Plugins

Appends ambiguity-handling rules to the dsh system prompt so the agent asks clarifying options before guessing.

Fruit price tracking skill: collect and query global wholesale fruit prices with RMB/kg conversion.

Global news monitoring skill: track world events and analyze impact across industries, currencies and commodities.

Product research skill: analyze market, technology and supply chain, then output three-tier product spec proposals.

Fridge manager skill: recognize ingredient photos and recommend recipes, with inventory and shopping reminders.

Tracking assistant skill: monitor product prices, celebrity news and custom metrics with change alerts and scheduled pushes.

Fishing assistant skill: spot analysis, fish activity and bait/lure recommendations from water photos and weather data.

Persona wardrobe for DeepSeek Harness: save multiple personas, apply them globally or per workspace, with a master switch, an adjustable API balance guard, and a lazy token-loving tired mode that speaks on the agent's behalf when quota runs out. Files-as-config, hot reload, zero dependencies.

Minecraft-survival game mode as a DSH agent preset: hard-settled HP/hunger/day-night/mobs, vanilla crafting gates and anvil repair, plus a browser status bar; built on the official preset plugin spec.

失败账本：DeepSeek Harness 工具失败自动记录·签名聚类·跨会话持久化 + 失败感知重试拦截，让 Agent 越用越少犯错，含 Web UI 面板

During long agent thinking, posts a hot-news or reminder message directly into the conversation stream.

Static HTML auditor for Chinese line-breaking: checks explicit-`<br>` orphan lines, keep-all CSS coverage, CJK line-start/line-end punctuation rules, and protected-phrase splits, then inserts <wbr> at natural breakpoints; exposes a cn_linebreak_audit tool for agents.

macOS desktop control for DeepSeek Harness: agent operates non-browser apps (CapCut, Photoshop, WPS, native clients) like a human — 14 tools for mouse, keyboard, scrolling, app activation, windows and screenshots, plus a vision closed-loop (desktop_see for image understanding + desktop_locate for UI element coordinates) that reuses dsh existing vision credentials. GUI-session service starts on demand, no login autostart.

DeepSeek Harness (dsh) approval answerer backed by the agent-governance sidecar — every approval/request is decided by the deny-by-default gate, L0–L4 rules and the organizational audit chain.

Replay real DeepSeek Harness turns against Standard, Minimal, Anchored, or plugin candidates with frozen request-surface evidence

Multiple chat panels for DeepSeek Harness: view and interact with several agent sessions side by side in a Mission Control grid

Adds a sentient floating eye to DeepSeek Harness — injects a taming kernel that makes the AI truly understand you, with 11 dynamic eye skins, ghost mode, and zero token cost when AI mode is off.

Adaptive Reversible Context (ARC) for DeepSeek Harness — provider-aware, cache-conscious context governance with reversible recovery.

Permission mode between workspace-write and full access: shell commands are checked by deterministic rules and a subagent review; irreversible or system-level actions require explicit approval.

Exposes dsh's internal capabilities (language, workspace registry, agent-status and approval events) over HTTP under /dsh-api on the local loopback socket dsh already listens on, with SSE fan-out at /dsh-api/events.

Human-in-the-loop tools (`human_task` / `human_task_ready_check`): pause the agent until the user performs a real-world action or returns an observation (GUI operation, game testing, visual verification, hardware checks), gated by session consent and an AFK presence check, returning a structured JSON result via a web task dialog.

DeepSeek Harness memory and knowledge hub that routes update, retrieve, and subscribe calls across compatible plugins, with an optional transactional Rust backend.

Ranked subagent delegation: role-based model routing with failover, per-route reasoning effort, model-transparent labels, fork delegation, and a visual route editor.

Tiered model routing for DeepSeek Harness: think/build model tiers, vision delegation (the vision model only describes images; the working model acts), image generation, and a high-impact guard.