Skip to content

About

Scan a plan, deck, or spec for buzzwords, unsourced claims, vanity metrics, and missing failure cases.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

5 Commits

Folders and files

Repository files navigation

Vaporware Detector

skills.sh License: MIT

What does this document actually say?

A Claude Code skill that scans a business plan, funding deck, product proposal, or requirements doc for vapor — buzzwords, unsourced claims, vanity metrics, undefined core terms, and missing failure cases — then gives a credibility score and a concrete rewrite.

Not a roast. A translation layer: every sentence that cannot be checked gets flagged, and every flagged sentence gets a rewrite that could be checked.


What it detects

Detector Example
Buzzword density "a full-stack, closed-loop ecosystem" / "seamless, world-class, end-to-end solution"
Unattributed claims "the market responded enthusiastically" — how many people, when, saying what?
Vanity metrics registrations instead of week-two retention; downloads instead of DAU
Undefined terms "our moat is the algorithm" — which algorithm, and why can't others copy it?
Missing negative space No risks, no failure conditions, no "we stop if X is false"

Scored 0–100 across five dimensions.


Install

Option 1 — skills CLI

npx skills add yangben2019/vaporware-detector

Option 2 — Claude Code plugin marketplace

/plugin marketplace add yangben2019/vaporware-detector
/plugin install vaporware-detector@vaporware-detector

Option 3 — manual

git clone https://github.com/yangben2019/vaporware-detector.git /tmp/vd
cp -r /tmp/vd/skills/vaporware-detector ~/.claude/skills/

Usage

What does this funding deck actually say? Audit it before I forward it.
Is this proposal real? Audit it for hype before I sign off.
Which claims in this requirements doc cannot be verified?

The skill quotes the original text for every finding — it will not paraphrase your document, because paraphrase reads as misrepresentation.


Design notes

  • Quote, don't summarize. Every finding cites the source sentence verbatim.
  • Separate "vague" from "wrong." Most problems are unclear claims, not lies. The wording targets unverifiability, never accuses the author of deception.
  • Rewrites must be checkable. "Add more data" is not a rewrite; "X dropped from 4h to 1.5h (internal, 2026 Q2)" is.
  • No industry judgment. The skill evaluates whether the argument holds, not whether the business will work.
  • Short documents get no total score. Under one page, there is not enough to score fairly.

License

MIT © 2026 yangben2019

About

Scan a plan, deck, or spec for buzzwords, unsourced claims, vanity metrics, and missing failure cases.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors