The project's README from github.com/Dominic789654/awesome-deepseek-harness. As one markdown file: README.md.
A curated list of plugins, skills, MCP servers, patch/profile layers, orchestrators, aggregators & UIs for DeepSeek Harness (DSH) — DeepSeek's official agent runtime built around the idea
Model + Harness = Agent.
English | 简体中文
DeepSeek Harness ("DSH") is DeepSeek's agent runtime / harness layer — the "hands" that turn the model's reasoning into real actions (context management, tool-call orchestration, execution sandbox, feedback loop, session persistence). Its defining feature is an open plugin ecosystem: the community contributes plugins, skills, MCP servers, orchestrators, aggregators, and UIs.
This list collects the best of that ecosystem. Contributions welcome — see Contributing.
Tip for authors: DeepSeek asks plugin repositories to carry the
#dshGitHub topic so they can be discovered. Add it to your repo, then open a PR here.
# Launch the DSH Web UI
npx @deepseek-ai/dsh web
# Install a community plugin (from this list) into your profile
dsh plugin --profile web add "github:owner/repo#main"
Before installing, confirm the target repo carries the #dsh GitHub topic so the community hub can index it.
deepseek-ai/deepseek-harness — DeepSeek's official agent runtime framework (Model + Harness = Agent); an "everything is a plugin" architecture built on Cordis (TypeScript, MIT). ⭐38238
deepseek-ai/awesome-deepseek-integration — Official curated list of DeepSeek API integrations. ⭐38654
deepseek-ai/awesome-deepseek-agent — Official list of agents/harnesses with DeepSeek support. ⭐5426
Temoa/dsh-rules-paths — Claude Code-style paths: rule injection for DeepSeek Harness (DSH).
mycodesite/dsh-rules — DeepSeek Harness (dsh) plugin: global + project-level two-tier Markdown rule injection (RuleBase).
Ruszero01/dsh-tang-governance — Three Departments and Six Ministries governance mode plugin for DeepSeek Harness | dsh 三省六部模式插件.
DSH's core composition mechanism: a profile stacks bundle patch layers, then your own cordis.patch.yml (profile-level, then $DSH_HOME-level, then --patch overlays) — letting you reshape the whole plugin tree without forking. This is the layer where task-specialized runtime configurations live: a long-horizon profile, a math-reasoning profile, a slides-editing profile are all just a different bundle stack + patch, not a different codebase. Tools and harnesses that operate at this layer (share/export a profile, or run DSH as a specialized backend under a task-specific patch) belong here rather than under generic plugins.
asdf17128/dshp — Manage DeepSeek Harness profiles: list, create, clone, diff, and share a whole dsh setup (plugin versions + bundle order + patch) as one portable file.
geohotstan/dsh-tabula-rasa-preset — Blank slate preset for DeepSeek Harness. Create your own adventure.
AMAP-ML/LongHorizon-Harness — Long-horizon computer-use harness with a DSH adapter: runs dsh --profile headless under an isolated DSH_HOME with role-scoped patches (workspace-write for executors, read-only for Manager/auditors) — a concrete example of a task-specialized DSH profile.
duyanta123/dsh-preset-scaffold — DSH agent preset: scaffold a standardized, runnable, verifiable project skeleton from scratch (architect persona + six template assets + a strict bootstrap flow).
light051001/dsh-preset-qa-mode — Ask-first DSH agent preset based on Standard: exhaustive structured clarifying questions (nine dimensions, max 5 rounds, interruptible) with a confirmation gate before any execution; plan-first for complex tasks.
Jungod1121/dsh-anchored-standard — Two-phase DSH preset: a minimal-aligned bootstrap (bash + read), then full Standard tools after the first tool call or reply.
lovstudio/dsh-llm-config — Reusable LLM configuration profile library exposed to browser consumers as a Remote (DeepSeek Harness plugin).
ZRui-C/dsh-minimal-first-turn — Installable Web bundle for a Minimal-compatible root-session first request, then restores the selected preset after the first tool call or reply; includes a persistent composer toggle.
songoao25/virtual-product-team — Product Team Mode — a DSH agent preset: boss-style conversation with a virtual product team (PM → Engineer → QA → Release) from idea to shipped product.
higekibaka/dsh-ciel — 夏尔 Ciel — a pre-planning advisor and convergent critic for DeepSeek Harness: a second, knowledge-rich model offering directions, prior art, pitfalls and verification checklists (ideas, never steps).
qwe225380/dsh-omni-router — Omni Router: an intelligent orchestration preset that routes tasks by complexity and thinking mode (spec/react/balanced), auto-collects project context, and adds Plan Mode, TDD, delivery-gate, Git workflow, and acceptance-checklist guidance. Supports dsh plugin add dsh-omni-router.
AythyaCrispus/dsh-minimal-msys2 — Windows Minimal Mode: persistent bash + str_replace_editor plugin — registers an agent preset, provides a working persistent-bash backend on Windows, and exposes a GUI-editable bash path in the plugin-settings section (persisted via the credentials domain).
CeilCelia/dsh-eli-mode — Eli Mode: an agent preset for DeepSeek Harness built around wiki-driven long-term memory and skills, on an extremely minimal Harness setup.
LiFenrir/dsh-scenario — Scenario-management plugin: bundle persona + model + permissions into named scenarios (dev / wiki / personal), one-click hot-switch from the settings page.
Saikel-Orado-Liu/dsh-coding-preset — Windows-adapted DSH coding-agent preset with persistent PowerShell 7 (pwsh) and str_replace_editor, mirroring the official minimal preset.
Scorp1o117/dsh-soul-md — Soul.md persona for DeepSeek Harness: a persona-card plugin (人设卡) that gives your agent a persistent character.
delightedMaster/dsh-anchored-standard-windows — Windows Anchored Standard agent preset for DeepSeek Harness with on-demand tools and Skills.
delightedMaster/dsh-subprocess-win32 — Windows subprocess Cordis runtime and Minimal/Anchored Standard presets for DeepSeek Harness.
brunhildzhou/dsh-all-warmup — Global frictionless warm-up layer plugin for DeepSeek Harness: the first turn of any session auto-warms up, full mode resumes from the second turn on.
jiatong-lab914/obsidian-knowledge-mode — DSH agent preset for a portable, AI-native knowledge system on Obsidian: anti-hoarding Layer 0, 80/20 distillation into Context/Claim, verifiable update loops, four read-only role agents, a source gate hook, and a zero-content starter template.
alllllllllli/Living-Dream-DSH — Complete DSH desktop config framework: 8+ MCP servers (computer-use, browser, memory, OCR, vision, history, markitdown, os-copilot), free model channels (AMD Radeon Cloud, DeepSeek v4-flash/v4-pro), mobile remote via Tailscale, vision patches (GLM-4V-Flash → Ollama fallback), and a one-click installer (PowerShell + offline 120 MB SFX). MIT.
JingMox/learner-preset — First-principles learning assistant preset for DeepSeek Harness: knowledge components with decaying mastery, an analogy lifecycle, and a prediction ledger.
hackerFish/awesome-dsh-presets — Tested DeepSeek Harness presets & rules: official-derived + original combos, every preset passes structure and package-existence validation (Chinese-first).
orziz/odai — Profile-wide DSH governance and routing bundle with a separately installable session-scoped Agent preset; compatible with DSH 0.1.0-rc.6 and 0.1.0-rc.7.
JohnXu22786/hooks-adapter — Hooks-config compatibility layer for dsh: reads hooks from .claude/settings.json, .codex/hooks.json, opencode.json and maps their lifecycle events to dsh extension points, running shell/webhook/oracle/proxy handlers.
Asher-2000/dsh-expert-mode — DSH expert-mode agent preset: a chief-coordinator persona plus 11 domain-expert subagents.
Socialist-Sister/dsh-survival-mode — DeepSeek Harness Survival Mode: a playful agent preset mixing Minecraft survival rules with real coding work.
baihejiangnan/dsh-plugin-pack-web — DeepSeek Harness plugin pack: one-click Profile/Web replication (DSH Plugin Pack).
hili986/dshpack — Export a dsh scenario into an installable, shareable, auditable pack, then install that pack as a standard dsh profile.
Kaalia0912/dsh-whale-musume-persona — Global whale-girl persona plugin for DeepSeek Harness.
keeshakulbida948-tech/dsh-plugin-fun — "阿蓝" (A-Lan) persona plugin for DeepSeek Harness: a warm, meme-literate whale-girl companion persona that knows when to send emoji stickers (happy / venting / food talk / cheering up), one per turn max, never for serious topics; pairs with dsh-plugin-sticker.
xingyingyuzhui/dsh-agent-identity — Claw identity files for DeepSeek Harness: bakes SOUL / AGENTS persona files into the prompt.
xingyingyuzhui/dsh-claw-suite — Claw Agent governance suite for DeepSeek Harness: persona, permissions, gates, delegation, and memory.
KannaKuron/dsh-ptc-cordis-preset — A creative mode built on top of PTC: DSH preset that composes Code Mode tool orchestration with self-referential Cordis tools and preset authoring guidance, materialized as the 'ptc-cordis' user preset.
Feiyang1997/dsh-preset-minimal-pwsh — DeepSeek Harness Windows minimal-mode agent preset: two tools (pwsh + str_replace_editor), built-in plugins only.
ztlovelsw/dsh-model-profile — Manually or automatically configure model thinking effort, context window, and max output tokens.
NIU-001-LIU/dsh-win-minimal — Windows minimal agent preset for DeepSeek Harness: one-line fixed persona, three tools, no runtime context. Installs via dsh plugin add.
Freakz2z/dsh-catgirl-plugin — A token-efficient persona runtime for DeepSeek Harness.
zhy201810576/dsh-zh-reasoning — Makes DeepSeek Harness's reasoning and final answers default to Simplified Chinese.
chaserchan/dsh-plugin-global-prompt — A global-prompt textarea in Settings > General, injected into every conversation system prompt.
peterwangze/dsh-reasoning-level — Unified default reasoning level (thinking effort) for all models: per-model defaults with capability probing, live call statistics, one-command install via dsh plugin.
zhangdong456/dsh-prompt-presets — Prompt Presets to manage your library.
SLAPaper/dsh-self-checking-profile — A drop-in dsh web profile that adds the Self Checking sandbox mode to DeepSeek Harness (dsh).
lynsucceed/dsh-openclaw-persona — Reuse OpenClaw persona files (SOUL/IDENTITY/USER/MEMORY/TOOLS.md) as the DSH agent persona, with a Web GUI editor — edit the .md files in the sidebar and the change takes effect on the next request.
XMoon/dsh-profile-settings — Per-profile settings overlay for DeepSeek Harness: global settings.yaml plus profiles//settings.patch.yml, transparently layered under ctx.settings.
muzyLink/dsh-subagent-profile — Subagent dispatch profile plugin for DeepSeek Harness: save named combos of preset/model/reasoning-effort/tool-scope for one-click dispatch, with built-in cost guardrails.
WSL043/dsh-reasoning-slider — Native, model-aware reasoning effort slider for DeepSeek Harness.
BaronCyrus/dsh-harness-ally — DeepSeek Harness Ally mode: freely combine DSH, Claude Code, Codex, and every configured model while preserving native agent lifecycle and real-time execution visibility.
GTC2080/dsh-APEX_Plugin — APEX experimental preset for DeepSeek Harness: Minimal anchoring, Pro-dominant mode, and a vision-enabled Flash Max bounded-collaboration flow with host-level web acceptance.
MengYuil/dsh-ponytail — Lazy senior dev mode for DeepSeek Harness — ponytail port (always-on minimal-code ruleset, /ponytail-review/audit/debt/gain/help).
sch246/dsh-warm-minimal — DSH warm-minimal mode: writes a high-quality first-turn trajectory (we/let's style) at session creation, with real input starting from the second turn; AGENTS.md and the skill catalog are injected naturally.
ninipa/oh-my-dsh-slim — Specialist subagent delegation preset for DeepSeek Harness — orchestrator + 5 roles (oracle/designer/fixer/explorer/librarian) with per-role models, effort, tool filters, scoped MCP and background-first delegation. Port of oh-my-opencode-slim.
Kian-Oraish/dsh-prompt-enchant — Prompt-enhancer magic wand: an adaptive prompt-enhancement plugin for the DSH Web composer input box.
blackteaYES/dsh-thinking-levels-settings — Per-model thinking-level settings page for custom DSH llm-pi-ai providers (official dsh client-plugin form).
TOBYCAI/dsh-patch-reasoning — Idempotent DSH model-tuning patch after upgrades: widens DeepSeek reasoning to 7 tiers and defaults workflow/ralph models to deepseek-v4-flash.
Starlight-bananice/dsh-zhushou — DSH sidebar assistant plugin: manage/select assistant personas from the sidebar and chat in-session under the selected persona; leaving none selected keeps the native session.
Anna-la/dsh-submodel-change — Lets you pick which model a subagent invokes for DeepSeek Harness.
drscrewdriver/dsh-thinking-levels — DeepSeek Harness (dsh) thinking-levels plugin: adjusts reasoning-effort strength.
lixiaoshuang79/dsh-helm — Multi-node control plane and MCP gateway for DSH ChatGPT Helm connectors: hub + node agents, authenticated WS JSON-RPC, compact session summaries, queue/steer routing, fail-closed routing.
fsrmqi/dsh-promptkit — Prompt building & enhancement toolkit for DeepSeek Harness: Studio + QuickEnhancer.
Motuo24/dsh-thinking-slider — Thinking-effort adjustment plugin for DSH that mimics the snap-to-gear reasoning slider of advanced Codex models (e.g. GPT5.6 Terra), unlocking Low/Medium/etc. gears.
9527ccccccc/dsh-build-plan-mode — Build and read-only Plan modes for DeepSeek Harness sessions — a pure user-layer Cordis plugin, zero DSH core modification.
yongshuai0314/dsh-i-have-adhd — ADHD-friendly output shaping for DeepSeek Harness: one system-prompt section with adhd_on/adhd_off/adhd_status tools, persisted across restarts. Inspired by ayghri/i-have-adhd (MIT).
sakka6868/dsh-prompt-optimize-plugin — "Prompt optimize" button for the DeepSeek Harness composer: one click rewrites a vague prompt into something clear, specific, and actionable, IDE-smart-suggestion style.
svgop/dsh-rich-context — Agent instruction manager for DSH — edit and template the AGENTS.md files the harness actually reads (global + per-workspace).
niyongsheng/meww — Pixel-art AI desktop-pet plugin for DSH: a small group of Pokémon-style sprites living in the corner of your Agent UI, with chat and persistent per-pet memory.
hackernotfound/dsh-tacit — Learns what you leave unsaid in your prompts and steers the DeepSeek Harness agent for you.
chenmzh/dsh-codex-reasoning-router — DeepSeek Harness preset: Luna executes, tool-less Sol advises.
NOirBRight/dsh-llm-grok — Grok subscription OAuth p