Agent collaboration · DeepSeek Harness plugins — Page 291

RAG-style skill router for DeepSeek Harness: embeds skills, matches user tasks by similarity, and auto-injects the relevant skill bodies.

DeepSeek Harness plugin adding the skill_search / skill_load / skill_ref tools and a 技能 tab that lists what the session actually loaded, so an agent can find and load any skill from a staged library instead of paying for every skill in the session catalog.

DSH skill 使用记分板：同时记录按会话去重与每次加载两种次数；设置页三选项卡（Skill 排行 / 会话榜 / 管理），Skill 排行固定按会话去重降序并分页（两种次数同时显示），会话榜按去重 skill 数排列并显示持久化会话标题；skill 工具与 read 直接读 skill 文件均记分；agent/pre-step 注入记分榜与 skill 实际路径；导入导出记分 JSON。

Community skill marketplace for DeepSeek Harness: browse, search, and one-click install skills from awesome-skills-cn and the wider Claude Code / OpenClaw skill ecosystem.

Skill studio for DeepSeek Harness: visualize all agent skills, view/edit their SKILL.md bodies, and enable or disable model/user invocation — from the web settings panel and via skillmgr_* tools

Dynamic skill runtime manager for DeepSeek Harness

DSH manager for user Skills already on disk in Agent home and DSH home

DSH skill manager — list, inspect, enable/disable skills on disk and delegate creation/editing to the host agent. · DSH 技能管理器：列出/查看/启用禁用本机技能，创建与编辑交由 agent 完成。

Lock, verify and govern AI agent skills: a DSH plugin that makes the agent check what a skill can do before installing or updating it, and notice when it changes.

跨平台 AI Skills 统一管理器：以 ~/.agents/skills 为中心技能库，通过系统符号链接把同一份技能一键安装/卸载到 Claude Code、Cursor、Codex、Gemini CLI、Trae、Windsurf 等多个 Agent 工具，仅展示本机已安装的平台，统一管理本地技能目录与只读平台展示。支持 Windows / macOS / Linux。DeepSeek Harness plugin.

Skills management tab for the Web Plugins settings: manage agent skills at the official hierarchy levels — global (host layer, <dshHome>/skills + <agentsHome>/skills), preset (<preset>/skills), and agent (live-agent runtime layer) — through a same-origin /skills-manager JSON API.

DeepSeek Harness plugin: skill management settings page. Lists all available agent skills (name, description, source, provider, invocation policy), supports search/filtering, and shows the full SKILL.md body/path on demand.

DeepSeek Harness 皮肤大师：主题壁纸/自定义皮肤插件 —— 全局背景图片/视频、毛玻璃、输入框背景、弹出框毛玻璃、滤镜、AI 回复文本与消息气泡颜色

Cross-session work ledger for agent harnesses: event-sourced work items with lifecycle tracking, audit history, and session context injection. Ships as a DeepSeek Harness (dsh) bundle.

Sleev routing observability for DeepSeek Harness

Turn a structured outline into a presentation for DeepSeek Harness (dsh): a self-contained HTML deck that presents offline with speaker notes and prints to PDF, plus optional editable .pptx export. Five committed themes.

SLI/SLO/错误预算：定义可衡量的可靠性目标、错误预算告警、SRE 实践。受 wshobson/agents（38k★ MIT）启发。

Adds a model parameter to the subagent tool so the model picks the subagent's provider/model per call

Two DeepSeek Harness agent presets: `dsh-smith`, which builds harness agents and Cordis plugins, and `dsh-forge`, a software-delivery agent. Each carries a thinking protocol, a layered memory discipline, and a standing team of named experts.

SQL 优化模式：EXPLAIN 分析、索引策略、N+1 解决、查询改写。受 wshobson/agents（38k★ MIT）启发。

Sleep-consolidated memory for DeepSeek Harness: resident identity + intentions, action-time associative recall, episodic / semantic / prospective memory, and dream-time consolidation via a Python sidecar.

SOUL.md-backed living persona for DeepSeek Harness: injects a trusted SOUL.md as a system-prompt section refreshed at every session start, registers soul_view / soul_remember / soul_edit / soul_rewrite so the agent maintains its own cross-session persona and memory, and serves a SOUL.md editor card in the Web Plugins configuration section.

「灵魂系统」状态面插件（DSH / DeepSeek Harness）：14 个 soul_* 工具（记忆 / 目标 / 待办 / 反思 / 反馈 / 成本 / 日程）+ 每轮账本摘要注入 + 零 token 自维护定时器 + 心跳唤醒（复用已有会话，默认每天 ≤5 次）+ 用户状态信号层 + 右侧栏面板（概览 / 成本 / 记忆；展示为主，记忆可就地遗忘与撤销遗忘）。账本是明文 Markdown，可读、可版本化、可删除。桌面通知与日历 / 提醒事项桥接为 macOS 专属，其余平台核心功能可用。

Play a notification chime in the DSH Web UI when the agent finishes responding.