brainstorming-extension

Brainstorming Extension - Mandatory Review Gate Bridge

Purpose

This skill bridges superpowers:brainstorming into the tribunal quality pipeline by enforcing the Design Review Gate after any design document is created. Without this bridge, brainstorming flows directly into writing-plans, bypassing the 5-agent design review that catches architectural, security, and requirements issues before expensive implementation begins.

This is a critical workflow enforcement point, not a passive extension.

---

The Problem This Solves

superpowers:brainstorming has a built-in terminal state: "The ONLY skill you invoke after brainstorming is writing-plans." This creates a pipeline bypass:

WITHOUT this extension (broken flow):

superpowers:brainstorming
    └── Commits design doc
          └── writing-plans (DIRECTLY — no review!)
                └── executing-plans
                      └── Implementation begins on UNREVIEWED design

WITH this extension (correct flow):

superpowers:brainstorming
    └── Commits design doc
          │
          ▼
    ┌─────────────────────────────────────────┐
    │  MANDATORY DESIGN REVIEW GATE           │
    │  (Enforced by CLAUDE.md + start-task)   │
    │                                         │
    │  5 parallel review agents:              │
    │  • Product Manager (use cases)          │
    │  • Architect (architecture)             │
    │  • Designer (UX/API)                    │
    │  • Security Design (threats)            │
    │  • CTO (TDD readiness)                 │
    │                                         │
    │  ALL FIVE must APPROVE                  │
    └─────────────────────────────────────────┘
          │
          ▼
    ALL APPROVED? ────No────► Iterate on design (max 3)
          │
         Yes
          │
          ▼
    writing-plans → plan-review-gate → orchestrated-execution

---

How Enforcement Works

YAML frontmatter triggers (auto_activate, triggers) are metadata hints — Claude Code does not enforce them automatically. Instead, this enforcement works through three redundant mechanisms:

1. CLAUDE.md Template (Primary)

The CLAUDE.md template contains a "Workflow Enforcement (MANDATORY)" section that explicitly states:

After brainstorming: STOP → Run Design Review Gate → Wait for all 5 agents to APPROVE → Only then proceed

This instruction is loaded into every conversation and overrides conflicting skill instructions.

2. start-task Command (Secondary)

The /start-task command's Problem Definition Phase includes an explicit "MANDATORY HANDOFF" block that requires the design review gate after brainstorming.

3. This Skill Document (Reference)

When this skill is loaded (either by name or auto-activation), it provides the detailed procedure below.

---

Procedure: After Brainstorming Completes

When superpowers:brainstorming commits a design document:

Step 1: Announce the Gate

## Design Review Gate Activated

Your design document has been committed. Before proceeding to implementation,
I'll run it through our 5-agent review panel.

Spawning reviews:
- Product Manager Agent (use case/requirements validation)
- Architect Agent (technical architecture)
- Designer Agent (UX/API design)
- Security Design Agent (threat modeling/security review)
- CTO Agent (TDD readiness)

This typically takes 2-3 minutes...

Step 2: Run Debate Phase (if configured)

Check tribunal.yaml for debate configuration. If debate.rounds > 0 and debate.agents is non-empty:

1. Invoke $debate with the design document and research findings
2. Wait for user to select an approach
3. The selected approach replaces the original design document

If debate is not configured (no tribunal.yaml or debate section), skip directly to the design review gate.

Step 3: Invoke the Design Review Gate

Invoke $design-review-gate with the path to the design document. This spawns all 5 review agents in parallel.

Step 4: Report Results

If ALL APPROVED:

## Design Review Gate: PASSED

All five reviewers have approved your design!

| Agent           | Verdict  | Notes                                 |
| --------------- | -------- | ------------------------------------- |
| Product Manager | APPROVED | Clear use cases, measurable benefits  |
| Architect       | APPROVED | Clean architecture, follows patterns  |
| Designer        | APPROVED | Good API design, clear error states   |
| Security Design | APPROVED | No high-risk threats, mitigations OK  |
| CTO             | APPROVED | TDD specs present, ready to implement |

### Next Steps
1. Create epic for this feature
2. Create implementation plan (writing-plans)
3. Run plan through Plan Review Gate (3 adversarial reviewers)
4. Set up worktree for isolated development
5. Begin orchestrated execution

Ready to proceed? [Yes / No]

If ANY NEEDS_REVISION:

## Design Review Gate: NEEDS REVISION

Some reviewers found issues that need to be addressed.

### Blocking Issues
[Agent-specific issues listed here]

### Questions Requiring Answers
[Questions listed here]

---

Please revise the design document and I'll re-run the review gate.
(Iteration 1 of 3)

Step 5: Iterate or Proceed

• If NEEDS_REVISION: Help user address issues, re-run gate (max 3 iterations)
• If ALL APPROVED: Proceed to planning phase
• After 3 failed iterations: Escalate to human decision (Override / Defer / Cancel)

---

Skip Conditions

The review gate can be skipped ONLY when:

• The user explicitly requests it ("skip the review gate")
• AND the agent confirms with the user before skipping
• AND the design is genuinely simple (< 1 day of work, < 3 files)

---

Related Skills

• superpowers:brainstorming — The upstream skill this bridges from
• $design-review-gate — The 5-agent review implementation
• $plan-review-gate — The next gate in the pipeline (after writing-plans)
• $orchestrated-execution — The execution framework after planning