Git for specs.
People review and approve intent. Agents turn approved intent into stacked, previewable commits. Code is the release artifact of a spec, not the thing you edit.
The loop
- 1
Write
Describe a change in one line. An intake agent reads your specs and open changes, asks what it needs to, and drafts an OpenSpec change you review.
- 2
Preview
Before any code: a mid-fidelity page mock for changes people will see, or an architecture diagram for ones they won’t. You approve it.
- 3
Build
Each task becomes one commit, stacked on the last, built by an agent in its own sandbox and checked against the spec’s acceptance checks, with a live preview per layer.
- 4
Publish
Stage a release candidate restacked onto main, QA it on its own preview, then merge and deploy exactly what you checked.
Coming from coopcli?
coopcli is now Specsesh. Your account, organizations and roles came with you: sign in with the account you already have.