startup-skill: the agent that tells you if your startup idea should die

npx skills add ferdinandobons/startup-skill

Ask a chatbot “I want to build X — is it worth it?” and you get the usual MVP-and-iterate pep talk. Nobody runs the searches for the incumbent that already owns the market or the startup that publicly died trying. startup-skill, an MIT-licensed skill bundle by ferdinandobons for Claude Code, turns the question into actual due diligence: market research, competitor battle cards, positioning, financial projections and validation experiments — 30+ structured documents from one prompt, starting with a five-minute Pre-Flight check that pours the first cold water. The README puts it plainly: “If your idea should die, it will tell you.” A short video on the WeChat channel AIFeed (September 28) introduced it to a wider audience; the repo had about 1.1k stars as of September 28, 2026.

Eight phases and two gates

startup-design is the main skill: eight sequential phases — INTAKE → BRAINSTORM → RESEARCH → CUSTOMER DISCOVERY → STRATEGY → BRAND → PRODUCT → FINANCIAL → VALIDATION — plus the five-minute Pre-Flight and two hard gates. Every phase writes Markdown deliverables and updates PROGRESS.md, so an interrupted session resumes from the last checkpoint. Pre-Flight does exactly two things: search for a well-funded incumbent with 10k+ customers on this exact problem, and search for companies that failed publicly at the same idea. The red flags land in document one, before any vision talk. Fail the Research Gate and the process stops there; the Customer Discovery gate won’t open until at least three real user interviews are logged. Say “quick validation” and a compressed pipeline still returns a go/no-go signal.

What it pins down

• Cold water before vision: Pre-Flight runs two or three targeted searches for dominant incumbents and public precedent failures, and flags them in the first document — you don’t walk eight phases on a hallucination.
• Gates that gate: the Research Gate needs go/no-go evidence; Customer Discovery needs at least three logged interviews. The agent cannot hand-wave its way through.
• Battle cards from real sources: startup-competitors runs three research waves over 5–8+ competitors; the pricing landscape and feature matrix come from real reviews, forums and community threads, not invented comparison tables.
• Pitch prep with the Q&A: startup-pitch produces 10-min, 5-min and 1-minute scripts, an investor email and a slide-by-slide deck outline, with a scoring rubric, Q&A prep and investor roleplay — and it works standalone.
• Resumable: everything lands in PROGRESS.md and phase files; when the token budget dies, say “resume” next session.

Who it’s for

Indie developers and small teams who want the idea hammered before writing code — exactly the “built it and nobody came” scenario the AIFeed video warns about. Three catches: a full run fans out multiple research agents and burns serious tokens (the README itself suggests Claude Max 5x); the latest release, v1.10.0, has been quiet since July 1, 2026 — less of a problem for a docs-and-process bundle than for an API-dependent tool; and the market data comes from agent web searches, so spot-check the numbers yourself. For harder Product Hunt data, pair it with the producthunt skill.