unfazed 0.6.0 → 0.8.0

package/CHANGELOG.md

@@ -0,0 +1,88 @@
+# Changelog
+
+## 0.8.0
+
+Fixes from a hands-on Code Mode test (2026-10-22), and the MCP tools the SDK
+lacked. Needs the platform change that adds `writeWorkspaceFiles`,
+`waitForDeployment`, the error and event queries and the new
+`restoreAppVersion` fields; on an older platform `writeFiles` and the deploy
+wait fall back to their 0.7 behaviour.
+
+**Breaking**
+
+- `writeFiles` returns a `WriteFilesResult` (`{ files, written, failed,
+  totalBytes, commit, commits, landed? }`) instead of `WorkspaceFile[]`, and a
+  file that fails throws `UnfazedWriteError` (its `result` and `failed`)
+  instead of `UnfazedGraphQLError`; `allowPartial: true` returns the result
+  instead. Its defaults are now 100 files and 32 MiB per request (were 20 and
+  4 MiB).
+- `getEvents` returns `DeploymentEvent[]` whose `type` is the event's own
+  dotted name (`"deployment.build_finished"`, `"service.vm_started"`), as MCP
+  shows it. The GraphQL enum moved to `typeEnum`; `rawType` stays, deprecated.
+
+**Changed**
+
+- `writeFiles` is ONE request and ONE commit on an app with platform git (the
+  `writeWorkspaceFiles` mutation MCP's `write_workspace_files` uses), with
+  `commitMessage`, and answers the commit: `sha`, `committed`, `waiting`,
+  `waitingCount`, `deploys`, `conflicts`, `writeSHA`, `paused`. A map over one
+  request's limits is several requests, one commit each. `writeFile` and
+  `writeBinaryFile` are one-file batches and answer the file with its
+  `commit`.
+- `deployService/deployStatic/deployBox/deploy({ wait: true })` and
+  `waitForDeployment` wait server-side (MCP `wait_for_deployment`): READY comes
+  back one round trip after it happens, with the full deployment, no poll
+  interval and no re-read. The 9.0 s vs 6.8 s gap in the report was the old
+  1 s × 1.5 poll backoff (READY seen at the 8.1 s poll) plus the full re-read.
+  The polling fallback starts at 0.5 s and caps at 3 s.
+- A deploy with `wait: true` carries `timings`: `platform` (`totalMs`,
+  `buildMs`, `promoteMs`, every measured phase as numbers) and `client`
+  (`createMs`, `waitMs`, `totalMs`, `waitRequests`, `waitMode`).
+  `timings: false` skips the extra read.
+- The platform holds at most 16 long waits of one account at once (and one
+  per GraphQL request). A deploy wait refused past that polls instead;
+  `waitForEvents` throws an error `isTooManyWaits` recognizes.
+
+**Added**
+
+- `help(topic?)` and `REFERENCE`: every method with its signature, in one
+  page (REFERENCE.md, shipped in the package).
+- `waitForCommit(workspaceId)` and `writeFiles(…, { waitForCommit })`; the
+  saved / committed / deployed states are documented.
+- History: `listCommits`, `getCommit`, `restoreVersion` (with `wait: true` it
+  follows the import the restore queues to its READY deployments).
+- Errors: `listErrors`, `getErrorGroup`, `resolveErrorGroup`,
+  `muteErrorGroup`, `reopenErrorGroup`.
+- Events: `listEventTypes`, `pollEvents`, `waitForEvents`.
+- `getDeploymentTimings(id)` and `parseDeploymentTimings(events)`.
+
+## 0.7.0
+
+`deleteApp`, and `deleteWorkspace` names every app it takes offline
+(`confirmAliases`, `liveApps`).
+
+## 0.6.1
+
+`npx unfazed login`, and the SDK uses the stored login.
+
+## 0.6.0
+
+`unfazed/workflow`: the unfazed World for Workflow SDK; `--backend unfazed`
+for Workflow SDK's CLI.
+
+## 0.5.0
+
+Runs: access mode, app alias, integrations on a run, `listRunHarnesses`.
+
+## 0.4.0
+
+Runs: `verify` commands and the harness result.
+
+## 0.3.0
+
+Agent runs.
+
+## 0.2.0
+
+`ensureWorkspace`, `ignoreNotFound`; `deleteWorkspace` fixed; a harder
+`rollback`.