What it does

This is a plain-English glossary of the jargon that accumulates around AI-assisted coding — terms like “harness,” “prefix cache,” “sycophancy,” and “vibe coding.” Each entry gets a concise definition, a note on what to avoid saying, and a short dialogue showing how developers actually use the term. The README is auto-generated from Markdown source files; the repo itself is mostly content, not code.

The interesting bit

The framing is deliberately skeptical: it calls out that “a whole VC-funded economy benefits from keeping it hard to understand.” The definitions are opinionated and practical — “training” is “a months-long process by the model provider” you can’t do yourself; “non-determinism” is something “there’s no setting you can flip to make it go away.” It’s less encyclopedia, more field guide written by someone who’s debugged the bill.

Key highlights

• Covers 50+ terms across seven sections: model internals, sessions/context, tools/environment, failure modes, handoffs, memory/steering, and work patterns
• Each entry includes realistic usage dialogue — e.g., explaining why your bill spiked because “every turn re-sends the whole session”
• Distinguishes “model” from “harness” (Claude.ai vs. Claude Code use the same model, different harnesses)
• Explicitly notes what terms to avoid and why (e.g., don’t say “word” when you mean “token”)
• Auto-generated from dictionary/*.md via npm run generate

Caveats

• The README is the entire product; there’s no searchable site or CLI tool in the repo itself
• Some entries are truncated in the source (the README cuts off mid-sentence on “prefix cache”)
• Strong editorial voice means some definitions are arguable — “vibe coding” gets defined as a pattern of work, which not everyone will agree with

Verdict

Worth bookmarking if you’re using AI coding tools and tired of pretending you know what “autocompact” means. Skip it if you already lecture people on attention budgets at parties.