dsh-agent-os-worker-protocol
by h565656445
DeepSeek Harness 个人适配声明(Personal Adaptation Notice) 本项目是 DeepSeek Harness 的 个人适配产物(personal adaptation) , 并非 DeepSeek Harness 官方文件(not an official DeepSeek Harness file)…
Personal Adaptation Notice for DeepSeek Harness — a personal adaptation, not an official DeepSeek Harness file.
安装
dsh plugin --profile web add github:h565656445/dsh-agent-os-worker-protocolGitHub 源码安装:首次需按提示配置 allowBuilds 构建授权后重试
安装与环境配置指引、插件开发教程见 DSH 中文社区文档 ↗
安装即在你的机器上以你的权限运行第三方代码——它可读写文件、使用凭据、访问网络,DSH 的工具审批不会为插件代码加沙箱。「检测到 manifest」仅代表发现 dsh.bundle / dsh.plugin 清单,不构成兼容性或安全审查;安装前请审阅源码,不熟悉的插件先在不含密钥的环境试用。
README
目录
- 用途 / What this is for
- Agent OS Worker Protocol v0.1 / Agent OS Worker 协议 v0.1
- Features / 功能
- What's inside / 目录结构
- Quick start / 快速开始
- DeepSeek Harness 衍生 / DSH Derivative
- License / 许可证
- 相关项目 / Related Projects
- Agent OS 内核 / Kernel
- Harness 基础设施 / Infrastructure
- 规格与文档 / Specs & Docs
- 适配器 / Adapters
- DSH 总集成 / Integration
DeepSeek Harness 个人适配声明(Personal Adaptation Notice)
本项目是 DeepSeek Harness 的个人适配产物(personal adaptation),并非 DeepSeek Harness 官方文件(not an official DeepSeek Harness file),随附功能、使用说明与个人产物(bundled with features, documentation, and personal artifacts),可与 DeepSeek Harness 搭配使用,也可独立使用。
This project is a personal adaptation for DeepSeek Harness, and is NOT an official DeepSeek Harness file, bundled with features, documentation, and personal artifacts. It can be used alongside DeepSeek Harness or standalone.
作者 / Author: h565656445
合作 / Collaboration: 如有项目可以一起合作,欢迎联系。微信:wohaishihenshuaide。If you have projects, let's collaborate. WeChat: wohaishihenshuaide.
用途 / What this is for
Agent OS 统一 Worker 协议:Agent/Skill/本机工具通过同一 Prepare/Accept/Inspect seam 接单,结果必须带可复算证据哈希。
Agent OS unified worker protocol: agents, skills and local tools take orders through one Prepare/Accept/Inspect seam; results must carry recomputable evidence hashes.
Agent OS Worker Protocol v0.1 / Agent OS Worker 协议 v0.1
Agent OS Worker 协议(AOS-002)用统一协议让 Agent、Skill 和本机工具接收受控工作包,并把结构化状态与可复算证据交回 Agent OS。协议不负责选择模型、自动创建 Agent、发布内容、修改核心规则或绕过 AOS 运行任务的人工门。
The Agent OS Worker Protocol (AOS-002) uses one unified protocol to let agents, skills, and local tools receive controlled work packages and hand structured state plus recomputable evidence back to Agent OS. The protocol does not choose models, auto-create agents, publish content, modify core rules, or bypass the human gate of AOS runtime tasks.
Features / 功能
三动作公开接口:
Prepare/Accept/Inspect,Prepare只接受state=running且 Ledger 完整的 AOS 运行任务。能力与权限适配器:适配器声明
agent/skill/local_tool类型、上下文槽位、允许根目录、字节与数量上限。最小上下文规则:请求只绑定能力声明中的槽位,Worker 输入不含源文件路径,只含作业内快照。
结构化结果与证据:结果必须通过 Schema 并至少提交一份证据,证据绑定内容模式与 SHA-256。
两段式 Worker Ledger:
prepared → succeeded|failed|blocked,事件绑定前序哈希、输入哈希与请求哈希。0.1→0.2 迁移兼容:旧链经
legacy_0.1严格 Schema 只读兼容,拒绝0.2 → 0.1降级。Three-action seam:
Prepare/Accept/Inspect;Prepareonly accepts AOS runtime tasks withstate=runningand an intact Ledger.Capability & permission adapters: adapters declare
agent/skill/local_tooltype, context slots, allowed roots, byte and count caps.Minimal-context rules: requests bind only declared slots; worker input carries no source paths, only in-job snapshots.
Structured results & evidence: results must pass the schema and submit at least one evidence bound to content mode and SHA-256.
Two-phase worker ledger:
prepared → succeeded|failed|blocked, each event binds previous hash, input hash, and request hash.0.1→0.2 migration: legacy chains are read-only compatible via strict
legacy_0.1schemas;0.2 → 0.1downgrades are rejected.
What's inside / 目录结构
dsh-agent-os-worker-protocol/
├── README.md # 双语说明(本文件)
├── LICENSE # MIT
├── src/AgentOSWorkerProtocol.psm1 # Worker 协议模块(准备、接收、检查)
├── runner/agent_os_worker.ps1 # 唯一公开入口(Prepare/Accept/Inspect)
├── schemas/ # 适配器/请求/输入/结果/回执 + legacy_0.1 Schema(9 个)
├── specs/Agent-OS-Worker-Protocol-v0.1.md # AOS-002 规格
├── tests/AgentOSWorkerProtocol.Tests.ps1 # Pester 5 测试
└── .dsh/ # DeepSeek Harness 衍生包
Quick start / 快速开始
# Prepare:为运行中的 AOS 任务创建最小 Worker 快照与 prepared 作业
$prepared = pwsh -File .\runner\agent_os_worker.ps1 -Action Prepare `
-RuntimeTaskPath "<task_path>" -AdapterManifestPath ".\adapter.json" `
-RequestPath ".\request.json" -AsJson
# Accept:复算权限与证据哈希,接受一个终态结果
pwsh -File .\runner\agent_os_worker.ps1 -Action Accept `
-JobPath "<job_path>" -ResultPath "<job_path>\output\worker_result.json" -AsJson
# Inspect:从 Worker Ledger 恢复状态并复验输入、回执与证据
pwsh -File .\runner\agent_os_worker.ps1 -Action Inspect -JobPath "<job_path>" -AsJson
# 运行测试(Pester 5)
Invoke-Pester .\tests\AgentOSWorkerProtocol.Tests.ps1
DeepSeek Harness 衍生 / DSH Derivative
本项目附带 DeepSeek Harness 衍生包,位于 .dsh/ 目录:
preset.yml— Agent 预设元数据agent.cordis.yml— Cordis 组装(基于 standard 预设,persona 已定制)skills/dsh-agent-os-worker-protocol/SKILL.md— 项目专属技能(skill)
安装与接入方式见 .dsh/README.md(双语)。
License / 许可证
相关项目 / Related Projects
这是 DeepSeek Harness 个人适配系列(共 31 个仓库)的完整导航。 / This is the complete navigation for the DeepSeek Harness personal-adaptation series (31 repos).
Agent OS 内核 / Kernel
dsh-agent-os-runtime · dsh-agent-os-planning · dsh-agent-os-scheduler · dsh-agent-os-worker-protocol(本仓库 / this repo) · dsh-agent-os-observability · dsh-agent-os-specs
Harness 基础设施 / Infrastructure
dsh-harness-core · dsh-graph-entry · dsh-async-job · dsh-file-identity · dsh-json-projection · dsh-manual-approval · dsh-observation-writer · dsh-provider-control · dsh-schema-negotiator · dsh-upgrade-governance
规格与文档 / Specs & Docs
dsh-harness-specs · dsh-novel-specs · dsh-architecture-guide · dsh-powershell-patterns · dsh-json-schema-driven-dev · dsh-llm-agent-harness-guide
适配器 / Adapters
dsh-short-story-engine · dsh-tutorial-video-state-machine · dsh-governance-kernel · dsh-sports-pipeline · dsh-motion-grammar
DSH 总集成 / Integration
dsh-integration · dsh-presets-pack · dsh-skills-pack · dsh-starter-kit
原始 README: https://github.com/h565656445/dsh-agent-os-worker-protocol/blob/main/README.md ↗
同类插件
查看全部 →
deepseek-harness
从仓库或系统描述生成经过校验的自包含交互式架构图、流程图、时序图、数据流图和生命周期图。

dsh-plugin
通过 DSH MCP 客户端挂载 Ouroboros 的纯配置包,在 DSH 中提供 36 个涵盖需求访谈、Seed、执行、评估与演化流程的工具。

dsh-tongflow
基于 TongFlow 的“片场”插件,用于图片、配音、音乐与视频制作:agent 为每个资产生成 TongFlow 工作流文件(.tongflow.json)并通过 TongFlow 插件执行,内嵌工作流画布,按镜头/角色/take 组织项目,附漫剧模板;以 @tongflow 开头的会话进入 Studio 界面。

helloagents
AI 编码 CLI 的工作流层:技能、项目知识、交付检查、更安全的配置写入与可恢复执行

dsh-ai-novel-writer
安装专用 AI 小说创作预设与工作台:提供带修订号的本地项目资产、紧凑侧边工作台,以及需要原生审批的逐文件变更。

rea
用 agent 逆向任何东西:从应用行为到原生二进制