v0.1.0 — first public release

First public release. Three packages, all MIT.

Package	What it is	Install
SignsOfAI.Core	The engine: rule packs, analyzers, scorer, originality comparison. Pure .NET, no browser.	dotnet add package SignsOfAI.Core
SignsOfAI.Cli	Check drafts from the terminal, or gate prose in CI with --max-score.	dotnet tool install --global SignsOfAI.Cli
SignsOfAI.Mcp	MCP server — the same engine as tools for Claude Desktop and any MCP client.	dnx SignsOfAI.Mcp --yes

What it does

Most AI detectors hand you a number. "87% AI" is unfalsifiable: a teacher can't accuse on it, a student can't appeal it, and a writer can't learn from it.

This one shows the evidence instead. It flags the specific tells — overused vocabulary, rhetorical crutches, syntactic patterns, and the statistical rhythm of sentence lengths (burstiness) — and for each one it says what to do about it. It also compares documents against each other and surfaces the passages they share, including reworded paraphrases across languages.

English and Spanish. The Spanish rule pack is derived from scratch, not machine-translated.

It runs on your machine. The browser app is Blazor WebAssembly; nothing is uploaded, no account. Two optional features (perplexity and cross-language paraphrase) call a server and say so before they do.

Also in this release

• Web app: https://peopleworks.github.io/SignsofAI/
• /signs-of-ai agent skill for Claude Code and other harnesses
• 75 tests, gating this release

It is a heuristic linter, not a verdict machine. That is deliberate.