Synara 0.3.7 release notes.

Synara 0.3.7

• Update downloads show their progress — When Synara is downloading a desktop update, the sidebar update button now shows a live percent badge so you can tell whether it is moving or nearly ready.

  The desktop update helper now clamps reported download percentages, hides them outside active downloads, and covers edge cases for null, negative, oversized, and fractional progress values.

• Claude auth checks are less jumpy — Claude sessions are less likely to be marked logged out during refresh-token races, especially when health checks and the macOS credential keepalive run near the same time.

  claude auth status is now serialized through a shared lock, retried once for structured false negatives, and can fall back to verified local credential metadata before showing an unauthenticated state.

• Turn checkpoints are harder to confuse — Synara preserves the first pre-turn snapshot more carefully, so transcript diffs and restore points better match what was on disk when you pressed send.

  Checkpoint capture now has single-flight behavior per repo/ref, a bounded capture timeout, first-writer-wins skipIfExists baselines, and extra recovery when a startup or backup path missed the original message baseline.

• The sidebar chat list behaves more naturally — The Chats section now lives with the rest of the scrollable sidebar content, has a familiar disclosure chevron, and keeps the footer focused on account/update controls.

  Sidebar chat collapse state now exposes aria-expanded, reuses the shared disclosure chevron, and separates chat rows from the footer while preserving sort and new-chat actions.

• First sends keep the workspace calm — Starting from an empty chat no longer opens the Environment panel by surprise after the first message turns the landing view into a transcript.

  ChatView now records a closed environment-panel preference when sending from the centered empty landing, preventing the default-open policy from popping the panel into a just-started conversation.

• Provider health parsing is easier to trust — Provider CLI output handling and Claude auth interpretation now have clearer, isolated parsing paths with more focused tests around failure and metadata cases.

  Generic CLI-output helpers moved out of ProviderHealth, Claude auth parsing moved into a pure module, and new tests cover auth JSON markers, credential summaries, the auth-status lock, checkpoint single-flight behavior, and provider health edge cases.