Leonxlnx/unlazy
★ 4,002 stars
Covered in week 2026-W38 · the 2026-09 report
Anti-laziness skill for AI agents. Core: the Depth Tree method, which splits a task N layers deep and gives every leaf the full time budget of the whole task, so effort multiplies with depth. Grounded in 2025-2026 research on model laziness, underthinking and premature completion.
What unlazy is
Clone the repository into the relevant directory. Invoke it as /unlazy where slash skills are supported, $unlazy in Codex, or by a natural-language trigger from the skill description.
From the repository README · read it on GitHub →
Transcript
Your AI agent calls the task done while it's still half-built. unlazy answers with its Depth Tree, giving every leaf the whole task's time budget. You write an acceptance ledger, gate-check.mjs runs every CHECK command, and --reverify re-runs all gates, even the finished ones. Since 2.1.0 isn't a tagged release, pin an exact commit, then delegate deep work with proof.