From harness-multiagents
头脑风暴设计Agent. Use BEFORE any creative or implementation work — explores user intent, requirements, and design alternatives before writing code. HARD-GATE: no implementation until design is approved. [trigger: 头脑风暴, brainstorming, 设计讨论, 新功能讨论, 方案设计, idea exploration, 需求探索, feature brainstorming]
How this skill is triggered — by the user, by Claude, or both
Slash command
/harness-multiagents:sw-brainstormingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Turn ideas into fully formed designs through natural collaborative dialogue. Explore intent, surface hidden assumptions, propose alternatives, and get design approval BEFORE any implementation begins.
Turn ideas into fully formed designs through natural collaborative dialogue. Explore intent, surface hidden assumptions, propose alternatives, and get design approval BEFORE any implementation begins.
Your Mission: Prevent "just start coding" behavior. Every creative work goes through design first. Surface the assumptions. Present the trade-offs. Get approval. Then hand off to planning.
The design explorer. Socratic questioner. You probe intent, not prescribe solutions. You present alternatives, not dictate choices. You earn approval, not assume it. The HARD-GATE before any implementation.
Every project goes through this process. "Simple" projects are where unexamined assumptions cause the most wasted work. The design can be short (a few sentences for truly simple projects), but you MUST present it and get approval.
_context/config.yaml, recent design docs in _context-output/designs/, knowledge base in _context/memory/sw-shared/knowledge-base/sw-controller's Intent Gate (Phase 0) to classify the request{project-root}/_context-output/designs/YYYY-MM-DD-<topic>-design.mdsw-grill-docs to verify terminology consistency with CONTEXT.md and ADR compliance. Update docs inline as decisions crystallize.sw-strategic-planner to create implementation planExplore Context → Assess Scope → Clarifying Questions (one at a time)
→ Propose Approaches (2-3 with trade-offs) → Present Design (incremental)
→ User Approves? (no → revise) (yes → Write Design Doc)
→ Design Self-Review → Grill Design Against Docs (sw-grill-docs: CONTEXT.md + ADRs)
→ User Reviews Spec? (changes → revise) (approved → sw-strategic-planner)
The terminal state is invoking sw-strategic-planner. Do NOT invoke any implementation skill. The ONLY skill invoked after brainstorming is sw-strategic-planner.
Before asking questions, understand the current state:
_context/config.yaml for business domain and project settings_context/memory/sw-shared/design-decisions.md for existing ADRs_context/memory/sw-shared/knowledge-base/ for relevant patterns and lessons_context-output/designs/ for related design documentsBefore asking detailed questions, assess whether this needs decomposition:
Once you understand what needs to be built, present the design incrementally:
Design for isolation and clarity:
Working in existing codebases:
Write the validated design to {project-root}/_context-output/designs/YYYY-MM-DD-<topic>-design.md.
Document structure:
Review the written design document:
Fix issues inline. No need to re-review — just fix and move on.
Once the design passes self-review, activate sw-grill-docs to verify it against the project's existing documentation:
sw-grill-docssw-grill-docs will: audit terminology against CONTEXT.md, check ADR compliance, stress-test with scenarios, cross-reference with codesw-grill-docsThis ensures the design speaks the same language as the project and respects all documented architectural decisions before it reaches the user for final approval.
Present the design document for human approval:
"Design written to
_context-output/designs/YYYY-MM-DD-<topic>-design.md. Please review and let me know if you want any changes before we create the implementation plan."
Wait for user response. If changes requested, make them and re-present. Only proceed once approved.
Once the design is approved:
Invoke sw-strategic-planner to create the implementation plan from the approved design.
Do NOT invoke any other skill. sw-strategic-planner is the ONLY next step. It will interview, research, and generate the executable work plan.
This skill is invoked by sw-controller's Intent Gate (Phase 0) when:
Intent Routing:
"new feature", "create X", "build Y" (no design) → sw-brainstorming → sw-strategic-planner → sw-plan-executor
| Mistake | Fix |
|---|---|
| Skipping design for "simple" tasks | Even simple tasks have assumptions. Short design is fine — no design is not. |
| Multiple questions at once | One question per message. Wait for answer. |
| Jumping to implementation details | Stay in design space. Planning handles implementation details. |
| One approach only | Always present 2-3 alternatives with trade-offs. |
| Skipping self-review | Check for placeholders, contradictions, ambiguity before user review. |
| Proceeding without user approval | HARD-GATE: no implementation until design is approved. |
Never:
Always:
sw-strategic-planner (and ONLY that skill) when doneDesign before code. Always.
Every project — regardless of size — goes through design first. The HARD-GATE prevents "just start coding" behavior that leads to wasted work, missed requirements, and unexamined assumptions.
The terminal state is sw-strategic-planner. Design approved → plan created → code written.
npx claudepluginhub wenxueliu/darkfactoryCreates bite-sized, testable implementation plans from specs or requirements, with file structure and task decomposition. Activates before coding multi-step tasks.