What does this skill do, and when should you use it?
vibe-check is an open-source Claude skill from a product manager with 12+ years of zero-to-one experience. It grills a fuzzy idea into a real problem, reality-checks it against Reddit threads and competitor reviews, then produces a full build plan and interactive PRD. Along the way it teaches build-time basics like Git and deployment, and its Checkup Mode tidies messy AI-generated codebases. Built for Claude Code, Codex, and Antigravity.
- Interviews the user one question at a time to surface the real problem, locking in a three-line north star: goal, current workaround, and why the workaround sucks
- Runs discovery: maps job steps via Jobs-to-be-Done, mines Reddit and review sites (G2, Capterra, app stores) through a five-lens sort with permalink verification
- Scores needs with ODI opportunity scoring (Opportunity = Pain + max(0, Pain − Served)) to separate the differentiator from table stakes
- Runs Crazy 8 sketching, an Experience Blueprint, and a Story Map to split features into V1 vs V2+
- Produces a markdown plan for the AI coding tool plus a self-contained interactive HTML PRD for the human
- Offers Checkup Mode (references/CODE-CHECKUP.md) to assess and safely tidy a messy existing codebase
- A first-timer with an app idea who wants to know what to build and whether it's worth building before touching code
- A vibe coder who can get something running on screen but needs product thinking to plan the full picture
- Someone who already did user research and wants to skip straight to planning the MVP with their evidence in hand
- A non-coder whose AI-built app has grown messy and scary to touch and needs a gentle code checkup
- A founder exploring a two-sided marketplace who needs both sides discovered plus a cold-start plan
- Experienced developers debugging, refactoring, or setting up CI/CD — the skill explicitly excludes this
- Users who want the AI to just start coding; the mandatory discovery pass can end in a documented no-go verdict
How do you install this skill?
- Discovery depends on Reddit and overseas review sites, largely unreachable from mainland-China networks; expect degraded results and reliance on the fallback ladder or manual user search.
- Runtime version checks and discovery involve outbound requests to third-party sites; mirror instances are unverified — assess in sensitive environments.
- This is a static review with no execution; discovery and output quality depend on model execution — trial on a small task first.
- Publisher is an unverified solo maintainer with no third-party maintenance guarantee.
- Shell / CLI
- Network access
- Local filesystem
Node.js (for the skills CLI installer)
The easiest route is the open skills CLI, which works across agents:
Claude Code / Codex / Antigravity
npx skills add TexasBedouin/vibe-checkOr clone it straight into your project:
Claude Code
git clone https://github.com/TexasBedouin/vibe-check .claude/skills/vibe-checkOther AI tools: copy the contents of SKILL.md into your system prompt or project instructions.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Use the vibe-check skill to help me plan my app.
- Is my idea worth building? Reality-check it with vibe-check.
- I already validated my idea. Use vibe-check to plan the build.
- My AI keeps breaking things and the project is a mess — run vibe-check's Checkup Mode on it.
After installing in Claude Code, trigger it in plain language, e.g. "Use the vibe-check skill to help me plan my app." It first asks a light question to gauge your experience (the confidence dial sets pace, jargon, and hand-holding), then routes you to one of three on-ramps: the full journey (default), validate-only, or plan-only. The flow runs through Phase 0 discovery (grilling plus Reddit/review reality-check and ODI scoring), Phase 1 vision, Phase 2 Crazy 8 experience sketching, and onward to the final plan and interactive PRD. Harmful ideas are named and declined, and weak evidence leads to an honest no-go with narrow/pivot/stop options. It silently checks for a newer version at session start and never blocks on it.
What are this skill's strengths and limitations?
- Covers the full pre-build journey from validation to buildable blueprint, filling the gap left by coding skills that assume you already know what to build
- Grounded method: beginner translations of JTBD, ODI opportunity scoring, Design Sprint, and User Story Mapping
- Built-in anti-hallucination protocol: quotes must pass permalink verification, and roughly three independent sources are required for a 'seen it' evidence tag
- Three on-ramps and three honest endings (full plan, findings summary, or no-go), backed by real example sessions including ClearList, a live product built with the skill
- MIT licensed, free, and updatable via the skills CLI
- Research relies on Reddit and review sites, which the author himself calls directional rather than statistical — not a substitute for real user research
- The skill is long and rule-heavy; sessions take real time and may feel heavy if you just want a quick concept
- Visual output depends on a bundled diagram engine or mermaid, with degraded rendering in tools lacking a browser or file system
- The plan-only on-ramp is only as good as the validation material you bring; thin material forces a partial discovery re-run
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 |
|---|---|---|---|---|
| vibe-check Product Planning Coach this page | 62 · Recommended | ★ 619 | 3mo ago | MIT |
| Product Architect | 54 · Use with care | ★ 115 | 1mo ago | MIT |
| Factory Floor — Startup Coach | 53 · Use with care | ★ 14 | 2mo ago | — |
| MVP Minimalist Advisor | 52 · Use with care | ★ 11k | 6mo ago | — |
| Validate Idea | 52 · Use with care | ★ 11k | 6mo ago | — |
The author explicitly positions it against Matt Pocock's grill-me: grill-me is for engineers and interrogates the code; vibe-check is for everyone else, aiming the same relentless questioning at the problem and the person, and covering the whole journey from validation through build guidance to code checkups.
How did FollowSkills review this skill?
Pure instruction-based skill: no malware, no credential collection; version check is non-blocking best-effort with silent skip on failure; explicitly forbids fabricated quotes, intercepts harmful ideas; MIT license with clear attribution. Deductions: runtime dependence on external fetching (Reddit, review sites, raw.githubusercontent.com), data flowing through unverified mirror instances, no explicit privacy/data-flow disclosure.
Highly self-consistent instructions: three-rung fetch ladder, fail-closed source verification, mermaid/table fallbacks when the diagram engine cannot render, and documented behavior on abnormal inputs. Deductions: static review cannot execute; no test suite or CI evidence covering the skill's key paths is present in the provided files, so the reliability ceiling of 10 applies.
Trigger conditions, three on-ramps, and non-fit boundaries (not for experienced dev debugging/CI) are exceptionally clear in the frontmatter, supporting accurate semantic invocation. Deductions: no Chinese-language support declared; core discovery depends on Reddit, G2, GitHub — services unreachable from mainland China — with only partial mitigations (Serper, hand-off to user).
Semantic versioning, a detailed CHANGELOG including self-reported defects, bump.sh with thorough pre-flight checks, RELEASING.md maintenance path, well-layered progressive disclosure across reference files, explicit license. Deductions: unverified solo maintainer with no third-party governance; self-documented fragility of the hardcoded master-branch version URL.
Claimed deliverables (structured plan plus self-contained interactive PRD) are backed by referenced examples and a clear value proposition over manual planning. Deductions: static review cannot verify representative output quality; SKILL.md is very long, implying substantial context cost; example files are not included in the provided evidence.
The CHANGELOG is an auditable evolution record including self-reported bugs and fixes; README cites real products (clearlist.me) and methodological sources; fact/inference separation is decent. Deductions: key artifacts (PRD, validation report, examples) are not within the provided files for cross-checking; execution claims (e.g., headless-browser render fix) rest on author assertion only.
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 →