdown artifact (e.g., `handoff-ux-to-ui.md`) that the next agent reads.\n- [ ] Artifact contracts specify: required input fields, output format, validation command.\n- [ ] Agent execution order respects data dependencies — independent tasks may parallelize.\n- [ ] Final orchestration log records every agent's input, output, and decision; conflicts are resolved with explicit rationale.\n\n",{"name":5,"author":9,"license":15,"version":24,"metadata":25,"description":7},"2.1.0",{"hermes":26},{"tags":27,"related_skills":33},[28,29,30,31,32],"orchestration","multi-agent","pipeline","handoff","cli",[34,35,36,37,38],"prompt-context-loading","core-system-prompt","ux-decision-framework","code-generation","review-critique","main","142c1bf4c5fa8f8c1bdff6f9620375d429cceb1d","skills\u002Forchestration\u002Fmulti-agent-workflow","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer","https:\u002F\u002Fgithub.com\u002Fgamemun01","2026-10-12 23:45:44",{"source":8,"skillName":5,"commands":46,"prompt":56,"zipAvailable":57},[47,50,53],{"tool":48,"command":49},"npx","npx skills add gamemun01\u002Fai-design-engineer --skill multi-agent-workflow",{"tool":51,"command":52},"bunx","bunx skills add gamemun01\u002Fai-design-engineer --skill multi-agent-workflow",{"tool":54,"command":55},"pnpm","pnpm dlx skills add gamemun01\u002Fai-design-engineer --skill multi-agent-workflow","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Forchestration\u002Fmulti-agent-workflow\u002FSKILL.md",true,[],[60,66,73,80,86,93],{"skillNo":61,"name":36,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":19},"01M7382QVG1QTWD8X9CWD9RBHG","Create UX decision documents from product context, user goals, journeys, task flows, information architecture, accessibility, and trade-offs. Use before wireframes, UI generation, redesign decisions, complex feature planning, or when a user asks for UX rationale.","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fux\u002Fux-decision-framework\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":19},"01M7382KR46P5ET7Z1TC8M9PYB","ui-generation-structured","Generate UI screens, variants, and implementation-ready prompts using the 8-layer prompt architecture and design-system constraints. Use when a UX decision exists and the next step is structured UI generation for dashboards, forms, landing pages, internal tools, or app screens.","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fui\u002Fui-generation-structured\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":19},"01M7382ES5CEZBZTXVQ0VCM3QT","design-system-governance","Enforce design-system rules, spacing scales, typography hierarchy, semantic color roles, accessibility constraints, and component contracts. Use when auditing generated UI, creating design tokens, reviewing component consistency, or preventing arbitrary styles.","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fui\u002Fdesign-system-governance\u002FSKILL.md",[],[],{"skillNo":81,"name":38,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":19},"01M73829S56RVV245HPGMPV1W2","Run a mechanical quality gate for visual quality, UX, accessibility, responsiveness, and engineering readiness using a 0-120 scorecard. Use before shipping UI\u002Fcode, after generation, after refinement, or when the user asks for critique, review, audit, or production readiness.","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fquality\u002Freview-critique\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":19},"01M73824S3Q7337443Q43AD90K","refinement-workflow","Iteratively improve generated UI or code after review findings by fixing high-impact issues, rescoring, and stopping when quality gates pass. Use after review-critique reports scores below threshold, when the user asks to refine or improve output, or when issues must be resolved without broad rewrites.","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fquality\u002Frefinement-workflow\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":19},"01M7382047ANKESKYEA2QEJD64","anti-patterns-detector","Detect AI design and frontend anti-patterns using the repository anti-pattern catalog, including visual, UX, accessibility, code, workflow, and production-readiness issues. Use for pre-shipping audits, recurring problem diagnosis, or when output feels inconsistent, inaccessible, overdesigned, or hard to maintain.","https:\u002F\u002Fgithub.com\u002Fgamemun01\u002Fai-design-engineer\u002Fblob\u002Fmain\u002Fskills\u002Fquality\u002Fanti-patterns-detector\u002FSKILL.md",[],[]]