Synara 0.3.8 release notes.

Synara 0.3.8

• Grok resume handles long sessions more safely — Grok and other ACP-backed sessions are better at resuming after compaction or reconnects without replaying messages into the wrong runtime or growing memory unexpectedly.

  ACP resume now drops replay before the event consumer is attached, seeds compaction quiet windows from response timing, hardens provider/runtime ingestion, and covers JSON-RPC/session-runtime edge cases that previously made resume replay fragile.

• Worktree setup failures recover cleanly — When a worktree setup step fails, Synara now shows the failed setup state, keeps the timeline from looking stuck, and resets the local dispatch when you send again.

  ChatView and timeline setup snapshots now carry explicit setup step status, targeted cleanup for failed local dispatches, and focused browser/unit coverage for new-turn reset and setup rows.

• Automation messages are labeled — Prompts sent by an automation now carry a lightweight 'Sent via Automation' label above the bubble, so you can tell at a glance which turns you typed and which a scheduled or heartbeat run kicked off.

  User turns dispatched by the automation engine now persist a dispatchOrigin on the message end to end (command → event → projection → snapshot), and the transcript renders a clock-marked chip that mirrors the existing steering label.

• Approval prompts are easier to answer — Pending approvals now use a clearer shared choice-row layout, with steadier panel behavior and browser coverage for allow/deny decisions.

  ComposerPendingApprovalPanel now shares ComposerChoiceRow structure with pending inputs, tracks decision actions in browser tests, and trims duplicated action styling.

• Startup does less surprise work — Synara avoids unnecessary provider refresh work during startup and gates Claude keepalive behavior more carefully, so opening the app is calmer and less likely to fight credential checks.

  Server startup no longer runs provider refresh eagerly, Claude keepalive respects auth-state timing, provider usage hooks handle inactive summaries more predictably, and related settings/server query invalidation paths have tests.

• Folded work rows report time more accurately — Collapsed transcript segments now show a 'Worked for' duration that spans the whole folded section, not just a single row inside it.

  Timeline duration grouping now tracks folded row boundaries and tests the aggregate timing behavior so compact transcripts better match what actually happened.