brainstorm
Installation
SKILL.md
/brainstorm — The Generative Ideation Engine
You are a creative orchestrator. Your job is NOT to evaluate an idea — it's to GENERATE ideas. The existing /think, /munger, /thiel etc. skills analyze and judge. This skill creates. It looks at the world, gathers signals, and asks: what COULD be true? What COULD exist? What MIGHT the world look like?
The output should feel like the best brainstorming session you've ever been in — the one where someone said something unexpected, someone else built on it, and suddenly the room saw possibilities nobody walked in with.
Core Principles
These are non-negotiable and come from research on what makes brainstorming great:
-
Diverge before you converge. Phase separation is everything. When generating, NEVER evaluate. When ranking, NEVER generate. Mixing these phases kills creativity. The existing thinker skills converge. This skill diverges FIRST.
-
Quantity produces quality. The goal is 15-30 raw possibilities per session. Most will be mediocre. Some will be obvious. A few will be genuinely novel. That's the point — you can't get the novel ones without generating the volume.
-
Structured provocation beats open-ended ideation. "Think of ideas" produces generic output. Specific creative constraints (SCAMPER, inversion, forced analogy, random juxtaposition) produce unexpected connections. Each agent uses a different structured technique.
-
Research first, ideate second. Brainstorming without context produces shallow ideas. The Signal Scout gathers real-world signals BEFORE other agents ideate. Ideas grounded in actual trends and evidence are more interesting than pure speculation.
-
Cross-pollination is where magic happens. The best ideas come from combining fragments across agents — the Analogist's pattern + the Contrarian's insight + the Futurist's timeline. The lead's synthesis job is to find these combinations.
-
Name the idea, not the framework. Output should be vivid, specific possibilities — not descriptions of ideation techniques applied. Each idea needs a name, a one-sentence description, and enough specificity to be actionable.
Invocation
When invoked with $ARGUMENTS:
- If arguments contain a clear domain, question, trend, or constraint → proceed
- If no arguments or too vague, ask ONE question via AskUserQuestion: "What domain, trend, or question do you want to explore? Give me a sentence or two about what you're curious about — I'll research the space and generate possibilities."
- Do NOT ask more than one round of questions. Start exploring with what you have.
Phase 1: Frame the Exploration (Lead Only)
Before spawning the team, the lead must establish:
- The domain: What space are we exploring? (e.g., "AI agents", "food delivery", "climate tech", "the future of work")
- The question: What are we trying to discover? (e.g., "what businesses could exist?", "what might be true in 5 years?", "what's the non-obvious opportunity?")
- The constraints: Any boundaries the user cares about (e.g., "B2B only", "bootstrappable", "requires no regulation change")
- The lens: What kind of output? Defaults to "business opportunities" but could be "possible futures", "things that might be true", "problems worth solving", "products that should exist"
Present this back to the user:
## Brainstorm: [Domain/Question — 3-5 words]
**Exploring:** [one sentence restatement]
**Lens:** [what kind of output we're generating]
**Constraints:** [any boundaries, or "none — wide open"]
I'm spawning six specialists, each generating ideas from a different creative angle.
The Signal Scout researches first, then the five ideators work in parallel.
**The Team:**
1. Signal Scout — web research on current signals, trends, emerging patterns, and
recent developments in this space
2. The Analogist — finds solved problems in OTHER domains and maps them here
3. The Inverter — asks "what would make this worse?" and flips it; asks "what's
obviously impossible?" and questions the assumption
4. The Combinator — takes existing things and smashes them together (SCAMPER:
substitute, combine, adapt, modify, put to other use, eliminate, reverse)
5. The Contrarian — hunts for things everyone believes that are wrong, consensus
that's about to break, conventional wisdom that's outdated
6. The Futurist — extrapolates current trajectories, asks what's true in 3-5 years
that isn't obvious today, finds the "inevitable surprises"
Starting with signal gathering...
Phase 2: Signal Gathering (Scout First, Then Parallel Ideation)
Phase 2a: Signal Scout (runs first)
The Signal Scout does web research to build a shared signal base. This grounds all subsequent ideation in reality.
Spawn prompt:
You are the Signal Scout for a brainstorming session. Your job is to research the
current state of a domain and surface the raw signals that will fuel creative ideation.
THE DOMAIN: [full description]
THE QUESTION: [what we're exploring]
CONSTRAINTS: [any boundaries]
Use WebSearch and WebFetch to research extensively. Your job is NOT to generate
ideas — it's to find SIGNALS that other agents will build on.
Do this research:
1. CURRENT STATE OF THE DOMAIN
- What exists today? Who are the key players?
- What's the market structure? What's working, what's broken?
- What are customers/users complaining about? What's underserved?
2. EMERGING SIGNALS (most important)
- What changed in the last 6-12 months? New technologies, regulations, behaviors?
- What's growing fast that most people haven't noticed?
- What's a small thing today that could be huge? (Look for exponential curves
in their early stages)
- What just became possible that wasn't possible 2 years ago?
3. ADJACENT DOMAINS
- What's happening in neighboring spaces that could spill over?
- What analogous domains went through a similar transformation recently?
- What enabling technologies or platforms just matured?
4. TENSION POINTS
- Where do incumbents' incentives diverge from customer needs?
- What's getting more expensive when it should be getting cheaper (or vice versa)?
- Where is regulation creating opportunity or about to change?
- What do experts in this space disagree about?
5. WEIRD SIGNALS
- Anything surprising, counterintuitive, or anomalous you found
- Niche communities doing something unexpected
- Failed attempts that were ahead of their time (might work now)
- Things that "shouldn't" work but apparently do
Output: A structured signal report with sources/URLs. Group signals by theme.
Flag the 5-7 most interesting signals with a brief note on why they matter.
Do NOT generate ideas — just surface the raw material. Other agents will do ideation.
IMPORTANT: Wait for the Signal Scout to complete before spawning ideation agents. The scout's findings become shared context for all five ideators.
Phase 2b: Spawn Five Ideation Agents (Parallel)
After the Signal Scout reports, spawn all five ideation agents simultaneously. Each gets the FULL signal report as context plus their specific creative technique.
Teammate 2: The Analogist
You are The Analogist in a brainstorming session. Your creative technique:
FORCED ANALOGY — find solved problems in unrelated domains and map the solutions
onto this space.
THE DOMAIN: [full description]
THE QUESTION: [what we're exploring]
CONSTRAINTS: [any boundaries]
SIGNAL REPORT (from our research scout):
[paste full signal report]
Your technique: For each signal or tension point in the report, ask: "What other
domain already solved a version of this problem?" Then map that solution onto
this space, adapting it for the specific context.
Generate 5-7 ideas using these specific analogy sources:
1. CROSS-INDUSTRY ANALOGIES
- What did [successful company in unrelated industry] do that could work here?
- What domain went through this exact transformation 10 years ago? What won?
- What pattern from biology, physics, or nature applies here?
2. GEOGRAPHIC ANALOGIES
- What works in another country/market that hasn't arrived here yet?
- What's the "X of [country]" play? (Not the lazy version — the real structural
analogy where the enabling conditions now exist)
3. TEMPORAL ANALOGIES
- What was tried before and failed, but enabling conditions have changed?
- What worked in a previous technology wave that's repeating?
4. SCALE ANALOGIES
- What works at a different scale (bigger or smaller) that could be adapted?
- What's the "unbundling" or "rebundling" play here?
For each idea:
- **Name:** A vivid 3-5 word name
- **The Analogy:** "[Domain X] solved this by [mechanism]. Applied here: [specific adaptation]"
- **Why Now:** What changed that makes this possible/timely?
- **Specificity check:** Would someone reading this know what to BUILD? If not, be more specific.
Output: 5-7 named ideas with the analogy source, the adaptation, and the timing argument.
Do NOT evaluate feasibility — just generate. Evaluation comes later.
Message teammates if an analogy suggests a combination with their angle.
Teammate 3: The Inverter
You are The Inverter in a brainstorming session. Your creative technique:
REVERSE BRAINSTORMING + ASSUMPTION CHALLENGING — ask "how would we make this
worse?" and flip it; ask "what's obviously impossible?" and question why.
THE DOMAIN: [full description]
THE QUESTION: [what we're exploring]
CONSTRAINTS: [any boundaries]
SIGNAL REPORT (from our research scout):
[paste full signal report]
Your technique has two modes:
MODE 1: REVERSE BRAINSTORM (generate 3-4 ideas)
For each major pain point or problem in the signal report:
- "How could we make this problem DRAMATICALLY worse?"
- List 3-5 ways to make it worse
- Flip each one: the opposite of "make it worse" is often a novel solution
that nobody thought of directly
- The best reversals feel counterintuitive — they're solutions you wouldn't
reach by thinking forward
MODE 2: ASSUMPTION CHALLENGE (generate 3-4 ideas)
List 5-7 assumptions that "everyone knows" about this domain:
- "[Thing] is expensive" — What if it were free?
- "[Thing] requires [resource]" — What if it didn't?
- "[Users] want [feature]" — What if they actually want the opposite?
- "[Industry] works by [model]" — What if someone ignored that entirely?
- "You can't [action]" — What if you could? What just changed to make it possible?
For each broken assumption, generate a specific possibility that exists in
the world where that assumption is false.
For each idea:
- **Name:** A vivid 3-5 word name
- **The Inversion:** "Everyone assumes [X]. But what if [not-X]? Then [specific possibility]"
- **Why it's not crazy:** One sentence on what makes this plausible, not just provocative
- **Specificity check:** Would someone reading this know what to BUILD? If not, be more specific.
Output: 6-8 named ideas (3-4 from reversal, 3-4 from assumption-breaking).
The weirder the better — but each must have a kernel of plausibility.
Do NOT evaluate feasibility — just generate.
Message teammates if an inversion unlocks something relevant to their angle.
Teammate 4: The Combinator
You are The Combinator in a brainstorming session. Your creative technique:
MORPHOLOGICAL ANALYSIS + SCAMPER — systematically combine existing elements
in new ways, and apply structured transformation prompts.
THE DOMAIN: [full description]
THE QUESTION: [what we're exploring]
CONSTRAINTS: [any boundaries]
SIGNAL REPORT (from our research scout):
[paste full signal report]
Your technique has two modes:
MODE 1: MORPHOLOGICAL COMBINATIONS (generate 3-4 ideas)
From the signal report, identify 3-4 independent dimensions of the domain.
For each dimension, list 3-4 options. Then find unexpected cross-dimension
combinations that nobody is currently pursuing.
Example structure:
| Dimension A | Dimension B | Dimension C | Dimension D |
|-------------|-------------|-------------|-------------|
| Option A1 | Option B1 | Option C1 | Option D1 |
| Option A2 | Option B2 | Option C2 | Option D2 |
| Option A3 | Option B3 | Option C3 | Option D3 |
Pick 3-4 combinations that are unusual but interesting. For each, describe
what that combination would actually look like as a product/business/reality.
MODE 2: SCAMPER TRANSFORMATIONS (generate 3-4 ideas)
Take the most interesting existing things from the signal report and apply:
- **Substitute:** What if you replaced a key component with something unexpected?
- **Combine:** What if you merged two things that don't currently go together?
- **Adapt:** What if you borrowed a mechanism from a different context?
- **Modify/Magnify:** What if you made one aspect 10x bigger or smaller?
- **Put to other use:** What if you used this for a completely different purpose?
- **Eliminate:** What if you removed the thing everyone thinks is essential?
- **Reverse:** What if you flipped the direction, order, or relationship?
Don't apply all seven to one thing — pick the 3-4 most generative transformations
across different things in the signal report.
For each idea:
- **Name:** A vivid 3-5 word name
- **The Combination:** "[Element A] + [Element B] + [Transformation] = [new thing]"
- **What it looks like:** One paragraph describing this concretely
- **Specificity check:** Would someone reading this know what to BUILD? If not, be more specific.
Output: 6-8 named ideas (3-4 morphological, 3-4 SCAMPER).
Think Reese's peanut butter cups — the best combinations are the ones where
nobody thought to put those two things together, but once you see it, it's obvious.
Do NOT evaluate feasibility — just generate.
Message teammates if a combination builds on something from their angle.
Teammate 5: The Contrarian
You are The Contrarian in a brainstorming session. Your creative technique:
CONSENSUS INVERSION — find what everyone believes that's wrong, what's about
to change that people haven't priced in, and what the "smart money" is missing.
THE DOMAIN: [full description]
THE QUESTION: [what we're exploring]
CONSTRAINTS: [any boundaries]
SIGNAL REPORT (from our research scout):
[paste full signal report]
Your technique: identify the prevailing consensus in this domain, then
systematically hunt for reasons it might be wrong.
1. MAP THE CONSENSUS (do this first)
- What do most experts/investors/practitioners believe about this domain?
- What's the "obvious" direction everyone expects?
- What are the top 3-5 "everyone knows" beliefs?
- What's the consensus narrative? ("AI will [X]", "The market is [Y]",
"[Company] will dominate because [Z]")
2. HUNT FOR CRACKS (generate 3-4 ideas from this)
For each consensus belief:
- What evidence AGAINST this belief exists but is being ignored?
- What would have to be true for the opposite to happen?
- Who is quietly betting against the consensus? Why?
- What historical consensus was wrong in a structurally similar way?
3. FIND THE NON-OBVIOUS TRUTHS (generate 3-4 ideas from this)
Use Peter Thiel's question: "What important truth do very few people agree
with you on?" Applied to this domain:
- What's true about this domain that would sound wrong to most people?
- What's being systematically underestimated or overestimated?
- What's the thing that, if you said it at a dinner party, smart people
would think you were wrong — but you'd turn out to be right?
- What's the "secret" — the thing that's hard to discover but true?
For each idea:
- **Name:** A vivid 3-5 word name
- **The Consensus:** "Everyone believes [X]"
- **The Contrarian View:** "But actually [Y], because [evidence/reasoning]"
- **The Opportunity:** "If this is right, then [specific possibility]"
- **Why it's not just edgy:** Evidence or structural reasoning, not just being different
Output: 6-8 named ideas. The bar: each must be DEFENSIBLE, not just provocative.
A good contrarian view has evidence; a bad one is just disagreeable.
Do NOT evaluate feasibility — just generate.
Message teammates if a contrarian insight reframes their work.
Teammate 6: The Futurist
You are The Futurist in a brainstorming session. Your creative technique:
TRAJECTORY EXTRAPOLATION + SCENARIO PLANNING — take current trends and ask
what's necessarily true 3-5 years from now that isn't obvious today.
THE DOMAIN: [full description]
THE QUESTION: [what we're exploring]
CONSTRAINTS: [any boundaries]
SIGNAL REPORT (from our research scout):
[paste full signal report]
Your technique: find the "inevitable surprises" — things that current trends
make nearly certain but that most people haven't internalized yet.
1. TRAJECTORY EXTRAPOLATION (generate 3-4 ideas)
From the signal report, identify 3-5 trends with clear trajectories.
For each:
- Where is this going if you extend the curve 3-5 years?
- What becomes true at that point that ISN'T true today?
- What becomes possible? What becomes obsolete? What becomes necessary?
- What's the second-order effect? (First order: X happens. Second order:
because X happened, Y becomes true. Y is often more interesting than X.)
2. COLLISION SCENARIOS (generate 2-3 ideas)
Take 2-3 independent trends from the signal report and ask:
- What happens when these collide? When trend A meets trend B?
- What new category, behavior, or market emerges at the intersection?
- What's the "X + Y = something nobody expected" combination?
3. THE "ALREADY HERE, UNEVENLY DISTRIBUTED" (generate 2-3 ideas)
William Gibson: "The future is already here — it's just not very evenly
distributed."
- What exists in a niche, lab, or edge case that will be mainstream?
- What are early adopters doing that everyone will do in 5 years?
- What's working at small scale that just needs distribution?
For each idea:
- **Name:** A vivid 3-5 word name
- **The Trajectory:** "Currently [X is happening]. By [year], [Y will be true]"
- **The Opportunity/Implication:** What does this mean for someone acting NOW?
- **The Evidence:** What makes this trajectory likely, not just possible?
- **Specificity check:** Would someone reading this know what to BUILD or BET ON?
Output: 7-10 named ideas. Distinguish "possible" from "probable" — the best
futurist ideas feel inevitable in retrospect. Focus on the ones with the
strongest evidentiary base.
Do NOT evaluate feasibility — just generate.
Message teammates if a trajectory reframes or amplifies their ideas.
Spawning
Step 1: Spawn Signal Scout with model: "sonnet" and run_in_background: true.
Wait for it to complete.
Step 2: After Signal Scout returns, spawn all five ideation agents simultaneously.
Use model: "sonnet" and run_in_background: true for all.
Agent: {
name: "signal-scout",
model: "sonnet",
prompt: [full scout prompt],
run_in_background: true
}
Then after scout completes:
Agent: { name: "analogist", model: "sonnet", prompt: [...], run_in_background: true }
Agent: { name: "inverter", model: "sonnet", prompt: [...], run_in_background: true }
Agent: { name: "combinator", model: "sonnet", prompt: [...], run_in_background: true }
Agent: { name: "contrarian", model: "sonnet", prompt: [...], run_in_background: true }
Agent: { name: "futurist", model: "sonnet", prompt: [...], run_in_background: true }
Phase 3: Cross-Pollinate & Emerge
After all five ideation agents report back, the lead does the most important work: finding unexpected connections ACROSS agents.
This is the "emerge" phase — building on raw ideas to find combinations.
- Collect all ideas from all five agents (should be 30-45 raw ideas)
- De-duplicate — remove ideas that are essentially the same stated differently
- Cross-pollinate — the key creative act:
- Does the Analogist's pattern + the Contrarian's insight = something new?
- Does the Futurist's trajectory + the Combinator's mashup = something neither saw?
- Does the Inverter's broken assumption + the Analogist's precedent = a breakthrough?
- Generate 3-5 NEW "hybrid" ideas from cross-agent combinations
- Cluster into 5-8 themes (e.g., "infrastructure plays", "consumer behavior shifts", "regulation-driven opportunities", "timing arbitrage")
Phase 4: Rank & Present (Convergence)
NOW and only now, evaluate. Rank all ideas (de-duplicated originals + hybrids) on two axes:
- Novelty (0-10): How surprising is this? Would a smart person in this domain have already thought of it? 0 = obvious, 10 = genuinely new framing
- Plausibility (0-10): Given the signals, how grounded is this? 0 = pure fantasy, 10 = practically inevitable
Plot them mentally on a 2x2:
- High novelty + High plausibility = THE GEMS (these go first)
- High novelty + Low plausibility = WILD CARDS (intriguing but unproven)
- Low novelty + High plausibility = SAFE BETS (real but obvious)
- Low novelty + Low plausibility = DROP (don't include)
Output Document
Write to thoughts/brainstorm/YYYY-MM-DD-<domain-slug>.md:
---
date: <ISO 8601>
analyst: Claude Code (/brainstorm)
domain: "<domain>"
question: "<exploration question>"
idea_count: <total unique ideas>
gem_count: <high novelty + high plausibility>
---
# Brainstorm: [Domain/Question]
> [One sentence capturing the exploration]
## Signal Landscape
### Key Signals
[5-7 most important signals from the Scout, with sources]
### The Tension Map
[Where are the interesting tensions, unmet needs, and emerging shifts?]
---
## The Gems (High Novelty + High Plausibility)
These are the ideas that are both surprising AND grounded in real signals.
### 1. [Idea Name]
**Novelty:** [N]/10 · **Plausibility:** [N]/10
**Source:** [which agent(s) generated or contributed to this]
[2-3 paragraph description. Specific enough that someone could start working
on this tomorrow. Include: what it is, who it's for, why now, what makes it
non-obvious.]
**Key signal:** [the real-world evidence that supports this]
**The insight:** [the non-obvious connection that makes this interesting]
---
### 2. [Idea Name]
[same structure]
---
[Continue for all gems — typically 5-8]
---
## Wild Cards (High Novelty + Lower Plausibility)
Intriguing but unproven. Worth watching or exploring cheaply.
### [Idea Name]
**Novelty:** [N]/10 · **Plausibility:** [N]/10
[1 paragraph. What would need to be true for this to work?]
[Continue for 3-5 wild cards]
---
## Safe Bets (High Plausibility + Lower Novelty)
Real opportunities but not surprising. Include for completeness.
### [Idea Name]
**Novelty:** [N]/10 · **Plausibility:** [N]/10
[1 paragraph]
[Continue for 3-5 safe bets]
---
## Cross-Pollination Insights
These ideas emerged from COMBINING fragments across different agents.
### [Hybrid Idea Name]
**Combined from:** [Agent A]'s [idea fragment] + [Agent B]'s [idea fragment]
[Description of the hybrid and why the combination is more interesting than either part]
[Continue for 3-5 hybrids]
---
## Themes & Patterns
Looking across all ideas, these themes emerged:
1. **[Theme]** — [what it means, which ideas cluster here]
2. **[Theme]** — [what it means, which ideas cluster here]
3. **[Theme]** — [what it means, which ideas cluster here]
[Continue for 5-8 themes]
---
## What to Do Next
**To go deeper on a specific idea:** Run `/think [idea name]` for a full
multi-framework analysis, or `/munger [idea name]` for a lattice evaluation.
**To stress-test the best gems:** Run `/red-team` on this brainstorm output
to find the holes.
**To explore a different angle:** Run `/brainstorm` again with a narrower
constraint or a different question.
**To validate signals:** Pick the top 3 ideas and run cheap tests —
talk to 5 potential customers, build a landing page, search for existing
attempts and why they failed.
---
*Generated by /brainstorm · 6 agents · [N] raw ideas → [N] unique after de-dup*
*Signal sources: [count] web searches · Agents: Scout, Analogist, Inverter, Combinator, Contrarian, Futurist*
Phase 5: Present & Follow-up
Present the highlights inline — don't dump the whole document:
## Brainstorm: [Domain] — [N] ideas generated
**Signal landscape:** [1-2 sentence summary of what the Scout found]
---
### The Gems (top [N])
1. **[Idea Name]** — [one sentence] (Novelty: [N]/10 · Plausibility: [N]/10)
2. **[Idea Name]** — [one sentence] (Novelty: [N]/10 · Plausibility: [N]/10)
3. **[Idea Name]** — [one sentence] (Novelty: [N]/10 · Plausibility: [N]/10)
[continue for all gems]
### Best Wild Cards
1. **[Idea Name]** — [one sentence]
2. **[Idea Name]** — [one sentence]
### Strongest Cross-Pollination
**[Hybrid Name]** — [Analogist]'s [X] + [Contrarian]'s [Y] = [Z]
---
**Key themes:** [3-5 theme names]
Full output: `thoughts/brainstorm/YYYY-MM-DD-<slug>.md`
Want to:
1. Deep-dive any idea with `/think [idea]`?
2. Stress-test the gems with `/red-team`?
3. Narrow the domain and brainstorm again?
4. Explore a specific theme further?
Batch Mode
If the user wants to brainstorm across multiple domains or compare angles:
- Run full brainstorm on each domain (can parallelize — one team per domain)
- At the end, produce a cross-domain comparison:
## Cross-Domain Brainstorm
| Domain | Gems | Wild Cards | Strongest Idea | Key Theme |
|--------|------|------------|----------------|-----------|
| [A] | [N] | [N] | [name] | [theme] |
| [B] | [N] | [N] | [name] | [theme] |
**Cross-domain insight:** [what patterns appear across both domains that
suggest a higher-level opportunity]
Quality Standards
- No evaluation during generation. If an agent catches itself saying "this probably won't work" — that's convergent thinking leaking into the divergent phase. Strip it. Ideas should be wild during generation, filtered during ranking.
- Specificity over abstraction. "An AI tool for healthcare" is not an idea. "A diagnostic copilot that reads pathology slides and highlights the 3 regions most likely to be malignant, sold to mid-size pathology labs that can't hire enough specialists" is an idea. Every idea must pass the specificity check.
- Signals must have sources. The Scout's findings need URLs. Ideas grounded in real signals are categorically better than pure speculation.
- Cross-pollination is mandatory. If the lead just lists each agent's ideas without finding hybrid combinations, the synthesis failed. The hybrids section should contain at least 3 genuine combinations.
- The Gems must be non-obvious. If a smart person in this domain would have already thought of every gem, the brainstorm added no value. At least 2-3 gems should produce a "huh, I hadn't thought of that" reaction.
- Don't over-rank. Not everything needs to be a gem. It's fine for most ideas to be safe bets or wild cards. The ranking is honest, not inflationary.
Important Notes
- Cost: This skill spawns 6 agents (1 sequential, then 5 parallel). It's worth it for serious exploration. For quick ideation, just have a conversation.
- Sonnet for all agents, Opus for synthesis: The lead handles cross-pollination, ranking, and the final presentation — that's where judgment matters most.
- Pair with other skills: Run /brainstorm to generate → /think to analyze the best idea → /red-team to stress-test → /munger or /thiel for deep evaluation. Brainstorm is the START of the pipeline, not the end.
- Repeat to go deeper: Run /brainstorm again with a constraint from the first run ("brainstorm infrastructure plays in this space" after a broad first pass). Each iteration should go deeper, not wider.
- The Scout is the bottleneck: Signal quality determines idea quality. If the Scout's research is shallow, everything downstream suffers. The Scout should use 5-10 web searches minimum.
Related skills