DWP v1: from a plan loop to a portable harness

Until now, Deep Work Plan has been a plan-writing loop that one agent follows inside a single session. This run of releases turns it into something a repository can adopt on its own terms — the first version dedicated to reach beyond a single agent and a single sitting.

The skill gains a trust statement with checksummed provenance, so a developer can see exactly what it reads and writes before installing it. Onboarding presets and supported agent hosts broaden considerably, testing discipline becomes a first-class part of every task instead of an afterthought, and a conformance checker lets a repository verify — objectively, not by assumption — that it meets the standard. A state-and-resume protocol means an interrupted plan can pick up where it left off instead of starting over. Under the hood, .claude and .cursor compatibility converge on one canonical .agents directory, so every supported agent reads the same kit.

Together, these releases lay the foundation for what comes next: the repository stops being just where the plan is stored, and starts becoming the agent’s working harness.

Review the skill’s release history for the full sequence.