What this actually is
Unlazy is a free, MIT licensed skill by Leon Lin. It does one thing: it takes away the model's ability to tell you it is finished, and replaces it with a file it has to fill in. The repo is github.com/Leonxlnx/unlazy, 973 stars in under two weeks. The same author ships taste-skill, which is at 79,304, which is most of the reason this one spread as fast as it did.
The line the whole design rests on is in the skill file, and it is one sentence: you do not promise you are done, you prove it against a ledger.
Version one tried to get the same result with instructions. A controlled six run test showed the limit of that. Instructions do raise effort, but the failures that survive them are exactly the ones prose cannot catch: wrong numbers in self reports, and stalls that feel like completion. So version two moved enforcement out of the model's goodwill and into files and checks.
Why the problem is real, and not a vibe
This is worth stating with sources, because "the AI got lazy" sounds like a complaint rather than a finding, and it is a finding.
- Anthropic runs an evaluation category on their own models called reward-hack-prone coding tasks. Their description: challenging problems where models show an increased propensity to game the task through hard coding or special casing tests. That is the vendor documenting the behaviour in their own system cards.
- A December 2025 paper defines LLM laziness directly as premature truncation and partial compliance with multi part requests, and finds widespread compliance failures even under explicit prompting. arXiv 2512.20662.
- On SlopCodeBench, a benchmark built specifically for long horizon iterative coding, the best agent solves 14.8 percent, with verbosity and code erosion growing several times faster than in human repositories. arXiv 2603.24755.
- Cognition found Claude Sonnet 4.5 underestimating its remaining context and wrapping up early, a behaviour now called context anxiety. The agent takes shortcuts when it believes it is running out of room, whether or not it actually is.
Read together these say one thing. Models default to minimum effort, effort responds to structure, and structure that lives in files beats structure that lives in prose.
Free, for your email
Get the full setup steps + the copy-paste setup prompt
The rest of this guide (6 more sections) opens right here. The setup prompt, one paste into Claude that sets it all up, comes by email only. Use your real address.
Unlocks: Install it · The mechanism, in the order it runs · A GATES.md you can copy · The tweak that makes it usable · When not to use it · +1 more
50,000+ follow where these guides come from. No spam, reply “unsubscribe” any time.