The video describes Unlazy as a completion-discipline skill for coding agents. Before substantial work begins, the agent writes observable outcomes into a gates file, with each runnable gate specifying a shell command and the output marker that proves success.
A gate counts only when its command exits successfully and the expected evidence appears. The checker records execution context and decisive output, while impossible gates must be explicitly abandoned with a reason rather than deleted or silently ignored.
Because gate commands are executable shell code, Unlazy separates status inspection from approval. The first run shows the resolved command without executing it; a user must review and approve the exact command, working directory, shell, timeout and environment before it can run. Any material change invalidates that approval.
For larger projects, the workflow decomposes work into owned leaves with separate ledgers and integration gates. Parents re-run child checks before accepting completion, and the final report is produced only after root-level verification. The trade-off is extra ceremony, so the video recommends it for long refactors, audits and parallel builds rather than trivial edits.
Watch on YouTube
