READMENOT — a marker for code not meant for humans

A short proposal by William Woodruff (yossarian-blog, ENOSUCHBLOG) for a READMENOT file: a marker a project (or its agent) can drop to honestly declare that the code within isn't intended for direct human consumption. Where a README tells a human where to start reading, a READMENOT tells them not to bother — use an LLM or another tool to interact with the project instead.

The argument

Woodruff spends a lot of his day reading code online, and increasingly that means accidentally reading "slop." His objection isn't slop's existence — it's the bait-and-switch. Like the written word, he wants to be informed before he spends human attention on machine output. He's a believer in giving people a way to express honest intentions: he sometimes wants to drop slop on the internet himself (to save for later, or for others to reuse without reading), without misleading anyone about the effort behind it.

So READMENOT is the honest-intentions signal. Its mere presence is an unambiguous warning that the code is unsuitable for unwitting human comprehension. A good default would be a short human-friendly note explaining why — e.g. "You're reading a project that isn't intended for direct human consumption. You may wish to use an LLM or another tool to interact with this project." He doesn't care what it's called or whether it's a file at all; any consistent marker would do. READMENOT is just funny.

His definition of slop

The useful, quotable part. Woodruff draws the line at either of two sufficient qualities — a codebase is slop if it either (1) is developed primarily without human supervision, or (2) reflects a fundamental lack of operator understanding. The two often co-occur but either alone suffices. This is the slop-marker-convention framing: slop is defined by its production process and the author's understanding, not by how the code reads on the surface.

He's careful to timestamp the claim (July 2026) because the target moves: "2026's slop is not 2025's slop" as models improve, so "slop" is hard to grant as a static qualifier.

The analogy that carries it

The mental model is generated code and the @generated marker. It's not annoying for generated code to exist; it's annoying to find yourself reading it because it lacks the marker that would have let you make an informed decision not to. READMENOT extends that courtesy to slop. Crucially, it's not "never read this" — there are good reasons to read slop as a human, security research being the obvious one — it only informs you what to expect if you do.

Where it sits

The mirror-image proposal is human-made-disclosure — declaring human authorship instead — argued for in no-ai-statements (James Zhan) with the handmade-goods labelling tradition as precedent. The incentives differ sharply: a READMENOT costs the producer something to attach, so declaring it is credible, while a "no AI was used here" footer is free to fake. Both rest on the premise catalogued in ai-detector-arms-race — provenance is not recoverable from the artifact's surface.

This is the honest-signal counterpart to the reader-side problem in credibility-as-slop-test: Theocharis argues you can't tell slop from the text's surface, so the burden falls on the author to stake credibility; Woodruff proposes the author instead flag the code up front so the surface question never arises. Both put the obligation on the producer. Compare the effort-signal collapse described in llm-as-average-democratizer and the broader mood in cult-of-vibe-coding. The @generated-style marker idea also rhymes with the provenance instinct behind reviewing-ai-code — knowing how code was produced changes how you should engage with it.