DeepSeek Harness plugins

Give text-only agents eyes with image Q&A, OCR, grounding, UI restoration, and pixel diff tools.

A full terminal interface for dsh with sessions, rewind, approvals, image support, and live agent state.

Create, edit, and manage sandboxed JavaScript tools from the DeepSeek Harness web interface.

Search and attach workspace files to prompts with fast, familiar Codex-style @file mentions.

Open any DeepSeek Harness workspace in VS Code directly from the web workspace menu.

Get desktop notifications when dsh turns finish, with outcome and keyword-based controls.

Roll a conversation and its workspace back to any turn with a persistent change ledger.

Render interactive HTML labs and charts as sandboxed cards inside the dsh conversation stream.

Add structured image evidence, OCR, layout, and semantic analysis to text-only agents.

Use OpenAI Codex through device-code OAuth, token refresh, and a web settings panel.

Navigate long sessions with a Git-style timeline, message previews, and one-click jumps.

Connect Lark or Feishu chats to dsh agents, replies, approvals, and interactive cards.

Add bounded, approval-gated, auditable memory across DeepSeek Harness sessions.

See conversation progress, preview turns, and jump to the current or earlier messages.

Discover verified dsh bundles and install pinned GitHub sources from the web settings UI.

Display DeepSeek balance, remaining ratio, and today’s spend in the sidebar footer.

Collect ten or more structured answers in one full-screen questionnaire and recap.

Let agents inspect, install, and verify apps on HarmonyOS devices through HDC tools.

Ask a read-only second model to review approval requests with auditable allow or deny reasons.

Give agents an offline guide to native dsh capabilities, workflows, and subagent patterns.