Refine Raw Ideas into Actionable Concepts

Idea Refine turns a raw idea into a focused, actionable concept through an interactive conversation. Its three phases expand possibilities, evaluate distinct directions, and produce a concrete Markdown one-pager. The approach starts with user experience and works backward to technology, favoring the simplest version that solves the real problem.

The first phase restates the idea as a concise How Might We problem. Three to five sharpening questions establish the target user, success criteria, constraints, previous attempts, and timing. The process waits for a clear understanding of the audience and success before moving forward.

It then develops five to eight considered variations. Lenses include inversion, removing constraints, shifting audiences, combining adjacent ideas, simplifying the concept, exploring scale, and applying an expert perspective. Each variation needs a reason to exist. In a codebase, architecture, patterns, constraints, and prior work ground the alternatives.

Use it when an idea remains vague, when assumptions should be stress-tested before committing, or when more options should be explored before choosing a direction. Trigger phrases include requests to refine an idea, ideate on a concept, or stress-test a plan.

Do not jump directly to the final artifact, generate large piles of shallow alternatives, or agree with weak concepts without examining their value. Avoid adding process steps merely for ceremony. A plan without surfaced assumptions or explicit exclusions misses central parts of the workflow.

The initial input is an idea plus answers about users, goals, constraints, and context. User reactions determine which alternatives advance. The convergence phase clusters promising ideas into two or three meaningfully different directions, then evaluates user value, feasibility, and differentiation.

For each direction, it identifies unvalidated bets, potential failure points, and things deliberately ignored. The evaluation asks who benefits, how much technical and resource work is required, which part is hardest, and whether users would switch from an existing approach. Feedback should challenge complexity and weak value with specific reasons.

The output contains a problem statement, recommended direction, assumptions with validation strategies, MVP scope, a Not Doing list with reasons, and open questions. The user confirms the direction before implementation. Saving the one-pager requires confirmation and normally uses docs/ideas/[idea-name].md, or another chosen destination.

An optional initialization command prepares the ideas directory:

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

Supporting resources frameworks.md, refinement-criteria.md, and examples.md provide additional ideation frameworks, evaluation criteria, and worked examples. These are selected to suit the idea rather than applied mechanically. Repository inspection supports concepts developed within an existing project. Commands, dependencies, and external service behavior require verification before applying changes. The skill is MIT-licensed.