Versions & releasing
Changing how the AI behaves works like releasing software: freeze a draft, test it, then go live.
Because behavior changes can break things quietly — a reworded glossary can send questions down the wrong path without anyone noticing. So every pack carries a test suite, and the database itself refuses to make a version live until every test has passed for that exact draft. Nobody can skip it — not through this screen, not through any other tool. Think of it as code review plus automated tests, for the AI’s behavior.
Releasing v3
See the test resultsTakes a snapshot of all 14 pieces exactly as they are right now.
The test suite runs against this exact draft — not against “the latest”, this one.
5 routing tests, 3 content tests, 1 quality score.
v3 becomes the live version; v2 is retired to history automatically.
Version history
| Ver | Status | Tests | Date | |
|---|---|---|---|---|
| v3 | Draft | Passed 9/9 | Jul 29, 2026 | |
| v2 | Live | Passed 9/9 | Mar 14, 2026 | Retires when v3 goes live |
| v1 | Retired | Before testing existed | Jan 30, 2026 |
Behind this page/templates/harnesses/:harnessPackKey/versions · catalog publisher
Retrieve
TemplateVersions filtered to TemplateType='HarnessPack', joined to the latest HarnessEvalRuns per version.Save
sp_SnapshotHarnessPack → sp_QueueHarnessEvalRun → sp_PublishHarnessPackVersion; the prior published version auto-archives.Success
- The gate lives in the proc — the UI can gray the button, SQL enforces it.
- A Passed run binds to the exact TemplateVersionId; re-snapshot means re-eval.
- The seed / upgrade worker only ever reads Published (sp_GetPublishedHarnessPack, base-first).
CLI handoff
Implement this scaffold from the structured contract, then remove hard-coded preview rows. The source of truth is CLI Handoff and admin-cli-manifest.json.
Server-inject identity and scope values; never trust browser-supplied account, app, tenant, user, entitlement, price, or permission identifiers. Preserve the loading, empty, forbidden, failed, retrying, and completed states shown by the preview.