ideation-copilot

Ideation Copilot

Turn a raw idea into a validated, investor-ready concept — using AI agents that research, score, and stress-test your thinking.

Installation

/plugin marketplace add kaminskypavel/ideation-copilot
/plugin install ideation-copilot@ideation-copilot

How It Works

You start with an idea. The copilot guides you through a loop of testing, scoring, and improving until the idea is solid — or killed. Each command tells you what to do next.

Command	Purpose
/idea:new [name "description"]	Scaffold a new idea with 6 structured docs
/idea:evaluate [idea-name] [vc|market|yc]	Score with parallel agents (all by default, or pick one)
/idea:pushback [idea-name]	Conversational stress-test with web research
/idea:update [idea-name]	Add new info to your docs (interviews, data, team changes)
/idea:forge [idea-name]	Synthesize everything into a consolidated summary
/idea:postmortem [idea-name]	Structured debrief when you kill an idea
/idea:setup	Check & configure optional integrations (Exa, etc.)

Step 1: Pitch your idea

/idea:new pawguard "Smart collar that detects early signs of illness in dogs using biometrics"

Scaffolds 6 structured docs (overview, brainstorm, lean canvas, assumptions, PMF strategy, experiments). You fill in what you know.

Step 2: Get scored

/idea:evaluate pawguard

Three AI agents run in parallel, each researching and scoring your idea from a different angle:

Agent	What it asks	Dimensions
VC	"Is this investable?"	Team, Timing, TAM, Technology, Moat, Business Model, GTM, Traction
Market Analyst	"Is the market real?"	Market Size, Competitive Landscape, Timing & Tailwinds, Customer Access, Regulatory Risk
YC Founder-Fit	"Should YOU start this?"	Problem Acuteness, Personal Demand, Successful Proxies, Commitment, Scalability, Idea Space Fertility

You get a combined score (0-100), deal-breakers flagged, and the single weakest dimension to fix first. Agents use web research — they'll check your TAM claims, find your competitors, and validate your timing.

Run a single agent: /idea:evaluate pawguard vc or market or yc

Step 3: Stress-test through dialogue

/idea:pushback pawguard

An adversarial sparring partner breaks your idea into testable claims and challenges each one. You defend, clarify, or concede. It uses named reasoning tools (inversion, base rate analysis, pre-mortem) and web research to back up its challenges.

This isn't a lecture — it's a conversation. You'll discover blind spots you didn't know you had.

Step 4: Fix what's weak

/idea:update pawguard

Low scores often mean your docs are incomplete, not that the idea is bad. Add what's missing — your team background, customer interview results, experiment outcomes, market data.

Step 5: Repeat, then synthesize

Run evaluate and pushback again. Your scores should improve. Keep iterating until you're confident — or until the evidence tells you to pivot.

When you're ready, forge synthesizes everything — score trajectory over time, what's validated vs still assumed, key pivots, and a pitch-ready summary:

/idea:forge pawguard

What's In an Idea Folder

Each idea lives in ideas/YYYY-MM-DD-idea-name/:

You write these:

File	What goes in it
00-overview.md	Problem, insight, solution, target customer
01-brainstorm.md	Problem/solution space exploration
02-lean-canvas.md	Lean Canvas — UVP, channels, revenue, costs
03-assumptions.md	Riskiest assumptions ranked, with evidence tracking
04-pmf-strategy.md	PMF ladder, go-to-market, milestones
05-experiments.md	Experiment backlog, results, pivot/persevere decisions

The copilot creates these: