idea-refine

Overview of idea-refine skill

idea-refine is a structured ideation skill for turning rough concepts into clearer, buildable directions through divergent thinking, critique, and convergence. It is best for founders, product leads, engineers, and AI agents doing Requirements Planning who need to decide what to build before they write specs or commit to implementation. The value of the idea-refine skill is not generic brainstorming; it is forcing ambiguity into a concrete one-pager with assumptions, scope, and a clear “not doing” list.

What idea-refine is good for

Use idea-refine when you have a vague feature, product idea, or workflow change and need to pressure-test whether it is worth pursuing. It helps when the real problem is still fuzzy, the target user is unclear, or the solution space is too broad. The skill is especially useful for early product discovery, feature framing, and idea-refine for Requirements Planning because it pushes from “interesting concept” to “specific direction.”

Why this skill is different

The idea-refine guide is built around a 3-step flow: understand and expand, evaluate and converge, then sharpen and ship. That matters because many ideation prompts jump straight to solutions. This skill first restates the idea, asks sharpening questions, and generates variants before selecting a direction. That reduces false confidence and makes the resulting plan more decision-ready.

When it is a strong fit

Choose idea-refine if you need a lightweight but disciplined way to explore options, identify assumptions, and leave with a markdown artifact you can share. It is a good fit when you want the assistant to behave like an ideation partner, not a feature factory. If you already know the exact requirements, a normal prompt may be enough; if you do not, idea-refine is usually the better install.

How to Use idea-refine skill

Install idea-refine

Install from the agent-skills repo with:

npx skills add addyosmani/agent-skills --skill idea-refine

If you want the optional local setup used by the repo, run the helper script:

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

That script initializes docs/ideas/ and is useful when you want outputs stored in a predictable place. This is the practical idea-refine install path for users who want the full workflow, not just the prompting behavior.

Give it the right starting input

The skill works best when your first message includes a rough idea plus context that narrows the problem. Good input tells it what the idea is, who it is for, and what constraint matters most. For example: “Refine a workflow tool for small agencies that reduces client approval delays without adding another dashboard.” That is much stronger than “help me ideate.”

For idea-refine usage, include:

• the target user or customer
• the problem or opportunity
• current workaround or competitor
• constraints like budget, time, platform, or scope
• the decision you need by the end

Use the repo files in the right order

Start with SKILL.md to understand the workflow, then inspect examples.md, frameworks.md, and refinement-criteria.md for how the skill thinks about ideation, comparison, and evaluation. Read scripts/idea-refine.sh if you want the directory setup behavior. That file order is the fastest way to understand the idea-refine guide without reading the entire repo first.

Turn a rough prompt into a better session

Instead of asking for “ideas,” ask for a refinement pass with a clear output target. A strong prompt might be: “Use idea-refine to evaluate three directions for a B2B onboarding assistant, then recommend one MVP with assumptions and a not-doing list.” That gives the skill a decision to make, which improves the quality of the final one-pager.

idea-refine skill FAQ

Is idea-refine only for early-stage startup ideas?

No. The idea-refine skill also works for feature planning, process redesign, internal tooling, and any requirement that is still too broad to scope cleanly. It is most valuable when the team needs to narrow options before writing implementation details.

How is this different from a normal brainstorming prompt?

A normal prompt often returns a list of ideas. idea-refine is designed to move through expansion, stress testing, and convergence, so the output is more actionable. For idea-refine for Requirements Planning, that means fewer loose ideas and more decision-ready structure.

Do beginners need to know ideation frameworks first?

No. The skill is usable without prior framework knowledge. If you do know frameworks like HMW or SCAMPER, they can help you ask better follow-up questions, but the core idea-refine usage does not depend on them.

When should I not use idea-refine?

Do not use it when the request is already well specified, when you need a final implementation plan immediately, or when the main task is writing code rather than choosing direction. In those cases, a narrower prompt or a planning skill is a better fit.

How to Improve idea-refine skill

Give it sharper constraints, not just more words

The biggest quality jump comes from adding boundaries: audience, business goal, platform, timeline, and what is explicitly out of scope. The idea-refine skill is strongest when it can trade off among real constraints instead of producing broadly appealing but weakly differentiated options.

Ask for a decision, not a summary

If you want better results, tell the assistant what judgment it must make: pick one direction, compare two approaches, identify the riskiest assumption, or define MVP scope. Without that, the skill can stay in exploration mode too long. Good idea-refine usage ends with a recommendation, not just possibilities.

Inspect and reuse the output structure

The repo’s output shape is a useful cue: problem statement, recommended direction, key assumptions, MVP scope, and not-doing list. If the first pass is too vague, ask it to tighten one section at a time rather than regenerate everything. That usually improves clarity faster than restarting the whole session.

Watch for common failure modes

The main risks are over-broad ideation, hidden assumption drift, and solutions that sound clever but do not map to a real user problem. Push back when the output does not name a specific user, does not distinguish painkiller from vitamin value, or does not explain why one direction wins. That is the fastest way to make idea-refine more useful for Requirements Planning.