unlazy

What problem does it solve? AI agents frequently report work as done when it is incomplete, silently reduce scope, or make unmeasured completion claims. This Skill makes completion testable by requiring acceptance gates written before work begins, executed checks with explicit approval, and re-verified evidence before any done report. ## Core Features & Use Cases - Acceptance gate ledgers: Write GATES.md files where each gate pairs a CHECK shell command with an EXPECT output match, so completion is proven by exit code plus output rather than a confident summary. - Approval-gated execution: CHECK lines run only after explicit --approve, with approvals bound to the exact command, expectation, working directory, shell, timeout, platform, and PATH stored outside the repository. - Orchestrated parallel work: Decompose large tasks with the Depth Tree into scoped pipelines under .unlazy/<scope>/, coordinate leaves with OWNS leases and native dispatch waves, and re-verify returned work with --reverify. - Use Case: Ask an agent to refactor a payment module across multiple files. The Skill writes per-leaf gate ledgers first, dispatches independent leaves in sealed launch waves, re-runs every gate on returned work, and blocks the final report until all gates pass or are explicitly abandoned as handoffs. ## Quick Start Ask the agent to use unlazy to tree 5 refactor the payment module and verify every migration path with runnable gates before reporting completion.