A Claude Skill that runs a skeptical-VC teardown on your startup idea

I'll pull the GitHub repo and HN thread for concrete facts, then draft the five analytical paragraphs from those only.A GitHub project from **zszendro** called **vc-teardown** packages a Claude Skill that stress-tests a startup idea as a skeptical VC, then rebuilds it into a plan with a defensible moat, MVP scope, go-to-market, and kill criteria. The public repo is at github.com/zszendro/vc-teardown. On Hacker News it appeared under Claude/Codex/Fable discussion with 1 point and 0 comments at the time of listing (news.ycombinator.com/item?id=49161394).

Technically the skill runs a fixed six-step workflow: intake (at most three clarifying questions), mandatory research into incumbents, free substitutes, adjacent software buyers already pay for, and evidence of real pain, then a teardown of 8–12 numbered challenges drawn from a library of 16 attack surfaces, a moat reframe over 9 patterns, a full plan (market reality, tiered use cases, revenue timing, unit economics with arithmetic, MVP scope and stack, GTM, risk register, falsifiable kill criteria), and delivery with an appendix mapping the original idea to outcomes. Implementation is a SKILL.md plus three reference files—challenge-library.md, moat-patterns.md, plan-template.md—loaded only when needed so context stays cheap until work starts. Install paths cover Claude apps via a .skill release, Claude Code via clone into ~/.claude/skills/vc-teardown, or a manual copy of the vc-teardown folder. License is MIT.

For engineers and builders the useful part is the order of work: research is non-negotiable before advice, every challenge must force a concrete plan revision, and the skill is calibrated to run honest beachhead math (the README’s example is a fully penetrated beachhead that only yields $18k of revenue) and to name the cheapest test that de-risks the biggest assumption—often time, not money—before any build. The 16 surfaces cover competition, free substitutes, monetization, willingness to pay, cold start, distribution, buyer incentive, sales cycle, data availability, regulatory and liability, founder-market fit, capital efficiency, defensibility, feature-not-a-company, retention, and scale honesty. Two moat patterns do most of the work in the design: generate the data instead of aggregating it, and find the incumbent’s metric mismatch.

In market context this sits among several Claude skills that score pitch decks, fundraising narratives, or named companies against simulated investor styles. vc-teardown starts earlier—on a raw idea—and ends on a reframe rather than a grade: kill the version that loses to what is already built, then rebuild around a different moat. The author positions it as a structured second opinion before code, not investment advice, not a substitute for customer conversations, and not a source of conviction.

Practical takeaway: feed it one line or a long brief, and state launch geography plus bootstrap vs. raise up front because those change the analysis. Triggers include describing a marketplace idea, asking to poke holes in a concept, grading a pitch outline, or questions like “validate this idea,” “what does the competition look like,” “find the moat,” and “write me a business plan.” Watch whether the challenge library and moat patterns get real teardown examples and better coverage for hardware, biotech, regulated finance, and deep tech—the areas the README flags as weak today—and treat every output as a hypothesis with kill criteria attached, not a green light to ship.