Repository navigation
v0.0.5
DSH Sidebar v0.0.5
Compatibility
This release adapts DSH Sidebar to the new DeepSeek Harness Remote protocol. It requires DSH 0.1.2-rc.1 or a compatible newer release and has been tested with 0.1.5-rc.1 (latest). DSH 0.1.1 and earlier are no longer supported by this extension release.
Update your local runtime before using this version:
npm install -g @deepseek-ai/dsh@latestWhat's new
- Support for the new authenticated DSH connection, event streams, session history, approvals, and follow-up questions.
- Reconnect to an existing runtime without restarting it or resending tasks. Queued messages, pending approvals, and background tasks are restored after reconnecting.
- Connect to an authenticated external runtime, or explicitly start a managed runtime.
- A session center for finding and switching conversations, with per-session drafts and attachments.
- Paste images directly from the clipboard, with sending delayed until pending image imports finish.
- Markdown tables and mathematical formulas now render in chat.
- Improved Permission, Plan, model, reasoning effort, and Agent preset synchronization, plus runtime plugin inventory and settings conflict protection.
Fixes and safeguards
- Restore tool result cards and native file review with the new DSH history format.
- Prevent runtime or workspace changes from interrupting background tasks.
- Discard unfinished response fragments after terminal connection failures and retry canceled attachment loads after reconnecting.
- Allow manual reconnects to tokenless reused runtimes without attempting token authentication.
Validation
- Typecheck and build passed; 506 automated tests passed, 1 skipped.
- Ubuntu and Windows CI passed for the compatibility changes.
- Real global DSH smoke tests covered streaming chat, history, images, slash commands, approvals, questions, queues, settings, and reconnects.
- Native VS Code tests covered debugger breakpoints, stepping, runtime values, file edits, Diff Editor, and background jobs.
Runtime integration tests used a local deterministic model endpoint; no paid model requests were made.