Awesome DeepSeek Harness
> 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 #dsh GitHub topic so they can be discovered. Add it to your repo, then open a PR here.
🆕 Recently added (last 7 days)
auto-generated by .scripts/dsh_recent_added.py
- ccch1mneyyy/dsh-TUI — Claude-Code-style full-screen terminal UI for DSH: pixel-whale top bar, live work-status line, streaming thoughts, double-Esc rollback, context progress bar + TPS gauge.
⭐3689· Team pick 2026-09-26 · post - omdsh-dev/DSH-better-sidebar — Open sidebar foundation: third-party tab registration plus built-in file rendering/editing, terminal, side chat, Git, and sub-agent pages.
⭐3842· Team pick 2026-09-27 · post - JackZo400/dsh-voice-transcribe — Local speech-to-text for dsh: SILK decoding + SenseVoice (sherpa-onnx int8), no API cost, about 0.1 s per clip.
- leolee9086/SAC_search — Multi-engine meta-search plugin for DeepSeek Harness: roughly 200 API-key-free search engines, zero runtime dependencies.
- qujinting/dsh-web-search-clinepass — DSH plugin (DeepSeek Harness): a web_search provider that searches with the session's selected model via gateway-native search tools, replacing the built-in DeepSeek-only provider.
- 9087/dsh-diff-approval — Make every agent edit approvable in DeepSeek Harness: full-diff review, per-file / per-block keep & revert, undo/redo.
- JackZo400/dsh-memory-search — Long-term memory for dsh agents: hybrid search (local embeddings + SQLite FTS5) over your markdown notes, with optional secrecy-level filtering.
- jipika/dsh-memory — Two-layer long-term memory for DeepSeek Harness (DSH): global + per-project markdown, live re-read, zero extra LLM cost, with a Settings panel.
- lpf20200901/dsh-memory-delta — Cross-session long-term memory for AI coding assistants: automatic injection, changes only. DSH plugin + zero-dependency CLI.
- Kihara777/dsh-api-balance — API usage / balance plugin for DeepSeek Harness: a "usage / balance" tab switch inside the Web UI usage ring, showing account balance and usage detail.
- JackZo400/dsh-onebot — OneBot 11 client channel for dsh: your agent in QQ groups and DMs, zero dependencies (forward WebSocket + HTTP).
- JackZo400/dsh-sticker-library — Sticker library for dsh: dedup, vision tagging, pick by meaning; unsafe stickers rejected at ingest.
…and 391 more this week.
Quick Start```bash
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.
Contents
- Official Picks
- Official
- Profiles & Patch Layers
- Harnesses & Runtimes
- Security & Permissions
- Session & Memory Management
- Cost & Usage Tracking
- Channel / IM Bridges
- Plugin Marketplaces & Ecosystem
- Visualization
- Slides / PPT
- Documents & Export
- Coding
- Agents
- Loops (Auto-Research, Self-Improve, etc.)
- MCP Servers
- Tools & Integrations
- Orchestrators & Aggregators
- UI / Clients
- Skills
- Resources
- Contributing
Official Picks
Highlights personally recommended by Tianyi Cui (@tianyi), DeepSeek Harness team lead, in his 【DSH 社区插件推荐】 series on X. Newest first; each entry links the original post. Kept in sync with README.zh-CN.md.
- oh-my-dsh/dsh-plugin-upgrade-skill — Upgrade skills for DSH plugin authors: 195 versioned upgrade cards (0.1.0-rc.8 → 0.2.0-rc.1) and 11 skills that list breaking changes, patch the plugin, test and publish — so Claude Code / Codex / Gemini can migrate an old plugin to a new DSH release, scored by a 63-task public benchmark.
⭐287· Team pick 2026-10-05 · post - bowenliang123/dsh-context — Context insight and management suite: a context dashboard, browser, sidebar and
contextcommand that show what the model's window is made of — composition vs. window size, per-request history, compression and injection events, per-message token stats, and multi-agent views.⭐2304· Team pick 2026-09-30 · post
🔍 Plugin Directories (team pick #3 · 2026-09-28)
- dshfind · hikariming/dshfind — Learning site + plugin marketplace in one: a systematic course from the basics to Cordis papers, plus a marketplace auto-aggregated from the GitHub
dsh-plugintopic, S/A/B/C ratings, a dev guide, and an author leaderboard.⭐280 - dshmarket · dsh-market/dsh-market — The only marketplace that installs into the DSH settings page: browse / search / one-click install / theme switch across 4000+ plugins, most take effect on refresh. Install:
dsh plugin --profile web add dshmarket⭐4837 - awesome-dsh-plugin · awesome-dsh-plugin/awesome-dsh-plugin — Community-curated directory: a repo must be installable via
dsh plugin addto be listed; bilingual categories, continuously human-verified.⭐17192 - In their words: “今天提名几个广泛使用、内容丰富、更新及时的 DSH 插件导航站吧~ dshfind.com / dshmarket.com / awesome-dsh-plugin.com。其中 dshmarket 也可直接作为 DSH 插件安装到设置页中。(不代表公司立场,不对非官方网站的内容负责。)”
-
Starter path: install dshmarket → filter for quality with awesome-dsh-plugin → learn the internals and read ratings on dshfind. post
-
ccch1mneyyy/dsh-TUI — Claude-Code-style full-screen terminal UI for DSH: pixel-whale top bar, live work-status line, streaming thoughts, double-Esc rollback, context progress bar + TPS gauge.
⭐3689· Team pick 2026-09-26 · post - In their words: “推荐一下从 DeepSeek Harness 内测期间就持续开发的 dsh-TUI。做得非常用心,补齐了 DSH 缺失的 TUI 界面,并持续随 DSH 版本更新及打磨功能。”
- Install:
dsh plugin --profile web add "github:ccch1mneyyy/dsh-TUI#main" - omdsh-dev/DSH-better-sidebar — Open sidebar foundation: third-party tab registration plus built-in file rendering/editing, terminal, side chat, Git, and sub-agent pages.
⭐3842· Team pick 2026-09-27 · post - In their words: “dsh-better-sidebar 插件为 DeepSeek Harness 添加了侧边栏、底边栏、分栏、可浮动栏等 UI 定制化能力。属于『为其它插件提供基础能力』的底座类插件,体现了 Cordis 插件系统中可组合性的特点。”
- Install:
dsh plugin --profile web add "github:omdsh-dev/DSH-better-sidebar#main"
Official
- 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.
-
Ndsanes/dsh-rules — OMP rules port for DeepSeek Harness: rulebook, always-apply injection, rule:// addressing, and time-traveling stream rules that interrupt violating model output.
-
bigunmd/gold — Adaptive AI engineering presets for DeepSeek Harness — architecture, development, QA, and DevOps with shared guidelines and configurable collaboration.
- qianai05233/dsh-codex-preset — Codex-mode agent preset for DSH: explore → plan → implement → verify → integrate.
- yue-xiaxing/dsh-chang-kaishen-mode — "Chang Kaishen mode": a purely-for-fun DSH persona toggle found under Settings → General; both the plugin and its release flow were produced entirely by AI.
- xuanke66/dsh-thinking-slider — DSH plugin that turns DeepSeek's reasoning levels (off/low/high/max) into a single slider.
Profiles & Patch Layers
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.
- Bryan-cmf/dsh-general-agentics — General Agentics distribution bundle for DSH: one install brings a white-label brand, curated plugins, bundled skills and MCP pre-configuration (MIT).
- lldwb/dsh-sub2api-provider — Sub2API third-party provider bundle for DeepSeek Harness, with Off / Low / High / Max thinking effort.
- cuddly-guacamole/dsh-orchestrator-preset — Thin-shell agent preset for DeepSeek Harness: nine lanes with declared tool boundaries, resident routing sections, four self-authored skills, and an evidence-based completion gate.
- AKI2253/Sidor_Character — SIDOR persona card: a DeepSeek Harness plugin that batch-imports persona cards (.persona.md / SillyTavern character cards), switches them freely, applies them live in-session, and persists the agent preset.
- LMQ00/dsh-spec-mode — Spec mode for DeepSeek Harness: interview-driven requirements capture with a read-only working tree.
- Cyrene-xl/dsh-cyrene-chat — DSH agent preset for a plain-text dialogue mode: swaps in a full replacement system prompt for in-character chat, keeping only web search and meme stickers (persona text adapted from the open-source Cyrene-Agent, MIT).
- ADkun/adg-multi-agent — DSH agent preset: one dispatcher agent routes tasks to nine capability-scoped specialist agents, with a bundled CDP browser toolchain and an add-agent skill.
- asdf17128/dshp — Manage DeepSeek Harness profiles: list, create, clone, diff, and share a whole
dshsetup (plugin versions + bundle order + patch) as one portable file. - geohotstan/dsh-tabula-rasa-preset — Blank slate preset for DeepSeek Harness. Create your own adventure.
- ggfgfgf-on/dsh-trajectory-anchor — Self-contained DSH bundle: first-round trajectory anchoring, EWMA trajectory scoring, drift rollback, anchorGate promotion, bootstrap context suppression, and Layer-4 training-data export.
- GBDJXB/dsh-workspace-presets — Per-workspace Agent preset bindings for DeepSeek Harness.
-
AMAP-ML/LongHorizon-Harness — Long-horizon computer-use harness with a DSH adapter: runs
dsh --profile headlessunder an isolatedDSH_HOMEwith role-scoped patches (workspace-writefor executors,read-onlyfor 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 provider and Web configuration plugin for DeepSeek Harness.
- raktim-mondol/dsh-researchcraft — ResearchCraft as a DeepSeek Harness (DSH) profile plugin: research persona, scientific skills catalogue, living lab notebook, and specialist subagents.
- WoyouWoyou/dsh-office-32k — DSH (DeepSeek Harness) office-document · 32K-mode Agent Preset: 32K context management + offline dependencies + Excel/Word/PPT skills.
- runfali/dsh-prompt-injector — Per-turn context-injection plugin for DeepSeek Harness: manage an enabled-prompts list from a settings page, and inject each enabled prompt as a "context injection" reminder line into every turn's model context — makes discipline rules (e.g. graph-consumption / wiki-first / memory-recall) reliably take effect.
- zhongjie10086/dsh-adaptive-native — Windows-native Adaptive preset for DeepSeek Harness.
- masknull/dsh-session-prompt — DSH plugin: injects a custom prompt at the very top of every session's system prompt, live-editable from the Web settings page.
- huanghai-lab/dsh-custom-instructions — DSH Web security custom-instructions manager: edit the global AGENTS.md with template support, Markdown preview, history restore, import/export, and concurrency protection.
- cloga/dsh-github-copilot — First-class GitHub Copilot model integration for DeepSeek Harness, with model discovery, reasoning, vision, and hosted search.
- ptonlix/dsh-forge — Building an auditable desktop distribution around DeepSeek Harness (DSH).
- tianyuegithub/dsh-pactflow — DSH PactFlow (zero-pulse mode) external Profile Bundle.
- zachshi-ai/newmind — Old-wisdom × new-intelligence lab: each classical school of thought solves one real AI problem. #1 zhizhi (知止) — a behavior-moderation layer for DeepSeek Harness inspired by the Tao Te Ching, aimed at agent reliability.
- chr003/dsh-subagent-model-picker — Separate per-session model and reasoning effort controls for DeepSeek Harness subagents.
- clearnature/dsh-math-proof — Mathematical-proof agent preset for DeepSeek Harness using the Agda kernel as the proof checker, with compute-then-verify workflows and a long-horizon ledger of construction methods, evidence levels and refuted claims.
- ABccgh/dsh-smith — DeepSeek Harness agent preset for building harness agents and Cordis plugins: a composition-design thinking protocol, a four-layer memory discipline, and a standing team of four named experts.
- PerryLink/dsh-autotier — Automatic strong/cheap model-tier routing for DeepSeek Harness: intent-gated tier landing on the agent/request waterfall, plan-mode handoff (strong plans, cheap implements), deterministic high-risk guards on tools/pre-execute, failure escalation with TTL fallback, a /tier command, and tier status tools.
- Robin1987China/dsh-plugin-preset-default-guard — Repairs a stale agent-presets.default so New Session stops failing silently after a preset rename (DeepSeek Harness plugin).
- jarvislee90s-dot/dsh-foxbell-pet — Foxbell: a draggable fox desktop pet in the DSH Web bottom-right corner, with multi-project status monitoring, completion voice alerts, and a show/hide toggle.
- gdrpzym/dsh-loomy-connect — Use the models that ship with the Loomy desktop app inside DeepSeek Harness — no API key; also adds a Settings card showing the signed-in account and Loomy credits (permanent + daily gift).
- persevere-dsh/perse-updater — perse-updater: carry DSH to its next version safely, with rollback (part of the "Persevere with DSH" project).
- 67-68/dsh-plugins — A dotfiles-style paradigm for a whole DSH configuration: a git repo is the source of truth and one idempotent
install.shsyncs presets, plugins and experience docs to~/.dsh, with arequirements.txtversion lock. - n0pe-sled/system-prompt-editor — dsh plugin: edit the assembled system prompt of every new session (custom text, persona, tool guidance) with live preview.
- my-dsh-plugin/agent-mode-switcher — Switch the current session's agent preset (mode) after the model answers, and keep chatting in the new mode.
- winston-hoo/dsh-spec-forge — Forge vague coding requirements into executable specs before any work starts, and distill each finished task into a reusable personal prompt-template library (plain Markdown under
~/.dsh/spec-forge). - Haven-hvn/dsh-persona — Versioned system-prompt persona bundle for DSH agents; the model resolves wallet address and balances live via wallet_info instead of hard-coded values.
- bychv/dsh-preset-enhance — SillyTavern-style preset mode for DeepSeek Harness: a macro engine and a preset editor.
- GitRuozhi/dsh-mini-win — Windows-friendly minimal agent preset for DeepSeek Harness: sandboxed pwsh, native cmd.exe, and str_replace_editor.
- ciskonc/dsh-context-imports — Claude Code-style @path imports for DeepSeek Harness: expands @imports in AGENTS.md/CLAUDE.md and injects the referenced files into context at session start.
- Zomcxj/ModelHarbor — Desktop GUI (Rust + egui) for visually editing agent and provider configuration: opencode, pi, oh-my-pi (omp) and DeepSeek Harness config files, with cross-format conversion and automatic
.bakbackups. - john-walks-slow/dsh-simulated-life — Simulated life context and world evolution for DeepSeek Harness: injects the past 24h of simulated events and provides a
life_reacttool. - BOWLUNA/dsh-custom-mode — A custom mode for DeepSeek Harness (dsh): an editable system prompt that takes effect on the next step, per-row plugin switches, and a settings page.
- z91772524-ai/pojia-next — Five desktop AI-client persona-patch tools merged into one zero-dependency Python script: one persona, three targets (DeepSeek Harness / Codex CLI / WorkBuddy), with mandatory backup before every change and one-step restore.
- lifecoder1988/dsh-agents-md-manager — AGENTS.md management for the DeepSeek Harness: every instruction level a session loads in precedence order (user-global first, then project root down to cwd), editable in place from a Settings page or from the model.
- Yu1Ko/dsh-reasoning-support — DSV4.1 reasoning support for DeepSeek Harness: lean analysis, images and attachments, and bounded acceptance/repair loops that aim to retain minimal-mode reasoning inside a full agent.
- shenA2024/whale-persona — Multi-host persona engine: one config.json lets DSH and ZCode share the same persona, work contract, chain-of-thought language and long-term memory; memory changes must be proposed by the AI and confirmed by a human (enforced in code).
- wwwangzilin/dsh-character-presets — DSH character preset pack: 10 characters sharing one six-layer emotion engine — Luna / Xiao Miao / Feiyin / Rin / Mei / Baiye / Alma / Xiaoling / Michiyo / Hui.
- jonah791/dsh-agent-toolface — Tool-surface tiering for DSH agent presets: narrow the tools a model can see inside an agent scope using an evidence-backed deny set, with one-click lean/full switching and an audit trail — cuts the fixed context cost of tool schemas and reduces tool-selection dilution.
…(正文过长,页面只展示前面部分,完整内容见仓库)
