v4.0.0 · Apache-2.0
A 267-skill research graph in pure markdown — 51 research operations built from 216 single-purpose steps, composed in any order with explicit backtracking. One npx install, no runtime, no MCP bindings. The AI is the researcher; you set the direction.
v0.3.0-rc.1 · GPL-3.0
DeepSeek Harness 长篇角色扮演agent(DSH酒馆插件):SillyTavern 角色卡导入、行动选项卡、分支对话管理、长篇记忆、关键词与语义混合检索、交互式角色卡创作、世界书、多角色 Agent 集群、文风预设、小说与角色卡导出、一键安装。 / Roleplay Agent 、Tavern plugin for DSH (DeepSeek Harness): branch conversation, long-term memory, hybrid keyword and semantic retrieval, SillyTavern cards, interactive card authoring, card export, one-click installers.
v0.1.26 · MIT
DSH plugin for semantic AST code skeletonization and test/build log compression
v0.3.7 · MIT
AI4Scholar for DeepSeek Harness (dsh): 38 native academic tools — Semantic Scholar, PubMed, Google Scholar, arXiv, bioRxiv/medRxiv, DOI resolution, PDF full text, auto-cite, scientific figures, credit balance — with an API-key card in the Web UI settings.
v0.1.0 · MIT
Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
v0.5.21 · MIT
Composable, view-based memory for DeepSeek Harness. Pluggable sources and strategies, with three-tier memory out of the box.
v3.50.0 · MIT
🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
v0.1.44 · MIT
DeepSeek Harness 的表情包插件——找得到、发得出、学得会,纯文本斗图、情绪主动发图、像 QQ/微信 一样发图、AI 自动学图、自定义表情包 、多种风格随意切换。
v1.1.0 · GPL-3.0
AI 小说创作软件:把灵感、角色、世界观、大纲、章节写作、审稿和修稿组织成可控流程;提供 Windows/macOS 桌面版,支持本地和在线模型。AI Novel Writing Software: Organizes inspirations, characters, worldbuilding, outlines, chapter drafting, review, and revision into a controllable workflow. Features desktop apps for Windows/macOS, Ollama integration, and a DeepSeek Harness (DSH) plugin preview.
v1.3.1 · MIT
Persistent project intelligence and memory plugin for DSH: architecture analysis, cross-session context, TODOs, and optional hybrid retrieval
v1.0.0 · no license
A dsh plugin that periodically scans reddapi.dev for new Reddit leads matching your one-sentence ICP, dedupes what you've already seen, and writes a dated markdown report. Also forwards 6 read-only Reddit search/lookup MCP tools.
— · MIT
A persistent, unified memory layer for all your AI agents (e.g. Claude Code, Codex, DSH), backed by Markdown and Milvus.
— · Apache-2.0
Open-source infrastructure that turns scattered SKILL.md files into curated, retrieval-ready agent-skill corpora—with retrieval and evaluation tooling included.
— · MIT
🧠 The memory that dreams — cross-session memory for DeepSeek Harness. Offline & private, auto-consolidates in its sleep (autoDream), visualized in a memory panel.
— · no license
Structural memory for AI coding agents. Bi-temporal graph, MCP-native, zero LLM calls. Cursor · Claude Code · Codex · DeepSeek Harness · Hermes · VS Code · Windsurf.
v0.29.0 · MIT
Cross-session memory plugin for DeepSeek Harness: seven-layer SQLite store (soul/user/project/fact/lesson/topic/rules), BM25 retrieval, per-window dream consolidation. 跨会话七层长期记忆插件。
v0.2.94 · MIT
Recent memories stay detailed. Older ones grow concise. Important ones live on.近期记忆保持详细,旧记忆逐渐精简,重要记忆长期存在。
v5.0.2 · MIT
DSH permission-gate for the DeepSeek Harness 0.1.5 line: a single self-sufficient, deterministic-first, fail-closed gate covering P0 hard-deny -> P1 session grant -> P2 static rule (allow/deny) chain -> P3 optional LLM semantic classifier -> P4 ask, with
v0.2.16 · Apache-2.0
Local document knowledge base for DeepSeek Harness: library_add/remove/list, hybrid semantic+keyword library_search with diversity re-ranking, relevance filtering and lost-in-the-middle avoidance, citation-aware injection, library_cite_check and library_d
v0.40.0 · MIT
Stable, editable PPTX generation for AI agents — semantic IR in, native DrawingML out
v0.25.4 · MIT
Evidence-backed UI delivery for coding agents working on existing Web products: declared criteria, criterion-bound artifacts, findings linked to declarations, and repair guidance. Evaluator review does not replace human semantic approval.
v0.8.7 · MIT
Requires dsh >= 0.1.7 (session format 4); on dsh 0.1.5 use dsh-repeat-tool-breaker@0.7. Escalating repeat detection for dsh: a synchronous ctx.tools.guard gate that counts semantic fingerprints (exact/cmd/net/host/sink) over a per-agent sliding window, wa