Refine Raw Ideas into Actionable Concepts

An ideation-partner skill that refines raw ideas into sharp, actionable concepts through structured divergent and convergent thinking, guiding the user through a three-phase conversation (not a rigid template). Philosophy: push toward the simplest version that solves the real problem, start with user experience and work backwards to technology, say no to a thousand things (focus beats breadth), and challenge "how it's usually done" as not a reason. Phase 1 (Understand & Expand, divergent) restates the idea as a crisp "How Might We" problem statement, asks 3-5 sharpening questions via AskUserQuestion (who this is for, what success looks like, real constraints, what's been tried, why now) and does not proceed until audience and success criteria are clear, then generates 5-8 idea variations using named lenses - inversion ("what if we did the opposite"), constraint removal, audience shift, combination with an adjacent idea, simplification to a 10x-simpler version, scaling to a 10x version, and an expert lens on what's obvious to insiders but not outsiders - grounding variations in actual codebase context (via Glob/Grep/Read) when run inside a project, and drawing selectively from an additional frameworks.md reference rather than running every framework mechanically. Phase 2 (Evaluate & Converge) clusters the ideas that resonated into 2-3 meaningfully distinct directions, stress-tests each against user value (painkiller vs. vitamin), feasibility (technical/resource cost, hardest part), and differentiation (would someone switch from their current solution) per a fuller rubric in refinement-criteria.md, then explicitly surfaces hidden assumptions for each direction - what's being bet on but unvalidated, what could kill the idea, and what's being knowingly ignored - since skipping this step is where most ideation fails; the skill is instructed to be honest rather than supportive, pushing back on weak ideas with kindness rather than acting as a yes-machine. Phase 3 (Sharpen & Ship) produces a markdown one-pager with a Problem Statement, Recommended Direction (2-3 paragraphs max), a checklist of Key Assumptions to Validate with how to test each, an MVP Scope, a Not Doing list with reasons (called out as arguably the most valuable section, since focus means saying no to good ideas too), and Open Questions - saved to docs/ideas/[idea-name].md only after explicit user confirmation. Documented anti-patterns: don't generate 20+ ideas (5-8 well-considered beats 20 shallow ones), don't be a yes-machine, don't skip "who is this for," and don't produce a plan without surfacing assumptions.

Use it when an idea is still vague, before stress-testing assumptions ahead of committing to a plan, or when options need expanding before converging on one - triggered by phrases like "help me refine this idea," "ideate on [concept]," or "stress-test my plan."

Input is a raw idea plus conversational answers to sharpening questions. Output is a markdown one-pager (problem statement, recommended direction, assumptions to validate, MVP scope, not-doing list, open questions) optionally saved to docs/ideas/[idea-name].md after user confirmation.

bash skills/idea-refine/scripts/idea-refine.sh

Uses the AskUserQuestion tool for sharpening questions, Glob/Grep/Read to ground variations in existing codebase context when applicable, and references frameworks.md and refinement-criteria.md for additional ideation lenses and evaluation criteria.

Anyone with a vague idea who wants a structured, honest ideation partner - one that pushes back on weak ideas, surfaces untested assumptions before they become plan-killers, and forces the discipline of an explicit Not Doing list rather than open-ended brainstorming without convergence.