brainstorm

Name: brainstorm
Author: Aedelon

108stars34forks

How to install brainstorm?

aedelon/claude-code-blueprint/brainstorm

$npx -y skills add Aedelon/claude-code-blueprint --skill brainstorm

Installs into the current project.

Summary

Multi-agent brainstorming and strategic thinking on any complex topic. Explores the subject deeply, then spawns specialized sub-agents in parallel for diverse expert perspectives. MUST BE USED PROACTIVELY when user faces any decision, architectural choice, design question, strate

Files · 1

View on GitHub

SKILL.md

1	# Multi-Agent Brainstorm
2
3	## Overview
4
5	Structured brainstorming that combines clarifying dialogue with parallel expert sub-agents.
6	The goal: understand the problem first, then get diverse expert perspectives, then synthesize.
7
8	## Process
9
10	```
11	Phase 1: CLARIFY → Understand the topic (2-3 targeted questions)
12	Phase 2: MAP → Select relevant expert agents
13	Phase 3: EXPLORE → Spawn agents in parallel
14	Phase 4: SYNTHESIZE → Combine and present findings
15	```
16
17	## Phase 1: Clarify the Topic
18
19	Before spawning any agent, understand what the user actually needs.
20
21	Ask 2-3 targeted questions (one at a time, prefer multiple choice):
22
23	- What is the core problem or goal?
24	- What constraints exist? (tech stack, budget, timeline, existing code)
25	- What does success look like?
26
27	Rules:
28	- One question per message
29	- Prefer multiple choice via AskUserQuestion when possible
30	- Max 3 questions — don't interrogate, get enough context to spawn useful agents
31	- If the topic is already clear and specific, skip to Phase 2
32
33	## Phase 2: Map to Expert Agents
34
35	Select 2-4 relevant agents from this mapping:
36
37	\| Domain \| Agent Type \|
38	\|--------\|------------\|
39	\| AI/ML Research \| research-synthesizer \|
40	\| Creative / Image Gen \| midjourney-expert \|
41	\| Finance/Crypto/Tax \| finance-advisor \|
42	\| Prompt Engineering \| prompt-engineer \|
43	\| Architecture/Design \| feature-dev:code-architect \|
44	\| Code Review \| feature-dev:code-reviewer \|
45	\| General research \| general-purpose \|
46
47	Selection criteria:
48	- Pick agents that bring different perspectives on the topic
49	- At least one agent should challenge the obvious approach
50	- For cross-domain topics, pick agents from different domains
51
52	## Phase 3: Explore via Parallel Agents
53
54	Spawn all selected agents simultaneously using the Agent tool.
55
56	Each agent prompt MUST include:
57
58	```
59	Brainstorm: [topic with context from Phase 1]
60
61	Context: [key constraints and goals identified]
62
63	Your task:
64	1. Propose 2-3 approaches from your domain expertise
65	2. For each approach: trade-offs, risks, and when it works best
66	3. Justify every recommendation with sources (docs, papers, experience)
67	4. State confidence level (HIGH/MEDIUM/LOW) for each claim
68	5. Flag anything you cannot verify
69	6. Highlight what you'd do differently from the obvious approach
70	```
71
72	Important:
73	- Use `model: "opus"` for complex/architectural topics
74	- Use `model: "sonnet"` for straightforward domain questions
75	- All agents run in parallel — never sequential
76
77	## Phase 4: Synthesize
78
79	After all agents return, present a unified synthesis:
80
81	```markdown
82	## Synthesis: [Topic]
83
84	### Consensus
85	[What multiple agents agree on — strongest signal]
86
87	### Divergent Views
88	[Where agents disagree, with rationale from each side]
89
90	### Recommended Approach
91	[Best path forward, justified by agent findings]
92	[State which agent perspectives informed this recommendation]
93
94	### Trade-offs to Consider
95	[Key decisions the user still needs to make]
96
97	### Sources
98	[All sources cited by agents, deduplicated]
99
100	### Confidence: [LEVEL]
101	[Based on source agreement and verification quality]
102	```
103
104	## Edge Cases
105
106	### Topic too vague
107	Ask one clarifying question. If still vague after 2 questions, brainstorm at the level of abstraction given.
108
109	### No obvious agent match
110	Use `general-purpose` agents with specific domain prompts. Always spawn at least 2.
111
112	### User wants to go deeper
113	After synthesis, offer to spawn additional agents on specific sub-topics that emerged.
114
115	### Quick brainstorm
116	If user says "quick" or the topic is narrow, skip Phase 1 and spawn 2 agents directly.

People who liked this also viewed

6 picks

Type

brainstorm

108stars34forks

View on GitHub

How to install brainstorm?

aedelon/claude-code-blueprint/brainstorm

$npx -y skills add Aedelon/claude-code-blueprint --skill brainstorm

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Multi-Agent Brainstorm
2
3	## Overview
4
5	Structured brainstorming that combines clarifying dialogue with parallel expert sub-agents.
6	The goal: understand the problem first, then get diverse expert perspectives, then synthesize.
7
8	## Process
9
10	```
11	Phase 1: CLARIFY → Understand the topic (2-3 targeted questions)
12	Phase 2: MAP → Select relevant expert agents
13	Phase 3: EXPLORE → Spawn agents in parallel
14	Phase 4: SYNTHESIZE → Combine and present findings
15	```
16
17	## Phase 1: Clarify the Topic
18
19	Before spawning any agent, understand what the user actually needs.
20
21	Ask 2-3 targeted questions (one at a time, prefer multiple choice):
22
23	- What is the core problem or goal?
24	- What constraints exist? (tech stack, budget, timeline, existing code)
25	- What does success look like?
26
27	Rules:
28	- One question per message
29	- Prefer multiple choice via AskUserQuestion when possible
30	- Max 3 questions — don't interrogate, get enough context to spawn useful agents
31	- If the topic is already clear and specific, skip to Phase 2
32
33	## Phase 2: Map to Expert Agents
34
35	Select 2-4 relevant agents from this mapping:
36
37	\| Domain \| Agent Type \|
38	\|--------\|------------\|
39	\| AI/ML Research \| research-synthesizer \|
40	\| Creative / Image Gen \| midjourney-expert \|
41	\| Finance/Crypto/Tax \| finance-advisor \|
42	\| Prompt Engineering \| prompt-engineer \|
43	\| Architecture/Design \| feature-dev:code-architect \|
44	\| Code Review \| feature-dev:code-reviewer \|
45	\| General research \| general-purpose \|
46
47	Selection criteria:
48	- Pick agents that bring different perspectives on the topic
49	- At least one agent should challenge the obvious approach
50	- For cross-domain topics, pick agents from different domains
51
52	## Phase 3: Explore via Parallel Agents
53
54	Spawn all selected agents simultaneously using the Agent tool.
55
56	Each agent prompt MUST include:
57
58	```
59	Brainstorm: [topic with context from Phase 1]
60
61	Context: [key constraints and goals identified]
62
63	Your task:
64	1. Propose 2-3 approaches from your domain expertise
65	2. For each approach: trade-offs, risks, and when it works best
66	3. Justify every recommendation with sources (docs, papers, experience)
67	4. State confidence level (HIGH/MEDIUM/LOW) for each claim
68	5. Flag anything you cannot verify
69	6. Highlight what you'd do differently from the obvious approach
70	```
71
72	Important:
73	- Use `model: "opus"` for complex/architectural topics
74	- Use `model: "sonnet"` for straightforward domain questions
75	- All agents run in parallel — never sequential
76
77	## Phase 4: Synthesize
78
79	After all agents return, present a unified synthesis:
80
81	```markdown
82	## Synthesis: [Topic]
83
84	### Consensus
85	[What multiple agents agree on — strongest signal]
86
87	### Divergent Views
88	[Where agents disagree, with rationale from each side]
89
90	### Recommended Approach
91	[Best path forward, justified by agent findings]
92	[State which agent perspectives informed this recommendation]
93
94	### Trade-offs to Consider
95	[Key decisions the user still needs to make]
96
97	### Sources
98	[All sources cited by agents, deduplicated]
99
100	### Confidence: [LEVEL]
101	[Based on source agreement and verification quality]
102	```
103
104	## Edge Cases
105
106	### Topic too vague
107	Ask one clarifying question. If still vague after 2 questions, brainstorm at the level of abstraction given.
108
109	### No obvious agent match
110	Use `general-purpose` agents with specific domain prompts. Always spawn at least 2.
111
112	### User wants to go deeper
113	After synthesis, offer to spawn additional agents on specific sub-topics that emerged.
114
115	### Quick brainstorm
116	If user says "quick" or the topic is narrow, skip Phase 1 and spawn 2 agents directly.

People who liked this also viewed

6 picks

Type