Plugins

Per-turn and per-step cost ledger for DeepSeek Harness: a Turns tab beside Chat and Trajectory with tokens, estimated cost, cache savings, and an inspector for tool arguments, raw results, and source. Client-only; prices are estimated from recorded model usage.

DeepSeek Harness activity as a GitHub-style contribution graph: an account-menu row opens a centred dialog with your avatar and name, a year of daily activity in five shades, a year picker and a tooltip on every day.

GitHub-style contribution grid in the Web sidebar showing how many prompts you sent each day, with a switch to count conversations, total tokens or output tokens instead. Counts only messages you sent yourself; injected context is excluded.

A customized usage-statistics-panel dashboard with a 52-week square heatmap, bar-only token trends, and ended-turn labels, using the upstream collector and history. Same package name as upstream HaoyueQin/dsh-usage-statistics-panel, so the two cannot be installed together.

Forensic analysis of past model-routing decisions. Replays each session, classifies every turn's task complexity from its request shape (deterministic, no LLM), and reports two kinds of waste: an expensive model on cheap work (the difference is measurable), and a cheap model on expensive work that failed on its own merits — where every retry re-sends the whole context, so the downgrade cost more than the upgrade. Distinguishes provider-capacity retries (RATE_LIMIT/TIMEOUT/TRANSPORT) from model failures so it does not cry wolf. Derives routing rules from the evidence, priced against a built-in per-million table.

Adds TXT attachments, session windows, batch archiving, plan comments, menu shortcuts, next-message suggestions and an optional Q theme. Computer use ships in 0.6.x, is on default and needs screen-recording and accessibility permission. Suggestions use the current model with extra usage.

Adds an Optimize prompt control next to the composer send button that rewrites the draft into a structured prompt (plain, sections, or role-task-goal) streamed from the session model, with Use, Undo, Copy and Close.

Composer UI for DeepSeek Harness that does two things. The reasoning-effort row in the model menu becomes a continuous slider drawn on the row itself, with the rungs read live from the model adapter so any ladder length works; each rung paints its own ember plume into a canvas behind the knob. Separately, the permission trigger renders in the theme error colour whenever full access is the active preset, including its row in the permission menu.

Model picker control for the DSH composer: a settings section beside the built-in Models page renames model display names and provider group names, adds a short coloured suffix capsule after a name, groups models from different providers into a new partition, reorders provider groups by drag or keyboard, sorts models inside a group by catalog order or name, and hides a provider or a single model without removing it from the catalog. Edits stay in a draft until you press Save, and the built-in 1017 tag and a custom capsule can be shown side by side.

Screenshot button for the DSH composer on Windows: drag a region over the frozen screen and the picture lands in the draft as a normal image attachment, with a global hotkey and an editable shortcut.

Four-tier prompt enhancer for the composer — the system tier always emits a LangGPT-structured system prompt (# Role / Profile / Skills / Rules / Workflows / Initialization), upgrading task drafts into the system prompt of the agent that runs them, while the other tiers rewrite as a light tidy-up, a single paragraph, or a five-section brief; the iterate workbench refines an editable base under an optional requirement, with fill-back and undo. Never changes intent, never invents requirements.

Download pets from petdex.dev and keep one on the desktop, following the DSH session status; the appearance is adjustable and an enterprise WeChat bot can push a notification when the agent needs confirmation or finishes a turn. Desktop mode needs macOS 12+ or Windows 10 1903+; the bundled companion is prebuilt.

Handwriting and diagram whiteboard: sketch on a floating or sidebar canvas and attach the cropped image to your message.

Diagnosis-first learning runtime for DeepSeek Harness — a live learning map, structured lessons, and persistent learner state, driven by the Universal Diagnostic Tutor skill.

ChatGPT-style composer predictions for DeepSeek Harness: after each assistant turn, suggest the user's next message as ghost text in the composer. Tab accepts, Escape dismisses.

Read-only page structure for DSH, attached to the CUA browser: table coordinates, shadow DOM fields, code blocks, and confidence labels.

rulemux rules inside DeepSeek Harness (dsh): asks the rulemux CLI for the rules it renders for this workspace and injects them into the session once, on the first turn (nothing written to disk for rules). Also syncs skills into the native skills dirs (.ds

装第三方插件之前，先看它会动什么

Sidebar tab that lists a session's file changes by turn, expands line-level diffs and undoes them, with a matching review card at the end of each chat turn. Needs dsh-better-sidebar installed separately. DSH 0.2.x port of Lzh3070/dsh-file-review-tab; do not install both.

Webpage watch agent: keep a watchlist of URLs, snapshot each page on every sweep, detect changes by bigram-Dice ratio and keywords, log important ones and push to a WeChat webhook; agent tools page_watch_add/list/check/history/remove. Renders JS anti-bot WAF pages via the optional lexmount extractor.

Full-screen screenshot panel for the DSH composer: marquee selection with rectangle, ellipse, arrow, pen, mosaic and text annotation, then insert into the conversation, copy or save. The selected region can also be recognized offline — PP-OCRv6 through onnxruntime-node, falling back to the built-in Windows OCR engine — and translated into nine languages with the model already configured in DSH. On Windows the marquee covers the whole screen including the taskbar and hides the DSH window first, so the capture is of the real desktop. The panel runs in an Edge or Chrome kiosk window, and falls back to an in-DSH capture flow when no browser can be launched. Note that DSH Desktop denies ordinary browser access by default, so the full-screen panel also needs "Allow opening this Profile in a browser" (compatibility mode) enabled in the profile settings; without it every capture goes through the in-DSH fallback. Windows only (DSH Desktop); the PP-OCRv6 models are downloaded from ModelScope on first recognition.

Annotation browser in the DSH sidebar: a real Chrome page streamed over CDP with input forwarded, where picking an element or selecting text attaches its source location, component stack, styles and a crop to the chat input. Needs a local Chrome (on Windows set chromePath).

Zero-key web search provider for the built-in web_search tool: scrapes Bing China, 360 and SearXNG through a backend fallback chain and returns real URLs with snippets and publish dates. No API key and no settings card are required. SearXNG is used only when you set your own instance URL, and installing this plugin switches web_search to it and turns off the official DeepSeek search row.

Hosted PrimeSync AI Studio strategy-research MCP configuration bundle using the official MCP client and mcp-remote OAuth bridge; eligible Studio access required. Experimental: OAuth consent and tool calls are not yet verified end-to-end. Complete manual CLI sign-in on the same computer before starting Harness; otherwise tools may be absent after startup. Targets DSH 0.2.1-rc.2 only; mcp-remote 0.15.2 is downloaded through npx on first launch.