| name | consult-polymaths |
| description | Use when the user wants multiple polymathic agents to analyze a problem, review code, debate an idea, or provide diverse expert perspectives. Triggers on 'consult polymaths', 'ask the polymaths', 'get polymath opinions', 'polymath review', 'run it by the polymaths', 'what would the polymaths say', or any request for multi-perspective expert analysis. Also trigger when the user says 'consult-polymaths' or wants to 'task the polymaths'. This is different from /consult (which queries external LLM APIs) — this skill spawns internal polymathic thinking agents. |
Consult Polymaths
Spawn multiple polymathic thinking agents in parallel to analyze a task from different expert perspectives. Each polymath applies a historical figure's cognitive architecture as structural constraints on reasoning — they're read-only consultants that analyze and advise.
Available Polymaths
Code & Architecture
| Agent | Lens |
|---|
| carmack | Performance, systems architecture, technical feasibility — finds the real constraint |
| linus | Code review, taste, BS detection, maintainability — demands elegance |
| tesla | Systems architecture, infrastructure, API design — complete mental model first |
| shannon | API design, simplification, hidden structure — strips to the invariant skeleton |
Product & UX
| Agent | Lens |
|---|
| jobs | Product vision, UX simplification, feature pruning — taste-first |
| rams | Product design, UI simplification — less but better |
| vangogh | UI/UX, color systems, emotional design — feeling before function |
| disney | Experience design, creative strategy — Dreamer/Realist/Critic triad |
Strategy & Decisions
| Agent | Lens |
|---|
| bezos | Customer-obsessed design, decision speed — press release first |
| thiel | Contrarian analysis, monopoly strategy — zero-to-one thinking |
| gates | Platform strategy, ecosystem design — decompose into atoms |
| andreessen | Market timing, technology adoption — spot discontinuities |
| suntzu | Competitive strategy, positioning — win before fighting |
| musk | Moonshot feasibility, aggressive simplification — delete before optimize |
Thinking & Analysis
| Agent | Lens |
|---|
| feynman | Debugging, first principles, cargo cult detection — freshman test |
| tao | Problem decomposition, cross-domain connections — toy models first |
| munger | Decision frameworks, bias detection — invert every problem |
| socrates | Assumption testing, exposing hidden ignorance — elenctic questioning |
| lovelace | Technology visioning, system abstraction — operational patterns |
| davinci | Cross-disciplinary synthesis, bio-inspired design — find the analog |
| aurelius | Decision-making under pressure, resilience — obstacle is the way |
Marketing & Content
| Agent | Lens |
|---|
| ogilvy | Copywriting, ad strategy — headline is 80% of the work |
| godin | Marketing strategy, audience building — smallest viable audience |
| graham | Startup strategy, product-market fit — do things that don't scale |
| mrbeast | Content strategy, attention engineering — retention curves |
Workflow
Step 1: Gather the Task
If the user provided a clear task in their message, use it directly. Otherwise, use AskUserQuestion to ask:
- "What problem or task do you want the polymaths to analyze?"
Step 2: Select Polymaths
Based on the task domain, select the most relevant polymaths. Consider:
- Code problems → carmack, linus, feynman, shannon, tesla
- Product decisions → jobs, rams, bezos, feynman, thiel
- Architecture → carmack, tesla, shannon, linus
- Strategy/business → bezos, thiel, gates, munger, suntzu
- UX/design → jobs, rams, vangogh, disney
- Debugging → feynman, carmack, linus, socrates
- Marketing → ogilvy, godin, graham, mrbeast
- Decision-making → munger, aurelius, bezos, socrates, feynman
- Innovation → davinci, lovelace, thiel, musk, tao
Cross-domain tasks benefit from mixing categories (e.g., a product architecture question might get carmack + jobs + shannon).
Step 3: Confirm with User
Use AskUserQuestion to present:
-
How many polymaths? — Offer 2, 3, 4, or 5 (suggest 3 as default for good coverage without overwhelming output). More than 5 gets noisy; fewer than 2 defeats the purpose.
-
Does this selection look good? — Show the selected polymaths with a one-line rationale for each. Let the user swap any out.
If the user wants to swap agents, adjust the selection and proceed. Don't re-confirm unless they explicitly ask.
Step 4: Spawn in Parallel
Spawn all selected polymaths simultaneously using the Agent tool with:
subagent_type matching the agent name (e.g., polymathic-feynman)- The same task prompt for each agent
- Each agent gets the full task context
The prompt to each agent should be the user's task, verbatim. Don't paraphrase or filter — let each polymath's cognitive architecture shape their unique response.
Step 5: Synthesize
After all agents return, provide a brief synthesis:
- Consensus — Where do the polymaths agree?
- Tensions — Where do they disagree, and why?
- Surprise insights — Anything unexpected that only one polymath surfaced?
- Recommended action — Based on the combined perspectives, what should the user do?
Keep the synthesis concise. The individual polymath responses carry the depth; the synthesis highlights patterns and conflicts.
