Release Notes 0.5.43

Release 0.5.43 ships the Ask Inbox sidebar workflow, richer composer routing, a new bundled React guidance skill, and stronger release packaging validation across macOS, Windows, and Linux artifacts.

Highlights

• Ask Inbox sidebar chat: Inbox Agent now includes a right-sidebar Ask Inbox tab with live mailbox-agent progress, final answers, matched email evidence, and a pinned composer for follow-up questions.
• Hybrid mailbox retrieval: Ask Inbox plans broad searches across local FTS, semantic mailbox embeddings, provider-native search, and attachment text before reading evidence and answering.
• Composer integration mentions: the message box @ picker now groups Agents, configured Integrations, and Files. Integration chips persist in prompts and task history while passing soft runtime guidance through integrationMentions.
• @Inbox routing: @Inbox and @inbox ... prompts from the main composer now open Inbox Agent and send the remaining query through Ask Inbox.
• React best-practices skill: the bundled skill set now includes react-best-practices for React and Next.js implementation, review, refactor, data-fetching, bundle-size, and rendering-performance work.

Release Validation

• macOS DMG smoke tests: release packaging now validates updater metadata, mounts the versioned DMG, checks the embedded .app bundle version, and verifies the app executable.
• Windows installer smoke tests: the Windows workflow now uses the shared smoke runner to validate updater metadata, install the versioned .exe silently, check the installed app version, launch it briefly, and uninstall it.
• Linux server smoke tests: the release workflow continues to build the Linux x64 server tarball, verify its checksum, and boot-smoke coworkd-node against /health before publishing artifacts.

Fixes

• Fixed duplicate @ rendering and the React removeChild crash when deleting a raw mention or integration chip.
• Cleared stale Google Workspace refresh tokens after refresh bad-request failures, including credential or scope changes before reconnect.
• Normalized long Azure OpenAI Responses fallback tool-call ids so integration-heavy turns no longer fail on call_id length limits.