with one click
wide-open-brainstorm
// Multi-model brainstorming room for product strategy and experience design — serious plus whimsical, multi-altitude, multi-round ideation.
// Multi-model brainstorming room for product strategy and experience design — serious plus whimsical, multi-altitude, multi-round ideation.
[HINT] Download the complete skill directory including SKILL.md and all related files
| name | wide-open-brainstorm |
| description | Multi-model brainstorming room for product strategy and experience design — serious plus whimsical, multi-altitude, multi-round ideation. |
| display_name | Wide-Open Brainstorm |
| brand_color | #C87941 |
| local_only | false |
| group | Better Products |
| usage | /wide-open-brainstorm:run |
| summary | Multi-model brainstorming room for product strategy — serious plus whimsical, multi-altitude ideation |
| default_prompt | Run a wide-open brainstorming room for this product idea. Move between big-picture strategy, project-level organization, tactical actions, and whimsical metaphors; then distill the strongest differentiated concepts. |
Run an expansive product/experience brainstorm that is both strategic and playful. Use this skill when the user wants to go wide, pull ideas from multiple perspectives, blend practical/actionable concepts with whimsical metaphors, and discover a product's deeper thesis rather than merely produce a feature list.
Treat the session as a brainstorming room: part product strategy, part systems design, part improv, part command-center fantasy, part ruthless distillation.
Optimize for:
Avoid:
Restate the user's idea in a richer prompt that preserves their language and expands the intent. Include:
If the user provided enough direction, proceed without questions. If not, ask one short question about the desired domain or artifact.
Before fan-out, detect which external LLM CLIs are available. Use the pre-mortem skill's detection script if it is installed, otherwise run the inline fallback:
# Preferred — reuse pre-mortem's script if the skill is installed:
bash "${SKILL_DIR}/../pre-mortem/scripts/detect-llms.sh" 2>/dev/null || \
for tool in agent ask-gemini gemini llm codex; do
command -v "$tool" >/dev/null 2>&1 && echo "found: $tool"
done
Determine your execution mode from the result:
| Available | Mode | Strategy |
|---|---|---|
| 2+ external LLMs | Full diversity | Subagents for project-aware roles + external LLMs for outsider/whimsy roles |
| 1 external LLM | Hybrid | Subagents + one external LLM for a contrasting outsider perspective |
| None | Single-agent | All roles as subagents with strongly differentiated mandates and tones |
Use the 6 core roles below. Run every available agent type; don't leave models idle.
| Role | What they generate | Ideal agent type |
|---|---|---|
| Strategist | Product thesis, priorities, what existing tools miss, positioning | Code-aware subagent (reads local README, docs, recent git) |
| Operator | Concrete workflows, statuses, actions, edge cases, day-to-day mechanics | Code-aware subagent |
| Cartographer | Views, maps, grouping patterns, zoom levels, navigation metaphors | External LLM (fresh eyes — no codebase access) |
| Archivist | Memory, synthesis, knowledge capture, long-term coherence, recovery flows | External LLM |
| Trickster | Whimsical metaphors, silly-but-useful delight, strange UI lenses, absurdist reframes | External LLM (most creative divergence happens with no anchoring context) |
| Skeptic | What becomes noise, what users ignore, what shouldn't ship, local maxima | Any model — assign to agent --fast if available for blunt, fast output |
| Executioner | The strongest honest case for never building this at all — wrong timing, wrong audience, wrong founder, already solved, fundamental flaw in premise | Any model with a blunt mandate — always included, never optional |
| Role | What they generate | Ideal agent type |
|---|---|---|
| Future Self | 6/12/24-month horizon ideas; what smart teams converge on; leapfrog moves | External LLM (e.g., agent --frontier or ask-gemini) |
| Outsider / Cultural Stranger | Assumptions the team never questioned; non-default user perspectives | External LLM |
| Customer Whisperer | Emotional arc the user goes through; delight moments; trust signals | External LLM |
| Devil's Advocate | The honest thing nobody wants to say; core assumption challenges | Any model with a blunt mandate |
Default panel for full-diversity mode: all 6 core + Executioner + Future Self + Outsider.
Default panel for single-agent mode: Strategist, Operator, Trickster, Skeptic, Executioner, Future Self — as subagents with strongly differentiated tones.
The Executioner is always included. A brainstorm without a kill test is a yes-machine.
Dispatch all roles simultaneously in a single turn. Do not wait for one to finish before launching the next.
Every role gets the seed brief (from step 1) plus a role-specific lens:
=== WIDE-OPEN BRAINSTORM ===
PRODUCT / IDEA: [from seed]
AUDIENCE: [who it is for]
CORE OUTCOME: [what "success" means to the user]
EMOTIONAL FLAVOR: [desired brand/experience tone]
PERMISSION: Be practical, silly, weird, and ambitious. All altitudes welcome.
YOUR ROLE: [role name]
YOUR MANDATE: [role-specific mandate — see below]
YOUR TONE: [e.g., "wide-open strategist" / "whimsical inventor" / "dry skeptic"]
INSTRUCTIONS:
1. Think across at least three altitudes:
- 10,000 feet: goals, strategy, direction, what matters and why
- 1,000 feet: projects, workstreams, groupings, operational patterns
- Ground level: exact items, statuses, actions, next clicks
2. Generate ideas across your assigned lens. Don't self-censor.
3. For whimsical ideas: explain what the metaphor REVEALS, not just what it is.
4. For practical ideas: name the specific workflow it enables.
5. Include a time-horizon pass:
- 6 months: what would smart builders do once the rough edges are fixed?
- 12 months: what do teams converge on when the pattern is legible?
- 24 months: what does the mature version assume as table stakes?
- Leapfrog: what bypasses the expected roadmap and gets closer to THE FUTURE now?
6. Name your 3 strongest ideas. Good names make ideas sticky.
7. End with: "The thing most people miss about this: [one non-obvious insight]"
FORMAT: Flowing prose or numbered list, your choice. No hedging. No meta-commentary.
| Role | Mandate |
|---|---|
| Strategist | Identify the product's true north star, what the competitive alternatives get wrong, and the one insight that makes this genuinely different. Think like a founder writing a first-principles memo. |
| Operator | Map the real workflow: what does the user actually do, step by step, when this product is working? Find the five micro-decisions the product must make easier. |
| Cartographer | Design the views, maps, and grouping systems. How does the user see everything at once? How do they navigate without getting lost? What metaphor makes the product's structure legible? |
| Archivist | Design the memory and synthesis layer. How does the product get smarter over time? What does it remember, surface, and forget? How does the user recover when context is lost? |
| Trickster | Generate whimsical reframes and strange metaphors. What if this product were a ship's bridge? A garden? A detective's evidence board? A jazz session? Push until at least one idea feels genuinely surprising. |
| Skeptic | Be the voice of restraint. What will users ignore? What adds noise without signal? What is the most common mistake products like this make? What should NOT be built? |
| Future Self | Assume it is now [18 months from today]. The category has matured. What do all the good versions of this product have in common? What leapfrog move gets there before the obvious roadmap does? |
| Outsider | Approach this product as someone who doesn't share any of the builder's assumptions. What is confusing about the framing? What assumption is being treated as obvious that isn't? |
| Customer Whisperer | Map the user's emotional arc: what do they feel when they discover this, when it first works, when it fails, when it becomes a habit? Where does delight live? Where does trust get built or broken? |
| Devil's Advocate | Make the strongest honest case against the core premise. What if the fundamental assumption is wrong? What would make this irrelevant? |
| Executioner | Your job is not to find flaws — it is to argue that this idea should be abandoned entirely. Find the single strongest reason it should never be built: wrong timing, already exists and is better, fundamental audience mismatch, wrong founder for this problem, market that won't pay, regulatory trap, or premise that sounds good but collapses on first contact with reality. You are not here to balance — you are here to kill the idea if it deserves killing. If you cannot find a compelling reason to kill it, say so explicitly: "This idea survived the kill test. Here's why." That verdict is a green light. |
| Role | Preferred model |
|---|---|
| Strategist | Code-aware subagent (general-purpose agent with local file access) |
| Operator | Code-aware subagent |
| Cartographer | agent --smart (gemini-2.5-pro, strong spatial/systems reasoning) |
| Archivist | agent --smart or agent (default flash) |
| Trickster | agent --fast (llama-4-scout, fast and generative, less anchored) |
| Skeptic | agent --fast (blunt, fast output) |
| Future Self | agent --frontier (claude-opus-4-5, widest horizon) |
| Outsider | agent (default, fresh framing via gemini-2.5-flash) |
| Customer Whisperer | agent --smart or any external LLM |
| Devil's Advocate | agent --fast — give the fastest, bluntest model |
| Executioner | agent --fast preferred; subagent with explicit "argue for abandonment" mandate otherwise |
Ensure the combined output covers all three levels:
Look for interaction patterns that let the user smoothly zoom out, zoom in, and jump precisely. If any altitude is thin, ask the relevant subagent to extend their output there.
After collecting role outputs, surface ideas from the future-self and leapfrog lens explicitly:
For each horizon, distinguish:
Do not treat the far future as sci-fi garnish. Use it to identify which near-term choices keep optionality open, which ideas are local maxima, and which design thesis can leapfrog the expected market path.
After collecting first-round outputs, run a second round when:
Send each agent a digest of what the other roles found, then ask:
Separate ideas into:
Note where multiple roles independently converged on the same idea — that convergence is a signal.
Name the strongest concepts. Good names make ideas reusable and spreadable.
Treat the Executioner's output separately from the other roles.
If the Executioner could not find a compelling reason to kill the idea:
If the Executioner found a compelling reason to kill the idea:
Use this structure unless the user requested another format:
Keep the final answer crisp enough to read, but rich enough to preserve the spark.
When the user asks to save the brainstorm, create a dated Markdown file in the most relevant project docs folder or durable notes location. Include:
Prefer project-local docs/ for project-specific brainstorms. Prefer the user's durable notes location for cross-project life/work strategy.
Load references/wide-open-brainstorm-prompt.md when a full prompt template is useful.