Multi Agent Brainstorming

1---
2name: multi-agent-brainstorming
3description: >
4  Use this skill when a design or idea requires higher confidence,
5  risk reduction, or formal review. This skill orchestrates a
6  structured, sequential multi-agent design review where each agent
7  has a strict, non-overlapping role. It prevents blind spots,
8  false confidence, and premature convergence.
9---
10 
11# Multi-Agent Brainstorming (Structured Design Review)
12 
13## Purpose
14 
15Transform a single-agent design into a **robust, review-validated design**
16by simulating a formal peer-review process using multiple constrained agents.
17 
18This skill exists to:
19- surface hidden assumptions
20- identify failure modes early
21- validate non-functional constraints
22- stress-test designs before implementation
23- prevent idea swarm chaos
24 
25This is **not parallel brainstorming**.
26It is **sequential design review with enforced roles**.
27 
28---
29 
30## Operating Model
31 
32- One agent designs.
33- Other agents review.
34- No agent may exceed its mandate.
35- Creativity is centralized; critique is distributed.
36- Decisions are explicit and logged.
37 
38The process is **gated** and **terminates by design**.
39 
40---
41 
42## Agent Roles (Non-Negotiable)
43 
44Each agent operates under a **hard scope limit**.
45 
46### 1️⃣ Primary Designer (Lead Agent)
47 
48**Role:**
49- Owns the design
50- Runs the standard `brainstorming` skill
51- Maintains the Decision Log
52 
53**May:**
54- Ask clarification questions
55- Propose designs and alternatives
56- Revise designs based on feedback
57 
58**May NOT:**
59- Self-approve the final design
60- Ignore reviewer objections
61- Invent requirements post-lock
62 
63---
64 
65### 2️⃣ Skeptic / Challenger Agent
66 
67**Role:**
68- Assume the design will fail
69- Identify weaknesses and risks
70 
71**May:**
72- Question assumptions
73- Identify edge cases
74- Highlight ambiguity or overconfidence
75- Flag YAGNI violations
76 
77**May NOT:**
78- Propose new features
79- Redesign the system
80- Offer alternative architectures
81 
82Prompting guidance:
83> “Assume this design fails in production. Why?”
84 
85---
86 
87### 3️⃣ Constraint Guardian Agent
88 
89**Role:**
90- Enforce non-functional and real-world constraints
91 
92Focus areas:
93- performance
94- scalability
95- reliability
96- security & privacy
97- maintainability
98- operational cost
99 
100**May:**
101- Reject designs that violate constraints
102- Request clarification of limits
103 
104**May NOT:**
105- Debate product goals
106- Suggest feature changes
107- Optimize beyond stated requirements
108 
109---
110 
111### 4️⃣ User Advocate Agent
112 
113**Role:**
114- Represent the end user
115 
116Focus areas:
117- cognitive load
118- usability
119- clarity of flows
120- error handling from user perspective
121- mismatch between intent and experience
122 
123**May:**
124- Identify confusing or misleading aspects
125- Flag poor defaults or unclear behavior
126 
127**May NOT:**
128- Redesign architecture
129- Add features
130- Override stated user goals
131 
132---
133 
134### 5️⃣ Integrator / Arbiter Agent
135 
136**Role:**
137- Resolve conflicts
138- Finalize decisions
139- Enforce exit criteria
140 
141**May:**
142- Accept or reject objections
143- Require design revisions
144- Declare the design complete
145 
146**May NOT:**
147- Invent new ideas
148- Add requirements
149- Reopen locked decisions without cause
150 
151---
152 
153## The Process
154 
155### Phase 1 — Single-Agent Design
156 
1571. Primary Designer runs the **standard `brainstorming` skill**
1582. Understanding Lock is completed and confirmed
1593. Initial design is produced
1604. Decision Log is started
161 
162No other agents participate yet.
163 
164---
165 
166### Phase 2 — Structured Review Loop
167 
168Agents are invoked **one at a time**, in the following order:
169 
1701. Skeptic / Challenger
1712. Constraint Guardian
1723. User Advocate
173 
174For each reviewer:
175- Feedback must be explicit and scoped
176- Objections must reference assumptions or decisions
177- No new features may be introduced
178 
179Primary Designer must:
180- Respond to each objection
181- Revise the design if required
182- Update the Decision Log
183 
184---
185 
186### Phase 3 — Integration & Arbitration
187 
188The Integrator / Arbiter reviews:
189- the final design
190- the Decision Log
191- unresolved objections
192 
193The Arbiter must explicitly decide:
194- which objections are accepted
195- which are rejected (with rationale)
196 
197---
198 
199## Decision Log (Mandatory Artifact)
200 
201The Decision Log must record:
202 
203- Decision made
204- Alternatives considered
205- Objections raised
206- Resolution and rationale
207 
208No design is considered valid without a completed log.
209 
210---
211 
212## Exit Criteria (Hard Stop)
213 
214You may exit multi-agent brainstorming **only when all are true**:
215 
216- Understanding Lock was completed
217- All reviewer agents have been invoked
218- All objections are resolved or explicitly rejected
219- Decision Log is complete
220- Arbiter has declared the design acceptable
221- 
222If any criterion is unmet:
223- Continue review
224- Do NOT proceed to implementation
225If this skill was invoked by a routing or orchestration layer, you MUST report the final disposition explicitly as one of: APPROVED, REVISE, or REJECT, with a brief rationale.
226---
227 
228## Failure Modes This Skill Prevents
229 
230- Idea swarm chaos
231- Hallucinated consensus
232- Overconfident single-agent designs
233- Hidden assumptions
234- Premature implementation
235- Endless debate
236 
237---
238 
239## Key Principles
240 
241- One designer, many reviewers
242- Creativity is centralized
243- Critique is constrained
244- Decisions are explicit
245- Process must terminate
246 
247---
248 
249## Final Reminder
250 
251This skill exists to answer one question with confidence:
252 
253> “If this design fails, did we do everything reasonable to catch it early?”
254 
255If the answer is unclear, **do not exit this skill**.
256 
257