What does this skill do, and when should you use it?
Idea Refine is for concepts that are still vague, need broader exploration, or require a plan stress test. It first expands the idea into multiple options, then converges through clustering, value assessment, feasibility analysis, and assumption testing. The workflow requires clarity about the target user and success criteria before recommending a direction. After user confirmation, it produces a Markdown one-pager at docs/ideas/[idea-name].md.
It restates a raw idea as a “How Might We” problem, asks 3–5 questions about the specific user, success criteria, constraints, prior attempts, and timing, and generates 5–8 variations using different ideation lenses. It then clusters promising options into 2–3 directions and stress-tests them for user value, feasibility, and differentiation. For each direction, it identifies unvalidated bets, possible deal-breakers, and deliberate exclusions. When used inside a codebase, it reads relevant repository context and is instructed to read frameworks.md, refinement-criteria.md, and examples.md. Its final artifact covers the problem statement, recommended direction, assumptions, MVP scope, items not being built, and open questions.
- A product or engineering lead has a rough concept and needs to clarify the real problem before committing development effort.
- A founder has several possible directions and wants to compare user value, implementation cost, and differentiation.
- A team needs to define a technical initiative but must first establish its target user, success criteria, and critical assumptions.
- A feature is being planned inside an existing codebase and should reflect current architecture, constraints, and prior art.
- A plan needs an adversarial review that makes risks and intentionally excluded scope explicit.
How do you install this skill?
- AskUserQuestion, Glob, Grep, and Read may not exist or behave consistently across agents, with no explicit fallback procedure.
- User confirmation is required before saving, but overwrite behavior, write failures, and recovery are unspecified.
- The verification checklist is procedural self-checking rather than automated coverage of key paths; publisher identity remains unknown.
- Shell / CLI
- Local filesystem
Bash
This skill is part of the addyosmani/agent-skills monorepo. Install the full collection with npx skills add addyosmani/agent-skills; the README also documents the single-skill format, which can be applied to Idea Refine as npx skills add addyosmani/agent-skills --skill idea-refine. The collection is MIT-licensed and targets multiple AI coding agents. The README does not provide a generic standalone folder-copy procedure for this skill.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/addyosmani/agent-skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/idea-refine" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Help me refine this idea,
- Ideate on [concept],
- Stress-test my plan.
In a compatible agent, provide an idea and use a trigger such as “Help me refine this idea,” “Ideate on [concept],” or “Stress-test my plan.” The skill proceeds as an interactive dialogue through Understand & Expand, Evaluate & Converge, and Sharpen & Ship. After the direction is confirmed, it saves the one-pager to docs/ideas/[idea-name].md, or to another location chosen by the user. Optionally initialize the ideas directory with bash skills/idea-refine/scripts/idea-refine.sh.
What are this skill's strengths and limitations?
- Provides a concrete three-phase workflow from exploration to a shippable concept brief.
- Requires explicit attention to the target user and success criteria.
- Makes hidden assumptions, failure risks, and non-goals visible.
- Can incorporate existing codebase context when ideating within a project.
- Produces a reusable Markdown artifact rather than leaving the result only in conversation.
- The core workflow depends on interactive dialogue and is not fully documented for non-interactive execution.
- Users must answer questions and confirm the final direction before the artifact is produced.
- It references platform-specific question, argument-substitution, and file-reading mechanisms, so other agents may require adaptation.
- The supplied source does not include the contents of frameworks.md, refinement-criteria.md, or examples.md, and provides no platform test evidence.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Idea Refine this page | 51 · Use with care | ★ 107k | 13d ago | MIT |
| Intent Interviewer | 59 · Recommended | ★ 107k | 13d ago | MIT |
| Context Engineering | 49 · Use with care | ★ 107k | 13d ago | MIT |
| Compound Engineering Brainstorm | 61 · Recommended | ★ 26k | 13d ago | MIT |
| Playwright Browser Automation Skill | 68 · Recommended | ★ 3.2k | 2mo ago | MIT |
The README names Superpowers and Matt Pocock's skills as comparison points and links to docs/comparison.md. That comparison document is not included in the supplied source, so no further differences can be verified.
How did FollowSkills review this skill?
The skill requires interactive clarification and explicitly says to save only after user confirmation; its initialization script only creates docs/ideas and exposes a simple JSON status. No network access, credential handling, covert exfiltration, or destructive default is shown. Deductions apply because it does not define write-failure recovery, rollback, sensitive-idea handling, or tool permission boundaries, and publisher identity is unverified.
The three-phase workflow, confirmation gate, anti-patterns, and verification checklist are broadly consistent, and the initialization script uses set -e with explicit status output. Deductions apply because AskUserQuestion, Glob, Grep, and Read are assumed without availability checks or fallback paths; empty input, malformed input, missing references, and save failures have limited diagnostic handling. Static review caps this at10, with no execution evidence.
The frontmatter identifies scenarios, trigger phrases, and an interactive-dialogue boundary; the examples cover product, codebase, and process ideation. Deductions apply because non-fit cases, input formats, and trigger-conflict behavior across agent environments are not sufficiently specified. There is no Chinese-language or mainland-China adaptation guidance, although the core function does not depend on overseas online services.
The documentation has clear usage, workflow, output, examples, anti-patterns, red flags, and verification sections, supporting progressive disclosure; MIT licensing and repository-level installation guidance are present. Deductions apply because the skill lacks its own version, changelog, maintenance owner, and update path, while dependency and fallback guidance for the referenced supporting files is incomplete.
The skill defines a directly usable Markdown one-pager containing the problem statement, recommendation, assumptions, MVP, and non-goals, and requires confirmation before persistence. Deductions apply because filename collisions, permissions, save failures, and final artifact validation are unspecified, and there is no independent evidence of output quality or marginal benefit versus manual ideation. Static review caps this at7.
The skill supplies auditable phase goals, an output template, and completion checks, while the repository shows a CI validation entry point. Deductions apply because the supplied test fixture is unrelated to this skill's key paths; there are no dedicated tests for triggering, phase transitions, confirmation, or file saving, and no third-party execution results or corroborating evidence. Static review caps this at5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →