# vibe-check: a planning skill that runs before you write any code > vibe-check is a single SKILL.md that guides a complete beginner from a vague app idea through problem discovery, experience mapping and a growth loop, then hands the AI tool a plan it can follow. Its output is one self-contained HTML PRD, and its sharpest dependency is reading Reddit. **TexasBedouin/vibe-check** — By a 12-year product manager who builds 0-to-1: takes a beginner from a vague idea to a buildable plan, then guides the build (GitHub basics, clean-code habits, a verify-and-iterate loop, a checkup for the mess). For Claude Code, Codex, and Antigravity. grill-me is for engineers, vibe-check is for everyone else. - Repository: https://github.com/TexasBedouin/vibe-check - Website: https://www.producthunt.com/products/vibe-check-6?launch=vibe-check-7 - Stars: 614 · Forks: 63 - Language: HTML - License: MIT - Published: 2026-09-17 · Updated: 2026-09-17 · Language: en - Canonical page: https://hysenlabs.com/projects/texasbedouin-vibe-check ## It is one SKILL.md, not a plugin or an app The entire product is a text file. There is no binary, no service and no dependency: SKILL.md sits at the repository root next to a VERSION file, a CHANGELOG, a RELEASING note, a bump script, an assets folder, a references folder and an examples folder. With Claude Code you install it and it becomes a skill; with another tool you paste the same file into your system prompt or project instructions. That explains the recorded primary language being HTML. The skill's own output is a self-contained PRD you open in a browser, and the examples folder is full of them: `clearlist-prd.html`, `clearlist-blueprint.html`, `plant-blueprint.html`, a `clearlist-session.md` transcript and an AuDHD validation report. What the skill produces is the artifact, and the artifacts are web pages. The positioning is narrow on purpose. The repository description splits the audience in two: grill-me is for engineers, and vibe-check is for everyone else. ## Three on-ramps decide how much work you repeat Most coding skills assume you already know what you are building. This one starts earlier, and it acknowledges that some readers have already done the earlier parts. There are three entry points and you pick one deliberately. The full journey takes a vague idea in and returns a validated, buildable plan; that is the default. Validate only asks whether the idea is worth building at all and returns a straight evidence-backed answer plus a findings summary, with the explicit promise that a dead idea gets no blueprint. Plan only assumes the validation already happened and takes your research or your findings summary as the starting point. The sibling skills stay out of the way: grill-me is aimed at engineers, this one at everyone else. ## The Reddit fetch ladder is the part that can fail Step one of the method is not writing a spec, it is finding out what you actually need: mining Reddit for real pain, running a competitor gap analysis, and scoring opportunity. The evidence then becomes an Opportunity Map of scored, evidence-tagged needs alongside a Competitor Matrix showing where the gap sits. That first step depends on a third-party site, and the release history shows the project repairing exactly that dependency. v2.6.1, published on 2026-07-17, is titled as a Reddit fetch ladder that reads reddit.com directly, which means earlier versions were reaching the same content by a route that stopped working. It is a small reminder about the whole approach: a plan whose problem statement comes from someone else's forum is only as good as the fetch on the day you run it. ## The output is one self-contained file you open in a browser The deliverable is a plan document structured as the AI tool's own onboarding manual, plus an interactive PRD that the human opens in a browser, with the whole session in one navigable, self-contained file. Self-contained is doing real work there: no build step, no server, no assets to lose. User flows arrive as drawings in the engine's visual style with mermaid as the fallback, and the architecture diagram carries beginner-friendly labels rather than component names. The rest of the plan is the ordinary list of sections a product person would hand a developer: the problem statement in your own words, Crazy 8 sketches and the converged direction, an Experience Blueprint, a Story Map with every capability sorted into V1, V2 and Later lanes, a feature breakdown, a tech stack with what each piece does and why and what it costs, a data model in plain language, a monthly cost breakdown with free tier details, and a distribution section that starts with your first ten users. ## Build Phase 0 is the riskiest assumption, not the first screen The one item in the plan worth singling out is the riskiest assumption, which is scheduled as Build Phase 0. It is the cheap test that runs before any real code exists, and the sequencing is the point: you are asked to find out whether the assumption is false while finding out is still inexpensive. The skill also front-loads the decisions people forget. Before the stack gets chosen, it surfaces auth, databases, payments, hosting and legal, since those are the choices that are expensive to reverse and easy to skip. For two-sided ideas it insists on discovering both sides rather than the one the founder happens to be, and then works through a cold-start plan so the product does not launch into an empty room. Growth is treated as part of the core flow rather than an afterthought, on the reasoning that a loop bolted on later needs a constant push while one in the flow compounds. ## Checkup Mode exists for the mess you create later The skill covers the part of the journey that comes after the demo works. Alongside the planning it teaches build-time basics in language for someone who has never touched code: the difference between local, GitHub and live, how to save and back up code through commit, push and deploy, and how to keep secret keys safe. Build checkpoints are built into each phase, so the AI stops and explains what was just built, why, and what comes next instead of running for twenty minutes. Checkup Mode is the later stage: it looks over a codebase that has grown messy and tidies it safely, so the tool keeps building cleanly rather than breaking what already works. Read together with the verify-and-iterate loop in the repository description, it is an admission that the interesting failure is not the first build, it is the fourth week of a project nobody has reviewed. ## Installing is an npx line or a clone into .claude/skills The easy route goes through the open skills CLI, which installs across agents: ```bash npx skills add TexasBedouin/vibe-check ``` The other route drops the folder straight into a project: ```bash git clone https://github.com/TexasBedouin/vibe-check .claude/skills/vibe-check ``` Which route you pick decides how updating works. With the CLI it is `npx skills update`; with the clone it is `git pull` inside `.claude/skills/vibe-check`. The same skill works for Codex and Antigravity, though there the instruction is to copy the contents of SKILL.md into the tool's system prompt or project instructions rather than to rely on a skills directory. Activation is conversational, with three phrasings shown: ask it to help plan the app, ask whether the idea is worth building and to reality-check it, or say the idea is already validated and ask for the build plan. ## MIT licensed, frequently released, and sold alongside The repository is MIT licensed and sits on the master branch rather than main, with the last push on 2026-07-17. Three releases landed in a week that month: v2.5.1 on 2026-07-10 marking a Product Hunt launch, v2.6.0 the same day described as the lean core release, and v2.6.1 on 2026-07-17 fixing the Reddit fetch path. A VERSION file and bump.sh suggest the version is maintained by script rather than by hand. The author also sells the same work commercially, listing idea validation, a vibe-coded app rescue audit and a validated MVP blueprint as paid services, with one free sample report linked from the README. The paid offering and the free skill line up on purpose: the free method is the discovery and planning work, and the paid version has someone else drive it. Judge the skill on whether it catches a bad idea, since that is the claim it makes. ## Conclusion Use vibe-check if you have an idea you cannot yet defend and no practice judging whether it is worth building, and you want the questions asked before code rather than after. It earns less if you already know the problem and want to go straight to a stack decision, since the discovery phase is the bulk of the session. Before trusting its evidence, spot-check the Reddit mining against the sources it cites, and treat the blueprint as something your tool executes rather than something it can skip steps of. ## FAQ ### What is the vibe-check skill? A single SKILL.md for AI coding tools that guides a complete beginner from a vague app idea to a buildable blueprint. It runs problem discovery, experience mapping, stack selection and a growth loop before any code is written. ### How do I install vibe-check with Claude Code? Run npx skills add TexasBedouin/vibe-check through the open skills CLI, or clone the repository into .claude/skills/vibe-check. Update later with npx skills update, or with git pull inside the cloned folder. ### Can I use vibe-check with Codex or Antigravity? Yes. The skill targets Claude Code, Codex and Antigravity. Where a skills directory is not available, copy the contents of SKILL.md into the tool's system prompt or project instructions. ### What does a vibe-check session produce? A plan document structured as the AI tool's onboarding manual plus a self-contained interactive PRD you open in a browser, containing an Opportunity Map, a Competitor Matrix, Crazy 8 sketches, an Experience Blueprint, a Story Map, user flows, architecture, tech stack, data model, cost breakdown and a distribution section. ### What does vibe-check do about code that was already built? Checkup Mode looks over a codebase that has become messy and tidies it safely, so the tool keeps building cleanly. The skill also teaches the basics of commit, push and deploy, local versus GitHub versus live, and keeping secret keys safe. ## Sources - [License: MIT](https://github.com/TexasBedouin/vibe-check/blob/master/LICENSE) - [Project website](https://www.producthunt.com/products/vibe-check-6?launch=vibe-check-7) - [README](https://github.com/TexasBedouin/vibe-check/blob/master/README.md) - [Releases](https://github.com/TexasBedouin/vibe-check/releases) - [TexasBedouin/vibe-check on GitHub](https://github.com/TexasBedouin/vibe-check) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/texasbedouin-vibe-check