Plugins

Enrutamiento adaptativo de modelos: clasificación de complejidad por solicitud con enrutamiento automático de proveedores.

Hooks de ciclo de vida al estilo Claude Code: comandos de shell configurados que se ejecutan antes/después de las llamadas a herramientas del modelo, con un payload JSON por stdin; una salida distinta de cero en el pre-tool bloquea la llamada.

Herramientas de Git de primera clase para el modelo: status / diff / log / commit / branch con mensajes y rutas validados, sin llamadas directas a git por shell.

Anulaciones por llamada de modelo, proveedor, persona y toolFilter para la delegación en subagentes, con referencias @preset: e identificadores compuestos de proveedor/modelo.

Las imágenes adjuntas en el editor se transcriben a texto mediante un modelo de visión compatible con OpenAI antes de llegar a los modelos de DeepSeek de solo texto.

Memoria local acotada con control de autoridad CBDC: afirmaciones en SQLite + FTS5, recuperación con alcance y decisiones explicables de usar/verificar/ignorar, con registro de auditoría completo, comandos /memory, inyección de ≤3 afirmaciones/1200 caracteres por llamada, sin llamadas adicionales al modelo.

Use a PackyCode (PackyAPI) API key in DeepSeek Harness through a built-in multi-protocol model route with key rotation and balance display

DeepSeek Harness LLM adapter for Qoder CN subscriptions, built on the official @qodercn-ai/qodercn-agent-sdk. Reuses the local qoderclicn login, and exposes a fully configurable settings page (per-model context window, session pool, timeouts, model deny l

DeepSeek Harness approval guardian: a reviewer model decides permission requests before they reach you, with a System One decision-API backend (TypeSafe Jev, SiliconFlow, self-hosted Laya), a configurable low-confidence policy and an allow-rule whitelist.

Connects one existing DSH session to QQ through OneBot/NapCat without starting a new session, and uses a fail-closed whitelist plus @-mention, nickname and configurable-probability wake rules to decide which messages reach the model.

Read-only Git and SVN code review using the current DSH agent model, with change previews, explicit context and approval-gated model requests.

Dependency-ordered multi-agent team runner for DeepSeek Harness: define roles with their own model, persona and tools, run a node graph with an independent review gate and bounded rework, and watch progress in the web UI.

Reads the session event stream, with no extra model calls, and turns corrections, failures and fixes into pending lesson candidates. Every candidate is judged against written criteria and carries the reason it passed or was refused; only confirmation writes a skill, and only through the single validated writer `learn_skill_manage`. An idle curator marks unused skills stale and archives them by moving rather than deleting, and spares any skill that is still being loaded. Five tools: `learn`, `learn_review`, `learn_curator`, `learn_skill_manage`, `learn_skills`. Registers a prompt section stating the criteria, and a tool guard that refuses direct edits to skill files so the writer cannot be bypassed.

Reasoning-effort gear presets for third-party models: 16 vendor/model presets with per-mode custom wire values, written into dsh-llm-pi-ai with pre-write validation and per-rule rejection; per-model switching between OpenAI Chat Completions, OpenAI Responses and Anthropic Messages with automatic route regrouping; plus a composer Optimize button and a settings section that rewrite the input text with a model you pick.

Desktop-only token statistics, provider quotas, custom quota queries, optional WorkBuddy credits and an advanced model selector.

Infinite Canvas — a node-based short-drama studio plugin for DeepSeek Harness: script, multi-shot storyboard, reference-consistent assets, one-shot video generation and final cut. Local-first, bring your own models.

Save DSH conversation turns into Markdown: a heart in the assistant action row, per-session collections in the sidebar, and a personal knowledge base the model can search on demand.

Multi-engine web search for DeepSeek Harness: an aggregated ctx.web search provider (auto-routing + sequential fallback, free engines keep search alive) plus a web_search_pro model tool with per-query engine choice, fan-out, batch and verticals — Brave, B

Exposes the models already configured in DeepSeek Harness through an OpenAI-compatible gateway, so other local projects can call them with any standard OpenAI client. Multiple provider candidates share one named group with ordered, balanced or random dispatch; upstream failures before any output switch to the next candidate transparently, and rate-limited requests are retried after the provider's backoff. Per-candidate call statistics (attempts, answers, failures, token/TTFT totals) appear as badges in the settings page, and daily aggregates persist to a local file with a 30-day heatmap and trend view. Until you create a key in the UI, the endpoint accepts calls from anyone who can reach it; keys (stored as SHA-256 hashes) then turn authentication on, and the endpoint stays locked even after the last key is revoked. Exposing the optional LAN listener without a key leaves the port open to the whole network. Runs on DSH 0.1.2-rc.1 through 0.1.7-alpha.1 and later.

Codebase Memory MCP bridge plugin for DeepSeek Harness: knowledge-graph code search, snippets, architecture, call-path tracing, and auto-started graph visualization UI.

DSH 的第三方模型补齐思考等级：在官方 Models 页每个模型行注入 Thinking 开关和强度/模态编辑器，在输入框模型菜单加供应商→模型选择器、视觉徽章与等级行，并带搜索、拖拽排序、供应商启用开关；以及按路由接管请求头与 user-agent。

Drive a DSH instance from another device: a host plugin opens a token-authenticated TCP port that a single-file Python CLI drives - listing and creating sessions, sending prompts with images, streaming replies, and switching models, permissions and workspaces.

Music composition inside the conversation: the model writes a score against a bundled composing skill, and a play_score tool renders it as an inline card with a piano roll, Web Audio playback and WAV/JSON export.

Registers external databases and document corpora as callable model tools: a per-source access boundary, a credential reference ledger that never stores values, and a fetch ledger that keeps fingerprints and content hashes instead of payloads.