brainstorm（头脑风暴）

Brainstorming Ideas Into Designs

Transform rough ideas into fully-formed designs through intelligent agent selection and structured exploration.

Core principle: Analyze the topic, select relevant agents dynamically, explore alternatives in parallel, present design incrementally.

Argument Resolution

TOPIC = "$ARGUMENTS"  # Full argument string, e.g., "API design for payments"
# $ARGUMENTS[0] is the first token (CC 2.1.59 indexed access)

STEP -1: MCP Probe + Resume Check

Load: Read("${CLAUDE_PLUGIN_ROOT}/skills/chain-patterns/references/mcp-detection.md")

# 1. Probe MCP servers (once at skill start)
ToolSearch(query="select:mcp__memory__search_nodes")
ToolSearch(query="select:mcp__sequential-thinking__sequentialthinking")

# 2. Store capabilities
Write(".claude/chain/capabilities.json", {
  "memory": probe_memory.found,
  "sequential_thinking": probe_st.found,
  "skill": "brainstorm",
  "timestamp": now()
})

# 3. Check for resume (prior session may have crashed)
state = Read(".claude/chain/state.json")  # may not exist
if state.skill == "brainstorm" and state.status == "in_progress":
    # Skip completed phases, resume from state.current_phase
    last_handoff = Read(f".claude/chain/{state.last_handoff}")

Phase Handoffs

STEP 0: Project Context Discovery

BEFORE creating tasks or selecting agents, detect the project tier. This becomes the complexity ceiling for all downstream decisions.

Auto-Detection (scan codebase)

# PARALLEL — quick signals (launch all in ONE message)
Grep(pattern="take-home|assignment|interview|hackathon", glob="README*", output_mode="content")
Grep(pattern="take-home|assignment|interview|hackathon", glob="*.md", output_mode="content")
Glob(pattern=".github/workflows/*")
Glob(pattern="**/Dockerfile")
Glob(pattern="**/terraform/**")
Glob(pattern="**/k8s/**")
Glob(pattern="CONTRIBUTING.md")

Tier Classification

If confidence is low, ask the user:

AskUserQuestion(questions=[{
  "question": "What kind of project is this?",
  "header": "Project tier",
  "options": [
    {"label": "Interview / take-home", "description": "8-15 files, 200-600 LOC, simple architecture", "markdown": "```\nTier 1: Interview / Take-Home\n─────────────────────────────\nFiles:    8-15 max\nLOC:      200-600\nArch:     Flat structure, no abstractions\nPatterns: Direct imports, inline logic\nTests:    Unit only, co-located\n```"},
    {"label": "Startup / MVP", "description": "MVC monolith, managed services, ship fast", "markdown": "```\nTier 3: Startup / MVP\n─────────────────────\nArch:     MVC monolith\nDB:       Managed (RDS/Supabase)\nCI:       GitHub Actions (1-2 workflows)\nPatterns: Service layer, repository pattern\nDeploy:   Vercel / Railway / Fly.io\n```"},
    {"label": "Growth / enterprise", "description": "Modular monolith or DDD, full observability", "markdown": "```\nTier 4-5: Growth / Enterprise\n─────────────────────────────\nArch:     Modular monolith or DDD\nInfra:    K8s, Terraform, Redis, queues\nCI:       Multi-stage pipelines\nPatterns: Hexagonal, CQRS, event-driven\nObserve:  Structured logging, tracing\n```"},
    {"label": "Open source library", "description": "Minimal API surface, exhaustive tests", "markdown": "```\nTier 6: Open Source Library\n──────────────────────────\nAPI:      Minimal public surface\nTests:    100% coverage, property-based\nDocs:     README, API docs, examples\nCI:       Matrix builds, release automation\nPatterns: Semver, CONTRIBUTING.md\n```"}
  ],
  "multiSelect": false
}])

Pass the detected tier as context to ALL downstream agents and phases. The tier constrains which patterns are appropriate — see scope-appropriate-architecture skill for the full matrix.

Override: User can always override the detected tier. Warn them of trade-offs if they choose a higher tier than detected.

STEP 0a: Verify User Intent with AskUserQuestion

Clarify brainstorming constraints:

AskUserQuestion(
  questions=[
    {
      "question": "What type of design exploration?",
      "header": "Type",
      "options": [
        {"label": "Open exploration (Recommended)", "description": "Generate 10+ ideas, evaluate all, synthesize top 3", "markdown": "```\nOpen Exploration (7 phases)\n──────────────────────────\n  Diverge        Evaluate       Synthesize\n  ┌─────┐       ┌─────┐       ┌─────┐\n  │ 10+ │──────▶│Rate │──────▶│Top 3│\n  │ideas│       │0-10 │       │picks│\n  └─────┘       └─────┘       └─────┘\n  3-5 agents    Devil's        Trade-off\n  in parallel   advocate       table\n```"},
        {"label": "Constrained design", "description": "I have specific requirements to work within", "markdown": "```\nConstrained Design\n──────────────────\n  Requirements ──▶ Feasibility ──▶ Design\n  ┌──────────┐    ┌──────────┐    ┌──────┐\n  │ Fixed    │    │ Check    │    │ Best │\n  │ bounds   │    │ fit      │    │ fit  │\n  └──────────┘    └──────────┘    └──────┘\n  Skip divergent phase, focus on\n  feasibility within constraints\n```"},
        {"label": "Comparison", "description": "Compare 2-3 specific approaches I have in mind", "markdown": "```\nComparison Mode\n───────────────\n  Approach A ──┐\n  Approach B ──┼──▶ Rate 0-10 ──▶ Winner\n  Approach C ──┘    (6 dims)\n\n  Skip ideation, jump straight\n  to evaluation + trade-off table\n```"},
        {"label": "Quick ideation", "description": "Generate ideas fast, skip deep evaluation", "markdown": "```\nQuick Ideation\n──────────────\n  Braindump ──▶ Light filter ──▶ List\n  ┌────────┐   ┌────────────┐   ┌────┐\n  │ 10+    │   │ Viable?    │   │ 5-7│\n  │ ideas  │   │ Y/N only   │   │ out│\n  └────────┘   └────────────┘   └────┘\n  Fast pass, no deep scoring\n```"},
        {"label": "Plan first", "description": "Structured exploration before generating ideas", "markdown": "```\nPlan Mode Exploration\n─────────────────────\n  1. EnterPlanMode($TOPIC)\n  2. Analyze constraints\n  3. Research precedents\n  4. Map solution space\n  5. ExitPlanMode → options\n  6. User picks direction\n  7. Deep dive on chosen path\n\n  Best for: Architecture,\n  design systems, trade-offs\n```"},
        {"label": "Iterative optimization", "description": "Try, measure, keep/discard, repeat (autoresearch-style)", "markdown": "```\nIterative Optimization (autoresearch-style)\n───────────────────────────────────────────\n  ┌──────────┐\n  │ Baseline │──measure──┐\n  └──────────┘           │\n       ┌─────────────────┘\n       ▼\n  ┌─────────┐  ┌─────────┐  ┌──────────┐\n  │ Try     │─▶│ Measure │─▶│ Keep or  │─┐\n  │ variant │  │ metric  │  │ Discard  │ │\n  └─────────┘  └─────────┘  └──────────┘ │\n       ▲                                 │\n       └─────────────────────────────────┘\n  Requires: one command + one metric\n  Runs until: user interrupts or plateau\n```"}
      ],
      "multiSelect": false
    },
    {
      "question": "Any preferences or constraints?",
      "header": "Constraints",
      "options": [
        {"label": "None", "description": "Explore all possibilities"},
        {"label": "Use existing patterns", "description": "Prefer patterns already in codebase"},
        {"label": "Minimize complexity", "description": "Favor simpler solutions"},
        {"label": "I'll specify", "description": "Let me provide specific constraints"}
      ],
      "multiSelect": false
    }
  ]
)

If 'Plan first' selected:

# 1. Enter read-only plan mode
EnterPlanMode("Brainstorm exploration: $TOPIC")

# 2. Research phase — Read/Grep/Glob ONLY, no Write/Edit
#    - Scan existing codebase for related patterns
#    - Search for prior decisions on this topic (memory graph)
#    - Identify constraints, dependencies, and trade-offs

# 3. Produce structured exploration plan:
#    - Key questions to answer
#    - Dimensions to explore
#    - Agents to spawn and their focus areas
#    - Evaluation criteria

# 4. Exit plan mode — returns plan for user approval
ExitPlanMode()

# 5. User reviews. If approved → continue to Phase 1 with plan as input.

Based on answers, adjust workflow:

• Open exploration: Full 7-phase process with all agents
• Constrained design: Skip divergent phase, focus on feasibility
• Comparison: Skip ideation, jump to evaluation phase
• Quick ideation: Generate ideas, skip deep evaluation
• Iterative optimization: Skip phases 2-6, enter autoresearch-style loop (see below)

If 'Iterative optimization' selected:

# 1. Ask for metric definition
AskUserQuestion(questions=[
  {"question": "What command produces the metric?",
   "header": "Metric command",
   "options": [
     {"label": "npm run benchmark", "description": "Node.js benchmark suite"},
     {"label": "pytest --tb=short", "description": "Python test suite"},
     {"label": "lighthouse --output=json", "description": "Web performance score"},
     {"label": "I'll type my own", "description": "Custom command"}
   ]},
  {"question": "How to extract the metric number?",
   "header": "Metric extraction",
   "options": [
     {"label": "grep from stdout", "description": "e.g. grep 'score:' output.log"},
     {"label": "JSON field", "description": "e.g. jq '.score' result.json"},
     {"label": "Exit code", "description": "0 = pass, non-zero = fail"},
     {"label": "I'll specify", "description": "Custom extraction"}
   ]},
  {"question": "Direction?",
   "header": "Optimization direction",
   "options": [
     {"label": "Lower is better", "description": "Latency, bundle size, error rate"},
     {"label": "Higher is better", "description": "Score, throughput, coverage"}
   ]}
])

# 2. Establish baseline
Bash(command="{metric_command} > .claude/experiments/baseline.log 2>&1")
baseline = extract_metric(".claude/experiments/baseline.log")
append_to_journal(baseline, "keep", "-", current_commit, "baseline")

# 3. Enter the optimization loop — see chain-patterns/references/experiment-journal.md
# LOOP (until user interrupts or trajectory == "stuck" for 5+ iterations):
#   a. Generate ONE idea (quick ideation, single agent)
#   b. Implement in worktree: Agent(isolation="worktree", ...)
#   c. Run metric command in worktree
#   d. Compare to previous best
#   e. If improved: merge worktree back, log "keep"
#   f. If not: discard worktree, log "discard"
#   g. Check trajectory — if "stuck" for 5+, try radical changes
#   h. NEVER STOP — continue until user interrupts

STEP 0b: Select Orchestration Mode (skip for Tier 1-2)

Choose Agent Teams (mesh — agents debate and challenge ideas) or Task tool (star — all report to lead):

1. Agent Teams mode (GA since CC 2.1.33) → recommended for 3+ agents (real-time debate produces better ideas)
2. Task tool mode → for quick ideation
3. ORCHESTKIT_FORCE_TASK_TOOL=1 → Task tool (override)

Fallback: If Agent Teams encounters issues, fall back to Task tool for remaining phases.

STEP 0c: Effort-Aware Phase Scaling (CC 2.1.76; xhigh added in 2.1.111)

Read the /effort setting to scale brainstorm depth. The effort-aware context budgeting hook (global) detects effort level automatically — adapt the phase plan accordingly:

# Effort detection — the global hook injects effort level, but also check:
# If user said "quick brainstorm" or "just ideas" → treat as low effort
# If user selected "Quick ideation" in Step 0a → treat as low effort regardless of /effort

Override: Explicit user selection in Step 0a (e.g., "Open exploration") overrides /effort downscaling.

CRITICAL: Task Management is MANDATORY (CC 2.1.16)

# 1. Create main task IMMEDIATELY
TaskCreate(
  subject="Brainstorm: {topic}",
  description="Design exploration with parallel agent research",
  activeForm="Brainstorming {topic}"
)

# 2. Create subtasks for each phase
TaskCreate(subject="Analyze topic and select agents", activeForm="Analyzing topic")          # id=2
TaskCreate(subject="Search memory for past decisions", activeForm="Searching knowledge graph") # id=3
TaskCreate(subject="Generate divergent ideas (10+)", activeForm="Generating ideas")          # id=4
TaskCreate(subject="Feasibility fast-check", activeForm="Checking feasibility")              # id=5
TaskCreate(subject="Evaluate with devil's advocate", activeForm="Evaluating ideas")          # id=6
TaskCreate(subject="Synthesize top approaches", activeForm="Synthesizing approaches")        # id=7
TaskCreate(subject="Present design options", activeForm="Presenting options")                # id=8

# 3. Set dependencies (sequential chain: 2→3→4→5→6→7→8)
for i in range(3, 9):
    TaskUpdate(taskId=str(i), addBlockedBy=[str(i-1)])

# 4. Before starting each task, verify it's unblocked
task = TaskGet(taskId="2")  # Verify blockedBy is empty
# 5. Update status as you progress
TaskUpdate(taskId="2", status="in_progress")  # When starting
TaskUpdate(taskId="2", status="completed")    # When done — repeat for each subtask

The Seven-Phase Process

Progressive Output (CC 2.1.76)

Output results incrementally after each phase — don't batch everything until the end:

For Phase 2 parallel agents, output each agent's ideas as soon as it returns — don't wait for all agents. This lets users see early ideas and redirect the exploration if needed. Showing ideas incrementally also helps users build a mental model of the solution space faster than a final dump.

Load the phase workflow for detailed instructions:

Read("${CLAUDE_SKILL_DIR}/references/phase-workflow.md")

When NOT to Use

Skip brainstorming when:

• Requirements are crystal clear and specific
• Only one obvious approach exists
• User has already designed the solution
• Time-sensitive bug fix or urgent issue

Quick Reference: Agent Selection

Always include: workflow-architect for system design perspective, test-generator for testability assessment.

Agent Teams Alternative: Brainstorming Team

In Agent Teams mode, form a brainstorming team where agents debate ideas in real-time. Dynamically select teammates based on topic analysis (Phase 0):

TeamCreate(team_name="brainstorm-{topic-slug}", description="Brainstorm {topic}")

# Always include the system design lead
Agent(subagent_type="workflow-architect", name="system-designer",
     team_name="brainstorm-{topic-slug}",
     prompt="""You are the system design lead for brainstorming: {topic}
     DIVERGENT MODE: Generate 3-4 architectural approaches.
     When other teammates share ideas, build on them or propose alternatives.
     Challenge ideas that seem over-engineered — advocate for simplicity.
     After divergent phase, help synthesize the top approaches.""")

# Domain-specific teammates (select 2-3 based on topic keywords)
Agent(subagent_type="backend-system-architect", name="backend-thinker",
     team_name="brainstorm-{topic-slug}",
     prompt="""Brainstorm backend approaches for: {topic}
     DIVERGENT MODE: Generate 3-4 backend-specific ideas.
     When system-designer shares architectural ideas, propose concrete API designs.
     Challenge ideas from other teammates with implementation reality checks.
     Play devil's advocate on complexity vs simplicity trade-offs.""")

Agent(subagent_type="frontend-ui-developer", name="frontend-thinker",
     team_name="brainstorm-{topic-slug}",
     prompt="""Brainstorm frontend approaches for: {topic}
     DIVERGENT MODE: Generate 3-4 UI/UX ideas.
     When backend-thinker proposes APIs, suggest frontend patterns that match.
     Challenge backend proposals that create poor user experiences.
     Advocate for progressive disclosure and accessibility.""")

# Always include: testability assessor
Agent(subagent_type="test-generator", name="testability-assessor",
     team_name="brainstorm-{topic-slug}",
     prompt="""Assess testability for each brainstormed approach: {topic}
     For every idea shared by teammates, evaluate:
     - Can core logic be unit tested without external services?
     - What's the mock/stub surface area?
     - Can it be integration-tested with docker-compose/testcontainers?
     Score testability 0-10 per the evaluation rubric.
     Challenge designs that score below 5 on testability.
     Propose test strategies for the top approaches in synthesis phase.""")

# Optional: Add security-auditor, llm-integrator based on topic

Key advantage: Agents riff on each other's ideas and play devil's advocate in real-time, rather than generating ideas in isolation.

Fork pattern (CC 2.1.89 — #1227): All brainstorm agents are fork-eligible: prompts are <500 words, no custom model, no worktree. CC shares the parent's cached API prefix across forks, reducing cost by ~60%. Do NOT add model= to agent calls. See chain-patterns/references/fork-pattern.md.

Team teardown after synthesis:

# After Phase 5 synthesis and design presentation
SendMessage(type="shutdown_request", recipient="system-designer", content="Brainstorm complete")
SendMessage(type="shutdown_request", recipient="backend-thinker", content="Brainstorm complete")
SendMessage(type="shutdown_request", recipient="frontend-thinker", content="Brainstorm complete")
SendMessage(type="shutdown_request", recipient="testability-assessor", content="Brainstorm complete")
# ... shutdown any additional domain teammates
TeamDelete()

# Worktree cleanup (CC 2.1.72) — for Tier 3+ projects that entered a worktree
# If EnterWorktree was called during brainstorm (e.g., Plan first → worktree), exit it
ExitWorktree(action="keep")  # Keep branch for follow-up /ork:implement

Fallback: If team formation fails, load Read("${CLAUDE_SKILL_DIR}/references/phase-workflow.md") and use standard Phase 2 Task spawns.

Partial results (CC 2.1.76): Background agents that are killed (timeout, context limit) return responses tagged with [PARTIAL RESULT]. When collecting Phase 2 divergent ideas, check each agent's output for this tag. If present, include the partial ideas but note them as incomplete in Phase 3 feasibility. Prefer synthesizing partial results over re-spawning agents.

PostCompact recovery: Long brainstorm sessions may trigger context compaction. The PostCompact hook re-injects branch and task state. If compaction occurs mid-brainstorm, check .claude/chain/state.json for the last completed phase and resume from the next handoff file (see Phase Handoffs table).

Session recap (CC 2.1.108+): After idle periods, use /recap to restore conversational context alongside checkpoint-resume. Enabled by default since CC 2.1.110 (even with telemetry disabled).

Push notifications (CC 2.1.110+): For long brainstorm sessions, use PushNotification to alert when synthesis is complete (requires Remote Control with "Push when Claude decides").

Manual cleanup: If TeamDelete() doesn't terminate all agents, press Ctrl+F twice to force-stop remaining background agents. Note: /clear (CC 2.1.72+) preserves background agents — only foreground tasks are cleared.

Key Principles

Related Skills

• ork:architecture-decision-record - Document key decisions made during brainstorming
• ork:implement - Execute the implementation plan after brainstorming completes
• ork:explore - Deep codebase exploration to understand existing patterns
• ork:assess - Rate quality 0-10 with dimension breakdown
• ork:design-to-code - Convert brainstormed UI designs into components
• ork:component-search - Find existing components before building new ones
• ork:competitive-analysis - Porter's Five Forces, SWOT for product brainstorms

References

Load on demand with Read("${CLAUDE_SKILL_DIR}/references/<file>"):