@knpkv/clockify-api-client 2.0.0 → 3.0.0

package/CHANGELOG.md

@@ -1,5 +1,38 @@
 # @knpkv/clockify-api-client
 
+## 3.0.0
+
+### Major Changes
+
+- [#521](https://github.com/knpkv/npm/pull/521) [`28c22ae`](https://github.com/knpkv/npm/commit/28c22ae49825b06472c0d35aa9d7ef92ed5748ff) Thanks [@konopkov](https://github.com/konopkov)! - `jcf auth clockify setup` reported "No workspaces found." and saved nothing, even when the key had workspaces. The generated client could not decode a real `/v1/workspaces` response, and setup turned that failure into an empty list.
+
+  - **`@knpkv/clockify-api-client`**: `WorkspaceDtoV1` now matches what the API returns.
+    - `featureSubscriptionType` is a `string` (e.g. `FREE_2026`), not a `FeaturePlan` object.
+    - `features` is a `ReadonlyArray<string>`.
+    - `workspaceSettings` is no longer modelled, because the upstream schema disagrees with live responses on many nullable and enum fields.
+    - `MembershipDtoV1.costRate` and `hourlyRate`, and `WorkspaceSubdomainDtoV1.name`, accept `null`.
+    - `FeaturePlan` and `WorkspaceSettingsDtoV1` stay exported.
+    - These are incompatible type changes to exported generated schemas, so this is a major release. The previous types could not decode a real response, so code relying on them could not have run.
+  - **`@knpkv/jira-clockify`**: setup now fails with a typed `ClockifyRequestError` that names the failed request, instead of reporting no workspaces. "Invalid API key" is reported only when Clockify rejects the key with 401 or 403; before, any failure looked up the user as an invalid key. The account lookup is exported as `loadClockifyAccount`, so it can be tested against real-shaped responses.
+
+- [#483](https://github.com/knpkv/npm/pull/483) [`da0e5ff`](https://github.com/knpkv/npm/commit/da0e5ff70ec5926beaeceff060a37b2b0f27b6f9) Thanks [@konopkov](https://github.com/konopkov)! - Discover every worked ticket from Claude Code and Codex sessions.
+
+  Presence is now a supervised turn: a typed prompt, including one queued while the agent was busy, opens a turn and the agent's work inside it counts until the turn ends, a task notification or auto-continuation takes over, or the idle cap passes. Task notifications, auto-continuations and `isMeta` lines no longer count as typed. Legacy Codex tool-call response items keep a turn alive. `decodeTranscript` now requires `idleCapMs`; `decodeSessionLines` and `SessionLine` are new.
+
+  Parallel stretches no longer drop tickets: every attributed ticket keeps a share, overlapping short tickets stay co-owners, and the web calendar shows suggestions down to Jira's one-minute minimum. When minutes are scarce, open-sprint tickets assigned to you rank first, then tickets not yet logged that day; the same facts appear as tie-breakers in the attribution prompt. Inside one unbroken stretch each ticket now gets a single block, ordered by first activity and packed with no gaps; a ticket that cannot reach a minute folds into the one ranked above it.
+
+  Ignore a ticket in every week with `jcf config set session-ignore <KEY>` or the web's Ignore button, and restore it from the Ignored tickets list. An ignored ticket is never suggested or offered to the attribution agent; a branch or path match to it falls through to the session's other candidates, and its parallel time goes to the tickets it ran alongside. Reports carry its raw time as `ignored`. In the web calendar, back-to-back short suggestions show as one card per stretch, and the page uses the full window width.
+
+  An orchestrating session nothing else places is split across the open-sprint tickets it mentions, by mention count, per active stretch (new attribution signal `split`); deterministic reads such as `jcf watch` leave it unplaced. The web lists low-confidence matches with a "Log as" action, and saved entries can be deleted or moved to another ticket; a delete releases its session claim so the time is suggested again, and a move carries the claim to the replacement. `SavedEntries` gains `remove` and an optional `ticketKey` on update.
+
+  Quick approvals are confirmed in batches of up to fifty under one provider re-read (new `/api/rows/confirm-batch`), Jira worklogs are read eight issues at a time, and idempotent Jira reads retry a dropped connection twice, so a long queue no longer waits on one full re-read per approval.
+
+  A provider window that needs manual review no longer fails a read: proposals carry a per-provider `writeBlocked` hold, the web shows the hold and disables writes to held providers, and writes keep refusing.
+
+  Ticket moves persist a replacement intent before creating provider time. Uncertain or partial moves stay held across restart and cannot create another replacement on retry; a verified pair can be resolved by explicitly deleting either entry. Ordinary replacements retain their verified ID and start so extending a reviewed window does not mistake the move for unknown earlier time. The private source ledger upgrades to version 5 while retaining existing claims and holds.
+
+  Breaking (`@knpkv/clockify-api-client`): `TimeEntryWithRatesDtoV1.costRate` and `hourlyRate` are now `RateDtoV1 | null`, matching the live API, which returns `null` when no rate applies.
+
 ## 2.0.0
 
 ### Major Changes