devos/Docs/DevOS-Platform-v2/34_DEPLOYMENT_INFRASTRUCTURE_RUNTIME.md

DevOS Deployment & Infrastructure Runtime

Version: 1.0
Status: Architecture Specification
Scope: Deployment / Infrastructure / Runtime Environment / Server / Container / CI/CD / Health / Rollback
Depends On: 01–33
Project: /home/dev/bun/devos

---

1. 文档目标

本文定义 DevOS Deployment & Infrastructure Runtime。

核心目标：

让 DevOS 能够统一管理项目从“构建”到“部署”再到“运行、监控、升级、回滚”的完整生命周期。

最终链路：

Project
   ↓
Build
   ↓
Artifact
   ↓
Deployment Plan
   ↓
Environment
   ↓
Infrastructure Provider
   ↓
Deploy
   ↓
Runtime Instance
   ↓
Health
   ↓
Observability
   ↓
Upgrade / Rollback

---

2. Deployment Runtime 定位

Deployment Runtime 不是：

• Docker CLI 的包装器
• SSH 脚本集合
• Kubernetes Dashboard
• 云厂商 SDK 集合
• CI/CD 脚本目录

它是：

DevOS 对“项目如何被部署、在哪里运行、如何升级和恢复”的统一抽象。

---

3. 核心模型

核心对象：

Project
Environment
Deployment
DeploymentPlan
DeploymentTarget
Artifact
RuntimeInstance
Service
Infrastructure
Provider
HealthCheck
Release
DeploymentStrategy
Rollback

关系：

Project
   ↓
Environment
   ↓
Deployment Plan
   ↓
Artifact
   ↓
Deployment Target
   ↓
Runtime Instance
   ↓
Health

---

4. Deployment 与 Infrastructure 的区别

必须明确：

Deployment
=
把应用发布到某个运行环境

而：

Infrastructure
=
提供运行应用所需要的基础资源

例如：

Infrastructure:
Server
Container
Network
Database
Object Storage
DNS
Load Balancer

Deployment:
把 my-app:v1.4.0 发布到 server-01

---

5. Provider

基础设施不能写死。

例如：

Docker
SSH
AWS
Cloudflare
Kubernetes
Hetzner
DigitalOcean
Vercel

都应该通过：

Provider Runtime

接入。

---

6. Provider Runtime

统一：

interface InfrastructureProvider {
  id: string;

  capabilities(): ProviderCapability[];

  validate(
    context: ProviderContext
  ): Promise<ValidationResult>;

  plan(
    request: DeploymentRequest
  ): Promise<DeploymentPlan>;

  deploy(
    plan: DeploymentPlan
  ): Promise<DeploymentResult>;

  health(
    target: DeploymentTarget
  ): Promise<HealthResult>;

  destroy(
    target: DeploymentTarget
  ): Promise<DestroyResult>;
}

---

7. Provider 不拥有 Deployment Runtime

Provider 只负责：

实际基础设施操作

不能自己创建：

Task Runtime
Event Runtime
Secret Runtime
Audit Runtime

统一使用 DevOS Runtime。

---

8. Deployment Architecture

User / Agent / CI / CLI
          ↓
     Security Context
          ↓
      Deployment API
          ↓
    Deployment Runtime
          ↓
      Deployment Plan
          ↓
    Provider Resolver
          ↓
 Infrastructure Provider
          ↓
      Infrastructure
          ↓
    Runtime Instance

---

9. Deployment Action

部署本身是 Action：

deployment.create
deployment.plan
deployment.deploy
deployment.cancel
deployment.rollback
deployment.restart
deployment.scale
deployment.destroy

进入：

Action Runtime

---

10. Deployment 不直接执行

错误：

Controller
 ↓
ssh
 ↓
docker run

正确：

API
 ↓
Action
 ↓
Action Runtime
 ↓
Deployment Runtime
 ↓
Provider
 ↓
Task
 ↓
Execution

---

11. Deployment Plan

部署必须先 Plan：

Project
 ↓
Artifact
 ↓
Environment
 ↓
Provider
 ↓
Deployment Plan

Plan 应包含：

Artifact
Target
Environment
Services
Config
Secret References
Network
Health Checks
Strategy
Rollback
Resource Requirements

---

12. Plan / Apply

必须支持：

devos deploy plan

以及：

devos deploy apply

原则：

Plan 与 Apply 分离。

---

13. Dry Run

支持：

devos deploy apply --dry-run

输出：

Target:
production/server-01

Artifact:
my-app:1.4.0

Services:
api
worker
web

Changes:
+ api container
~ worker image
~ web image

Database:
migration pending

Risk:
HIGH

---

14. Environment

Deployment 永远属于：

Environment

例如：

development
staging
production

环境来自：

Environment Runtime

不是 Deployment 自己创建一套环境系统。

---

15. Environment 隔离

必须：

development
    ≠
staging
    ≠
production

至少隔离：

Config
Secret
Credential
Provider
Database
Storage
Network
Deployment
Task

---

16. Production Guard

Production 默认：

protected = true

高风险操作：

Deploy
Destroy
Database Migration
Rollback
Credential Change
Infrastructure Change

可以要求：

Approval Runtime

---

17. Deployment Target

Target 表示：

项目最终运行在哪里。

例如：

server-01
docker-host-02
k8s-cluster-a
aws-ecs-prod

---

18. Target Schema

interface DeploymentTarget {
  id: string;

  organizationId: string;

  providerId: string;

  type: string;

  environmentId: string;

  region?: string;

  metadata: Record<string, unknown>;
}

---

19. Runtime Instance

Deployment Target 是：

目标

Runtime Instance 是：

实际运行实例

例如：

Target:
server-01

Instances:
api-01
worker-01
web-01

---

20. Service

一个 Project 可以包含多个 Service：

services:
  api:
    type: web

  worker:
    type: worker

  agent:
    type: agent

  web:
    type: frontend

---

21. Service 与 Project

Project：

业务项目

Service：

运行单元

例如：

SEO Agent Project
 ├── web
 ├── api
 ├── worker
 └── agent

---

22. Service Definition

interface ServiceDefinition {
  id: string;

  projectId: string;

  type: ServiceType;

  artifact: ArtifactReference;

  replicas: number;

  ports?: PortDefinition[];

  resources?: ResourceRequirements[];

  healthCheck?: HealthCheckDefinition;
}

---

23. Service Types

初始：

web
api
worker
agent
scheduler
mcp
cron
frontend

不要过度细分。

---

24. Artifact

Deployment 不应该直接从源码部署。

正确：

Source
 ↓
Build
 ↓
Artifact
 ↓
Deploy

Artifact 通过：

Artifact Runtime

管理。

---

25. Artifact Examples

Docker Image
npm Package
Static Bundle
Tarball
Binary
Server Bundle

Artifact 必须记录：

ID
Version
Checksum
Build
Project
Source Revision
CreatedAt

---

26. Build 与 Deployment

Build：

产生 Artifact

Deployment：

发布 Artifact

不要把两者混成：

deploy.ts

---

27. Release

Release 是：

某个 Project 在某个版本下的一组可部署 Artifact。

例如：

Release 1.4.0

api: image-a
worker: image-b
web: image-c

---

28. Release Schema

interface Release {
  id: string;

  projectId: string;

  version: string;

  sourceRevision?: string;

  artifacts: ArtifactReference[];

  createdAt: Date;
}

---

29. Immutable Release

Release 一旦发布：

1.4.0

不应该修改成另一个内容。

如果代码变化：

1.4.1

或者：

1.4.0+build2

具体策略由版本 Runtime 决定。

---

30. Reproducible Deployment

生产部署必须尽量：

Release
+
Environment
+
Config Snapshot

决定实际运行内容。

不能：

deploy latest

---

31. 禁止 latest

生产环境禁止：

image: latest

应该：

image: my-app@sha256:...

或者：

release: 1.4.0
artifactChecksum: ...

---

32. Deployment Lock

Deployment 应记录：

DevOS Version
Project Version
Release Version
Artifact Version
Plugin Versions
Module Versions
Config Version
Provider Version

这样可以回答：

这个生产实例到底运行了什么？

---

33. Configuration

Deployment 配置来自：

Configuration Runtime

例如：

DATABASE_URL
LOG_LEVEL
PUBLIC_URL
MODEL
WORKER_CONCURRENCY

但业务代码不能：

process.env.DATABASE_URL

直接读取。

---

34. Secret

Secret 来自：

Secret Runtime

Deployment 只能引用：

secret://production/database

不能保存：

password: "xxxxx"

---

35. Credential

Provider 认证使用：

Credential Runtime

例如：

AWS Credential
GitHub Credential
Cloudflare Credential
SSH Credential
Docker Registry Credential

Deployment 不直接读取 Credential 数据库。

---

36. Network

网络配置属于：

Network Runtime

至少抽象：

Port
DNS
Domain
TLS
Firewall
Private Network
Public Network

第一阶段可以提供：

Docker Network
SSH Host
HTTP Endpoint
DNS Provider

---

37. Domain

Domain 例如：

api.example.com

可以绑定：

Service
Deployment
Environment

通过：

DNS Provider

实现。

---

38. TLS

TLS 可以由：

Provider

提供。

例如：

Let's Encrypt
Cloudflare
Managed Certificate

Deployment Runtime 只描述：

TLS Required

---

39. Health Check

部署完成不代表成功。

必须：

Deploy
 ↓
Start
 ↓
Health Check
 ↓
Ready

---

40. Health Types

至少：

Process Health
HTTP Health
TCP Health
Container Health
Dependency Health
Application Health

---

41. Readiness

服务：

running

不代表：

ready

例如：

Database Migration
Loading Model
Warming Cache

期间：

running = true
ready = false

---

42. Health Runtime

复用：

Health Runtime

统一：

liveness
readiness
dependency
health

---

43. Deployment State

建议：

pending
planning
approved
deploying
starting
health_checking
ready
failed
rolling_back
rolled_back
cancelled
destroying
destroyed

---

44. Deployment State Machine

pending
  ↓
planning
  ↓
approved
  ↓
deploying
  ↓
starting
  ↓
health_checking
  ↓
ready

失败：

deploying
   ↓
failed
   ↓
rollback

---

45. Deployment Task

部署属于 Task：

Deployment
 ↓
Task
 ↓
Jobs

例如：

Pull Artifact
Prepare
Stop Old
Start New
Health Check
Switch Traffic
Cleanup

---

46. Worker

Deployment Job：

Queue
 ↓
Worker

Worker 执行：

Provider Action

而不是：

Deployment Runtime

自己创建 worker。

---

47. Deployment Timeout

每一步必须有：

timeout

例如：

Pull Image: 5m
Start Container: 2m
Health Check: 3m
Migration: 10m

避免：

Deployment hanging forever

---

48. Deployment Cancellation

支持：

devos deploy cancel <id>

但取消必须区分：

Safe Cancellation
Unsafe Cancellation

例如：

Pull Image

可以取消。

而：

Database Migration

可能只能：

Wait

或者：

Rollback

---

49. Deployment Strategy

初始支持：

recreate
rolling
blue-green
canary

第一阶段：

recreate
rolling

优先。

---

50. Recreate

Old
 ↓
Stop
 ↓
New

优点：

Simple

缺点：

Downtime

---

51. Rolling

Old × N
 ↓
New × 1
 ↓
Health
 ↓
New × 2
 ↓
...

必须：

Health Check

通过后才能继续。

---

52. Blue-Green

Blue = Current
Green = New

Deploy Green
 ↓
Health
 ↓
Switch Traffic
 ↓
Blue remains

适合：

Production

但第一阶段可以先作为抽象，不必立即完整实现。

---

53. Canary

95% Old
5% New

然后：

Metrics
 ↓
Health
 ↓
Increase

这是后续增强能力。

---

54. Traffic Switch

流量切换通过：

Load Balancer
Reverse Proxy
DNS
Service Mesh
Provider

实现。

Deployment Runtime 不应该自己实现所有网络层。

---

55. Rollback

Rollback 不是：

git checkout

而是：

将 Deployment 恢复到一个已知的 Release。

例如：

Current:
1.5.0

Rollback:
1.4.2

---

56. Rollback Preconditions

必须确认：

Artifact Exists
Config Compatible
Database Compatible
Plugin Compatible
Provider Available

否则：

Rollback Blocked

---

57. Database Migration 与 Rollback

最危险的是：

App Upgrade
+
Database Migration

例如：

v1
column A

v2
column B

直接回滚代码：

v2 → v1

可能失败。

所以 Migration 必须支持：

Expand
Migrate
Contract

策略。

---

58. Expand / Contract

推荐：

Release N
 ↓
Add new field
 ↓
Release N+1
 ↓
Use new field
 ↓
Release N+2
 ↓
Remove old field

避免：

One-step destructive migration

---

59. Deployment Rollback Plan

每次生产部署前生成：

Rollback Plan

包括：

Previous Release
Previous Config
Database Compatibility
Traffic Strategy
Artifact

---

60. Deployment Approval

生产：

Deployment Plan
 ↓
Risk Evaluation
 ↓
Approval
 ↓
Apply

高风险项：

Database Migration
Permission Change
Secret Change
Infrastructure Destruction

必须更加严格。

---

61. Risk Level

部署风险：

low
medium
high
critical

例如：

Frontend only:
low

API:
medium

Database Migration:
high

Production Destruction:
critical

---

62. Deployment Policy

可以配置：

production:
  requireApproval: true

highRisk:
  requireApproval: true

critical:
  requireTwoPersonApproval: true

最终由：

Policy Runtime

执行。

---

63. CI/CD

CI/CD 不是独立 Runtime。

它应该调用：

Project Factory
Build Runtime
Artifact Runtime
Deployment Runtime
Task Runtime

---

64. CI Pipeline

推荐：

Git Push
 ↓
CI
 ↓
Test
 ↓
Build
 ↓
Artifact
 ↓
Release
 ↓
Deployment Plan
 ↓
Approval
 ↓
Deploy

---

65. Git Provider

GitHub/GitLab/Bitbucket：

Plugin

负责：

Repository
Webhook
Commit
Pull Request
Release

---

66. CI Provider

第一阶段不要自己实现：

GitHub Actions
GitLab CI
Jenkins

统一抽象：

CIRuntime

后续通过 Provider 接入。

---

67. Deployment Trigger

支持：

Manual
Git Push
Tag
Release
Webhook
Schedule
Agent
Workflow

都进入：

Event Runtime

然后：

Event
 ↓
Workflow / Deployment Action

---

68. Git Push

例如：

push main

产生：

git.push

然后：

Workflow
 ↓
Build
 ↓
Release
 ↓
Deploy Staging

生产：

Approval

---

69. Scheduled Deployment

可以：

每周日凌晨部署

但 Scheduler 仍然是：

Scheduler Runtime

Deployment Runtime 不创建第二套 Cron。

---

70. Agent Deployment

Agent 可以提出：

将 staging 上当前 release 部署到 production。

Agent：

Goal
 ↓
Capability Discovery
 ↓
deployment.plan
 ↓
Risk
 ↓
Approval
 ↓
deployment.deploy

Agent 不可以：

ssh root

直接部署。

---

71. MCP Deployment

MCP 可以提供：

deployment.plan
deployment.status
deployment.deploy
deployment.rollback

但：

MCP
 ↓
Action Runtime

不能：

MCP
 ↓
Docker Socket

---

72. SSH Provider

SSH 是基础设施 Provider。

可以：

Connect
Execute Allowed Command
Upload
Download
Check Health

但必须：

Credential Runtime
Security Runtime
Audit Runtime

---

73. SSH Security

禁止默认：

root
password auth
unrestricted shell

优先：

SSH Key
Limited User
Command Policy
Network Policy
Audit

---

74. Docker Provider

Docker Provider 可以：

Image Pull
Container Create
Container Start
Container Stop
Container Remove
Network
Volume
Health

但：

Docker Provider

不拥有：

Deployment State
Task Queue
Audit
Secret

---

75. Docker Socket

生产环境：

Docker Socket

属于高权限资源。

不能：

Plugin
Agent
MCP

直接访问。

只有：

Infrastructure Provider

在受控权限下使用。

---

76. Kubernetes Provider

后续支持：

Kubernetes

但不要第一阶段直接把 Kubernetes API 暴露给：

Agent
Plugin
MCP

必须：

Deployment Action
 ↓
Kubernetes Provider

---

77. Infrastructure Resource

未来可以统一抽象：

Server
Container
Volume
Network
Database
Bucket
DNS
Certificate
LoadBalancer

这些属于：

Infrastructure Resource

---

78. Infrastructure Resource 与业务 Resource

不要混淆。

seo.website

是：

Business Resource

而：

server-01

是：

Infrastructure Resource

但两者都可以遵循：

Resource → Action → Runtime

---

79. Infrastructure Actions

例如：

infra.server.create
infra.server.start
infra.server.stop
infra.server.restart

infra.container.deploy
infra.container.restart

infra.dns.create
infra.dns.delete

统一进入：

Action Runtime

---

80. Infrastructure Query

查询：

server list
container list
deployment status
health

通过：

Query Runtime

而不是：

Controller → Provider → random JSON

---

81. Infrastructure State

Provider 返回的状态：

External State

DevOS 保存：

Desired State
Observed State

---

82. Desired vs Observed

核心模型：

Desired
   ↓
Deployment Plan
   ↓
Provider
   ↓
Observed

例如：

Desired replicas = 3
Observed replicas = 2

则：

Drift Detected

---

83. Drift Detection

必须支持：

deployment.drift.detect

检测：

Image
Replicas
Ports
Environment
Network
Domain
TLS
Resources

---

84. Reconciliation

后续可以：

Desired
 ↓
Observed
 ↓
Diff
 ↓
Reconcile

但是第一阶段：

只检测，不自动修复。

生产自动修复属于后续能力。

---

85. Infrastructure Discovery

可以：

devos infra discover

发现：

Servers
Containers
Networks
Volumes
Domains

然后建立：

Infrastructure Registry

---

86. Infrastructure Registry

记录：

Provider
Resource
External ID
Environment
Organization
Project
Status
Metadata
Last Observed

---

87. External ID

必须保留：

externalId

例如：

AWS Instance ID
Docker Container ID
Kubernetes Pod ID

不能只使用 DevOS UUID。

---

88. Import Existing Infrastructure

DevOS 应支持：

Import

例如：

已有 Docker Server
 ↓
Discover
 ↓
Import
 ↓
Register

避免强制用户重新创建所有基础设施。

---

89. Infrastructure Ownership

资源可以：

managed
external
imported
orphaned

只有：

managed

默认允许自动修改。

---

90. Orphan Detection

例如：

DevOS Deployment 已删除
但 Docker Container 还存在

应该：

orphaned

而不是自动删除。

---

91. Cost

Infrastructure Provider 可以提供：

estimatedCost
actualCost
usage

进入：

Usage / Cost Runtime

例如：

Project
 ↓
Deployment
 ↓
Provider
 ↓
Cost

---

92. Resource Quota

组织可以限制：

CPU
Memory
Storage
Deployments
Servers
Containers
Network
Cost

通过：

Quota Runtime

统一管理。

---

93. Deployment Budget

Agent Deployment 还可以有：

Max Cost
Max Instances
Max Duration
Max Network

例如：

Agent:
不要让部署成本超过 $20/月。

Resolver / Policy 必须验证。

---

94. Observability

Deployment 必须自动接入：

Logging
Metrics
Tracing
Audit
Health

所有日志包含：

deploymentId
projectId
environmentId
releaseId
providerId

---

95. Deployment Timeline

UI 显示：

12:00 Plan
12:01 Approval
12:02 Pull Artifact
12:03 Stop Old
12:04 Start New
12:05 Health Check
12:06 Traffic Switch
12:07 Completed

数据来自：

Task
Execution
Event
Audit

而不是 UI 自己维护状态。

---

96. Deployment Logs

Provider 日志：

stdout
stderr
command result
provider response

进入：

Logging / Artifact Runtime

敏感内容必须：

Redact

---

97. Secret Leakage Protection

绝对禁止日志出现：

API_KEY
PASSWORD
TOKEN
PRIVATE_KEY

例如：

docker login -p XXXXX

必须：

Redacted

---

98. Deployment Events

建议事件：

deployment.created
deployment.planned
deployment.approved
deployment.started
deployment.progressed
deployment.ready
deployment.failed
deployment.rollback.started
deployment.rollback.completed
deployment.cancelled
deployment.destroyed
deployment.drift.detected

---

99. Notifications

事件可以触发：

Notification Runtime

例如：

Production Deployment Failed

通知：

In-App
Email
Chat

Deployment 不直接发 Email。

---

100. Deployment API

建议：

POST /v1/deployments
POST /v1/deployments/:id/plan
POST /v1/deployments/:id/apply
POST /v1/deployments/:id/cancel
POST /v1/deployments/:id/rollback

GET /v1/deployments
GET /v1/deployments/:id
GET /v1/deployments/:id/tasks
GET /v1/deployments/:id/logs
GET /v1/deployments/:id/health

---

101. Infrastructure API

GET /v1/infrastructure/providers
GET /v1/infrastructure/targets
GET /v1/infrastructure/resources
POST /v1/infrastructure/discover
POST /v1/infrastructure/import

所有 API：

Security
Authorization
Scope
Audit

---

102. Deployment SDK

建议：

@devos/deployment-sdk

提供：

deployments.plan()
deployments.apply()
deployments.cancel()
deployments.rollback()
deployments.status()
deployments.health()

---

103. CLI

最终：

devos deploy plan
devos deploy apply
devos deploy status
devos deploy logs
devos deploy cancel
devos deploy rollback
devos deploy history

Infrastructure：

devos infra list
devos infra discover
devos infra health
devos infra doctor

---

104. Deployment UI

建议：

Deployments
 ├── Overview
 ├── Releases
 ├── Environments
 ├── Targets
 ├── Services
 ├── Health
 ├── Logs
 ├── Timeline
 └── Rollbacks

---

105. Project Detail

Project 页面可以：

Project
 ↓
Deployments

显示：

Development
Staging
Production

例如：

Production
Release: 1.4.0
Status: Healthy
Target: server-01
Services: 4/4

---

106. Deployment Detail

显示：

Release
Artifact
Environment
Provider
Target
Services
Health
Task
Logs
Events
Audit
Rollback

---

107. Agent Deployment UI

Agent 操作显示：

Goal:
Deploy SEO Agent to production.

Plan:
1. Validate release
2. Run migration
3. Deploy API
4. Deploy worker
5. Health check
6. Switch traffic

Risk:
HIGH

Approval:
Required

而不是：

Agent is thinking...

不展示隐藏 Chain-of-Thought。

---

108. Deployment History

必须保留：

Release
Actor
Time
Environment
Target
Result
Rollback

可以回答：

谁在什么时候把哪个版本部署到了生产？

---

109. Deployment Audit

至少记录：

deployment.plan
deployment.approved
deployment.started
deployment.completed
deployment.failed
deployment.rollback
deployment.destroy

---

110. Multi-Tenant Security

所有 Deployment：

organizationId
projectId
environmentId

必须继承：

SecurityContext

不能：

Project A
 ↓
Deploy
 ↓
Project B

---

111. Environment Permission

权限示例：

deployment.read
deployment.plan
deployment.deploy
deployment.rollback
deployment.destroy

environment.production.deploy
environment.production.destroy

---

112. Agent Permission

Agent 权限：

deployment.plan

不等于：

deployment.deploy

更不等于：

production.destroy

---

113. Plugin Permission

Plugin 例如 Docker Plugin：

只能获得：

docker.container.manage

不能：

environment.production.admin

除非明确授予。

---

114. Provider Network Security

Provider 必须经过：

Network Policy

控制：

Allowed Hosts
Allowed Ports
Regions
Protocols

避免任意外连。

---

115. SSRF

特别是：

HTTP Health Check
Webhook
Provider
Metadata API

必须防：

169.254.169.254
localhost
127.0.0.1
private network

除非显式允许。

---

116. Command Execution

基础设施操作涉及：

SSH
Docker
Shell

必须：

Command Allowlist
Argument Validation
Timeout
Resource Limit
Audit

禁止：

eval(userInput)

---

117. Shell Injection

错误：

exec(`docker run ${image}`);

正确：

Structured Command
 ↓
Argument Validation
 ↓
Command Runtime

---

118. Infrastructure Secrets

禁止：

SSH private key
Cloud API token
Registry password

进入：

Project Manifest
Git
Artifact
Logs

只保存：

Credential Reference
Secret Reference

---

119. Deployment Config Snapshot

部署时记录：

Config Version

而不是：

实时 Config

否则：

昨天部署的版本

可能因为今天 Config 改了而无法重现。

---

120. Secret Snapshot

Secret 不保存 plaintext。

Deployment 记录：

secretRef
secretVersion

这样可以知道：

使用的是哪个 Secret 版本。

---

121. Configuration Drift

如果：

Desired Config
≠
Observed Config

报告：

Config Drift

---

122. Deployment Drift

统一：

Desired
Observed

检查：

Artifact
Config
Replicas
Network
Domain
TLS
Resources

---

123. Deployment Repair

第一阶段：

Detect
Report

第二阶段：

Plan Repair

第三阶段：

Auto Reconcile

不要第一阶段就自动修复生产。

---

124. Disaster Recovery

Deployment Runtime 需要考虑：

Server Failure
Provider Failure
Database Failure
Storage Failure
Deployment Failure

但：

Disaster Recovery 不应该重新创建一套 Storage/Backup Runtime。

复用：

Storage Runtime
Artifact Runtime
Backup Provider
Task Runtime

---

125. Backup

生产部署前可以：

Database Backup
Config Snapshot
Artifact Retention

Backup 通过：

Backup Provider

实现。

---

126. Restore

支持：

devos backup list
devos backup restore

恢复必须：

Approval
Audit
Environment Protection

---

127. Rollback 与 Restore

两者必须区分：

Rollback
=
恢复应用 Release

Restore
=
恢复数据 / 基础设施状态

不能混为一谈。

---

128. Disaster Recovery Plan

Project 可以声明：

recovery:
  rpo: 1h
  rto: 30m

最终交给：

Backup / Infrastructure / Deployment Runtime

实现。

---

129. Availability

服务可以定义：

availability:
  replicas: 2
  strategy: rolling

Provider 决定实际实现。

---

130. Resource Requirements

Service 可以声明：

resources:
  cpu: "1"
  memory: "1Gi"
  storage: "10Gi"

Provider 转换为：

Docker
Kubernetes
VM
Cloud

各自配置。

---

131. Resource Abstraction

不要：

docker:
  memory: 1024

作为业务 Manifest。

优先：

resources:
  memory: "1Gi"

Provider 自己映射。

---

132. Deployment Profile

Project Factory 可以声明：

deployment:
  profile: docker

后续：

docker
kubernetes
serverless
cloud
bare-metal

---

133. Profile 与 Provider

Profile：

运行方式

Provider：

实际供应商

例如：

Profile:
docker

Provider:
remote-docker

或者：

Profile:
kubernetes

Provider:
aws-eks

---

134. Deployment Environment Matrix

项目可能：

development → local docker
staging → remote docker
production → kubernetes

Manifest：

deployment:
  environments:
    development:
      profile: docker

    staging:
      profile: docker

    production:
      profile: kubernetes

---

135. Promote Release

推荐：

Build once
Deploy many

即：

Build
 ↓
Artifact
 ↓
Staging
 ↓
Production

不要：

Build Staging
Build Production

两个不同 Artifact。

---

136. Promotion

devos release promote 1.4.0 \
  --from staging \
  --to production

实际：

Same Artifact

---

137. Release Approval

Production Promotion：

Release
 ↓
Tests
 ↓
Staging Health
 ↓
Approval
 ↓
Production

---

138. Deployment Verification

部署成功必须至少验证：

Process
Health
Endpoint
Dependencies
Metrics
Error Rate

---

139. Post-Deploy Checks

例如：

HTTP /health
HTTP /ready
Database Connection
Queue Connection
Storage Connection
Provider Health

失败：

Deployment Failed

根据策略：

Rollback

---

140. Smoke Test

可以定义：

verification:
  smoke:
    - GET /health
    - GET /ready
    - GET /api/version

Smoke Test 通过后：

Deployment Ready

---

141. Deployment Verification 与 E2E

Smoke：

快速验证

E2E：

完整业务流程

两者都应该进入：

Task Runtime

---

142. Deployment Health Window

部署成功后可以观察：

5 min
10 min
30 min

检查：

Error Rate
Latency
CPU
Memory
Restart

异常：

Auto Rollback

这是后续增强功能。

---

143. Auto Rollback Policy

示例：

rollback:
  automatic:
    enabled: false
    errorRate: 0.1
    window: 5m

默认：

false

生产自动回滚必须显式开启。

---

144. Deployment Notification

部署：

started
success
failed
rollback

通过 Notification Runtime：

In-App
Email
Slack
Discord
Webhook

---

145. Communication Provider

Deployment 不应该直接：

sendSlack()
sendEmail()

而：

Event
 ↓
Notification
 ↓
Communication Runtime

---

146. Deployment Cost

每次 Deployment 可以记录：

Build Cost
Storage Cost
Provider Cost
Network Cost
Compute Cost

进入：

Usage Runtime

---

147. Deployment Metrics

至少：

deployment.duration
deployment.success
deployment.failure
deployment.rollback
deployment.health_check
deployment.drift

---

148. Infrastructure Metrics

infra.cpu
infra.memory
infra.disk
infra.network
infra.container.restart
infra.provider.error

具体采集由：

Observability Provider

完成。

---

149. Deployment Tracing

Trace：

deployment action
 ↓
task
 ↓
job
 ↓
provider
 ↓
external request

所有步骤共享：

traceId
correlationId
deploymentId

---

150. Deployment Tables

建议：

projects
environments
deployment_targets
deployment_plans
deployments
deployment_services
releases
release_artifacts
runtime_instances
health_checks
deployment_events
infrastructure_resources

---

151. Project 与 Deployment

Project
 1
 │
 N
 ↓
Deployment

Deployment 必须属于：

Project
Environment
Release
Target

---

152. Deployment 与 Release

Release
 1
 │
 N
 ↓
Deployment

同一个 Release 可以：

staging
production

分别部署。

---

153. Deployment 与 Artifact

一个 Release：

api artifact
worker artifact
web artifact

Deployment 使用：

exact artifact

---

154. Infrastructure Registry

建议包：

packages/infrastructure/

包含：

resource
provider
target
discovery
drift
health

---

155. Deployment Packages

建议：

packages/deployment/
packages/release/
packages/infrastructure/
packages/deployment-provider/
packages/deployment-testing/
packages/backup/

---

156. Provider Packages

例如：

plugins/
├── docker/
├── ssh/
├── kubernetes/
├── cloudflare/
└── aws/

它们通过：

Plugin Runtime

注册 Provider。

---

157. Docker Plugin

Docker 不应成为：

Core

而应：

Official Plugin

实现：

Infrastructure Provider

---

158. SSH Plugin

同样：

Official Plugin

实现：

SSH Infrastructure Provider

---

159. Cloud Provider

例如 AWS：

Official / Verified Plugin

提供：

Compute
Storage
Network
DNS
Container

但每一项都通过：

Provider Capability

声明。

---

160. Provider Capability

interface ProviderCapability {
  id: string;

  version: string;

  resourceTypes: string[];

  actions: string[];

  regions?: string[];
}

例如：

docker.container.deploy
aws.ec2.create
cloudflare.dns.create

---

161. Capability Discovery

Agent 不应该知道：

AWS API
Docker API

它只发现：

deployment.deploy
infra.container.deploy

然后：

Provider Resolver

选择具体实现。

---

162. Provider Health

每个 Provider：

healthy
degraded
unavailable
unknown

Provider Runtime 应定期：

Health Check

---

163. Provider Failover

可以配置：

Primary Provider
Secondary Provider

但生产自动 Failover 必须：

Policy
Approval
Compatibility

不能简单：

provider A failed
→ random provider B

---

164. Region

Provider 支持：

region
zone
location

Project Manifest 不应该强绑定：

AWS us-east-1

除非用户明确要求。

---

165. Infrastructure Scheduling

Deployment 可以选择：

region
capacity
cost
latency
compliance

最终：

Placement Policy

决定 Target。

---

166. Placement Policy

例如：

placement:
  region: ap-northeast-1
  maxCost: 50

Resolver：

Provider
 ↓
Available Targets
 ↓
Policy
 ↓
Selected Target

---

167. Compliance

未来可以支持：

data residency
region restrictions
network restrictions
provider restrictions

但这些属于：

Policy Runtime

而不是 Deployment 自己硬编码。

---

168. Deployment Testing

必须覆盖：

Plan
Apply
Cancel
Retry
Health
Rollback
Drift
Provider Failure
Network Failure
Artifact Failure
Config Failure
Secret Failure

---

169. Provider Contract Test

每个 Provider 必须通过：

Provider Contract Suite

验证：

validate
plan
deploy
health
destroy

---

170. Fake Provider

测试环境提供：

FakeInfrastructureProvider

可以模拟：

Success
Timeout
Failure
Partial Failure
Drift
Health Failure

---

171. Deployment Integration Test

例如：

Project
 ↓
Build
 ↓
Artifact
 ↓
Deployment
 ↓
Fake Docker
 ↓
Health
 ↓
Ready

完整测试。

---

172. E2E Test

至少：

Create Project
 ↓
Generate
 ↓
Build
 ↓
Release
 ↓
Deploy Staging
 ↓
Health
 ↓
Promote
 ↓
Deploy Production
 ↓
Rollback

---

173. Chaos Tests

后续增加：

Provider Timeout
Network Disconnect
Worker Crash
Container Crash
Health Failure
Database Failure

验证：

Recovery

---

174. Worker Crash

Deployment Worker 在：

deploying

时崩溃。

系统应该：

Lease Expire
 ↓
Job Retry
 ↓
Check Actual State
 ↓
Resume

不能盲目重新执行破坏性操作。

---

175. Provider Timeout

Provider：

timeout

不一定代表：

operation failed

可能：

external operation succeeded
but response lost

因此必须：

Query Observed State

再决定 Retry。

---

176. Idempotency

Deployment Actions 必须支持：

Idempotency-Key

例如：

deploy(project=A, release=1.4.0)

重复请求：

same operation

不能创建两个 Deployment。

---

177. Deployment Concurrency

同一个：

Project + Environment

默认只能有：

one active deployment

否则：

Deploy A
Deploy B

互相覆盖。

---

178. Deployment Lock

使用：

Redis Lease

协调：

projectId
environmentId

但真正状态：

PostgreSQL

持久化。

---

179. Queue

Deployment Queue 统一：

Task
Job
Queue
Worker

不要创建：

deploymentQueue

第二套系统。

可以：

queue:deployment

作为逻辑队列。

---

180. Backpressure

大量项目同时部署：

1000 deployments

必须：

Queue
Concurrency
Rate Limit
Provider Limit

控制。

---

181. Deployment Priority

例如：

critical
high
normal
low

但 Priority 不能绕过：

Authorization
Approval
Environment Guard

---

182. Deployment Maintenance Window

生产可以定义：

maintenance:
  allowed:
    - "02:00-04:00"

由：

Scheduler / Policy Runtime

判断。

---

183. Emergency Deployment

可以支持：

Break Glass

但必须：

Short TTL
Strong Audit
Reason
Actor
Approval / notification

---

184. Production Protection

最终生产保护原则：

PRIVATE
SCOPED
AUDITED
APPROVED
IDEMPOTENT
REVERSIBLE

---

185. Deployment Project Manifest

Project Factory 可以生成：

deployment:
  profile: docker

services:
  api:
    artifact: api
    replicas: 2

  worker:
    artifact: worker
    replicas: 1

  web:
    artifact: web
    replicas: 2

environments:
  staging:
    target: staging

  production:
    target: production
    approval: required

Manifest 描述：

Desired Deployment。

具体 Provider 实现由 Runtime 决定。

---

186. Deployment Profile Example

deployment:
  profile: docker

services:
  api:
    type: api
    port: 3000

  worker:
    type: worker

Docker Provider 可以转换：

Docker Network
Container
Port
Health Check

---

187. Kubernetes Profile

未来：

deployment:
  profile: kubernetes

同样的 Service：

api
worker
web

转换成：

Deployment
Service
ConfigMap
Secret Reference
Ingress

但：

Kubernetes-specific objects 应属于 Provider/Template，而不是业务 Manifest 核心模型。

---

188. Deployment Template

复杂部署可以使用：

Deployment Template

例如：

single-server
high-availability
worker-cluster
web-api-worker

Project Factory 可以选择。

---

189. Project Factory Integration

完整：

Project Manifest
 ↓
Deployment Profile
 ↓
Service Definition
 ↓
Build
 ↓
Artifact
 ↓
Release
 ↓
Deployment Plan

---

190. Deployment Upgrade

升级：

Release 1.4
 ↓
Release 1.5

流程：

Compatibility
 ↓
Migration Plan
 ↓
Approval
 ↓
Deploy
 ↓
Health
 ↓
Promote

---

191. Deployment Version Compatibility

检查：

Project Version
Runtime Version
Module Version
Plugin Version
Artifact Version
Database Version
Config Version

复用：

Compatibility Runtime

---

192. Deployment Rollback Compatibility

Rollback 前：

Previous Release

必须检查：

DB Schema
Config
Plugin
Runtime
Artifact

---

193. Deployment Garbage Collection

旧 Artifact / Container：

保留 N 个 Release

然后：

Garbage Collection

但不能删除：

Current Release
Rollback Release
Protected Artifact

---

194. Artifact Retention

例如：

retention:
  releases: 10
  artifacts: 30d

通过：

Artifact Runtime

执行。

---

195. Infrastructure Cleanup

删除 Project 后：

默认：

Do Not Destroy External Infrastructure

必须明确：

destroy infrastructure

并可能要求：

Approval

---

196. Project Delete

正确：

Delete Project
 ↓
Disable Deployment
 ↓
Keep Infrastructure
 ↓
Keep Artifacts

除非用户明确：

Destroy

---

197. Infrastructure Destroy

Destroy 是：

critical

必须：

Explicit Action
Strong Authorization
Production Guard
Audit
Approval

---

198. Deployment Doctor

CLI：

devos deploy doctor

检查：

Provider
Credential
Network
Artifact
Config
Secret
Target
Health
Task
Queue
Worker

---

199. Infrastructure Doctor

devos infra doctor

检查：

Provider Connectivity
Credentials
Targets
Resource Drift
Health
Capacity
Network

---

200. Deployment Architecture Rules

必须：

Deployment → Action Runtime
Deployment → Task Runtime
Deployment → Event Runtime
Deployment → Provider Runtime
Deployment → Config Runtime
Deployment → Secret Runtime
Deployment → Artifact Runtime
Deployment → Health Runtime
Deployment → Observability Runtime

禁止：

Deployment → raw Docker
Deployment → raw SSH
Deployment → raw AWS SDK
Deployment → raw DB

---

201. Package Layout

推荐：

packages/
├── deployment/
│   ├── domain/
│   ├── runtime/
│   ├── plan/
│   ├── strategy/
│   ├── rollback/
│   └── verification/
│
├── release/
├── infrastructure/
│   ├── domain/
│   ├── runtime/
│   ├── provider/
│   ├── target/
│   ├── discovery/
│   └── drift/
│
├── deployment-sdk/
└── deployment-testing/

---

202. Official Providers

第一阶段建议：

Official Docker Provider
Official SSH Provider
Official Local Provider

后续：

Kubernetes
AWS
Cloudflare
GCP
Azure

通过 Plugin。

---

203. Local Provider

开发环境非常重要：

Local Provider

支持：

Local Process
Local Docker

用于：

dev
test

---

204. Mock Provider

测试：

Mock Provider

不连接真实基础设施。

---

205. Provider Development

开发者可以通过：

@devos/plugin-sdk

创建：

Infrastructure Provider

而不是修改 DevOS Core。

---

206. Provider Certification

Provider 需要：

Basic
Runtime Safe
Secure
Verified
Official

与 Plugin Certification 对齐。

---

207. Provider Permission Manifest

Provider Plugin 声明：

permissions:
  - infrastructure.read
  - infrastructure.deploy

network:
  allowed:
    - "*.docker.local"

高权限 Provider：

approval

---

208. Provider Isolation

未来支持：

in-process
restricted
isolated
container

与 Plugin Sandbox 共用。

---

209. Deployment Runtime 不拥有 Cloud SDK

例如 AWS：

禁止：

packages/deployment/aws.ts

大量写：

new EC2Client(...)

应该：

AWS Plugin
 ↓
AWS Provider
 ↓
Deployment Runtime

---

210. Provider Adapter

Provider 负责：

External API
 ↓
Normalize

例如：

AWS response
Docker response
Kubernetes response

统一为：

DeploymentResult
HealthResult
InfrastructureResource

---

211. Error Normalization

Provider 错误：

DockerError
AWSError
SSHError

统一为：

ProviderError
DeploymentError
InfrastructureError

并记录：

code
retryable
safeMessage
providerCode

---

212. Retry

Deployment Retry：

Provider Retry Policy

复用：

Task Retry Runtime

不要：

for (let i = 0; i < 5; i++)

散落在 Provider。

---

213. Circuit Breaker

Provider 连续失败：

Provider
 ↓
Circuit Open

防止：

1000 Deployments
 ↓
全部攻击 Provider

复用：

Provider Runtime

---

214. Rate Limit

限制：

Provider
Organization
Project
Environment
Action

复用：

Rate Limit Runtime

---

215. Deployment Observability

每个 Deployment：

deploymentId

贯穿：

Action
Task
Job
Provider
Event
Audit
Logs
Metrics
Trace

这样可以做到：

一次部署全链路可追踪。

---

216. Deployment Cost Attribution

成本可以归属：

Organization
Project
Environment
Deployment
Release
Agent
Actor

例如：

Agent-generated deployment

可以统计：

agent cost
infrastructure cost

---

217. Deployment Security Defaults

默认：

DENY
PRIVATE
SCOPED
AUDITED
APPROVED
IDEMPOTENT

生产：

PROTECTED

Secret：

SHORT-LIVED

Provider：

LEAST-PRIVILEGE

---

218. Zcode Implementation Protocol

实现本章必须按照：

1. Deployment Domain
2. Release / Artifact Integration
3. Environment Integration
4. Provider Runtime
5. Mock Provider
6. Deployment Plan
7. Deployment Task
8. Health Verification
9. Docker Provider
10. SSH Provider
11. Rollback
12. Drift
13. UI
14. API / SDK / CLI
15. Security
16. Observability
17. Tests

---

219. 第一 Vertical Slice

实现：

Local Project
 ↓
Build
 ↓
Artifact
 ↓
Local Provider
 ↓
Deployment
 ↓
Health

成功后再做：

Docker

---

220. 第二 Vertical Slice

Project
 ↓
Docker Artifact
 ↓
Docker Provider
 ↓
Deploy
 ↓
Health
 ↓
Logs

---

221. 第三 Vertical Slice

增加：

Task
Job
Queue
Worker

验证：

Deployment

不会阻塞 API。

---

222. 第四 Vertical Slice

增加：

Release
Promotion
Staging
Production
Approval

验证：

Build Once
Deploy Many

---

223. 第五 Vertical Slice

增加：

Rollback

验证：

Release 1
 ↓
Release 2
 ↓
Failure
 ↓
Rollback
 ↓
Release 1

---

224. 第六 Vertical Slice

增加：

SSH Provider

验证：

Remote Deployment

但：

Credential
Security
Audit

必须完整接入。

---

225. 第七 Vertical Slice

Agent：

Deploy staging

Agent：

Plan
 ↓
Risk
 ↓
Approval
 ↓
Deploy
 ↓
Observe

---

226. 第八 Vertical Slice

自然语言：

把当前 staging 版本发布到生产，如果健康检查失败就停止。

转换：

Goal
 ↓
Deployment Plan
 ↓
Policy
 ↓
Approval
 ↓
Deploy
 ↓
Health

Agent 不直接执行 Shell。

---

227. Deployment Definition of Done

第一阶段：

[ ] Deployment Domain
[ ] Release
[ ] Artifact Integration
[ ] Environment Integration
[ ] Deployment Plan
[ ] Plan / Apply
[ ] Task Integration
[ ] Queue Integration
[ ] Worker Integration
[ ] Mock Provider
[ ] Local Provider
[ ] Docker Provider
[ ] Health Check
[ ] Logs
[ ] Deployment Events
[ ] Audit
[ ] Rollback
[ ] Deployment Lock
[ ] Idempotency
[ ] Security
[ ] Production Guard
[ ] API
[ ] SDK
[ ] CLI
[ ] UI
[ ] Contract Tests
[ ] Integration Tests
[ ] E2E Tests

---

228. Infrastructure Definition of Done

[ ] Infrastructure Resource
[ ] Provider Runtime
[ ] Target
[ ] Discovery
[ ] Import
[ ] External ID
[ ] Health
[ ] Drift Detection
[ ] Docker Provider
[ ] SSH Provider
[ ] Credential Integration
[ ] Network Policy
[ ] Provider Health
[ ] Provider Contract Tests
[ ] Security Tests

---

229. Zcode 禁止事项

禁止：

Controller → Docker

禁止：

Controller → SSH

禁止：

Agent → SSH

禁止：

MCP → Docker Socket

禁止：

Plugin → Production Infrastructure

禁止：

Deployment → Secret plaintext

禁止：

Deployment → custom queue

禁止：

Deployment → custom scheduler

禁止：

Project → infrastructure implementation

---

230. Zcode Deployment Master Prompt

You are implementing the DevOS Deployment & Infrastructure Runtime.

Deployment is an Action Runtime capability.

The correct flow is:

Project
→ Release
→ Artifact
→ Environment
→ Deployment Plan
→ Approval
→ Deployment Task
→ Provider
→ Runtime Instance
→ Health
→ Event
→ Observability

Never let business code call:
Docker
SSH
AWS SDK
Kubernetes API
Cloud APIs
directly.

All infrastructure operations must go through:
Infrastructure Provider
Provider Runtime
Security Runtime
Credential Runtime
Action Runtime
Task Runtime

Deployment must use exact immutable artifacts.

Never deploy "latest" in production.

Plan and Apply must be separate.

Production deployments must respect:
Authorization
Environment Guard
Risk Policy
Approval

Rollback must restore a known Release, not simply change Git code.

Provider failures and timeouts must check observed state before retrying.

Deployment must be idempotent.

Do not create a second queue, scheduler, event bus, secret system,
logging system, or configuration system.

Reuse existing DevOS Runtime components.

The Infrastructure Provider is an adapter to external infrastructure,
not a second DevOS Runtime.

---

231. 最终 Deployment Architecture

                         USER
                          │
                         AGENT
                          │
                         CI
                          │
                         CLI
                          │
                          ↓
                  Security Context
                          │
                          ↓
                   Deployment API
                          │
                          ↓
                   Action Runtime
                          │
                          ↓
                  Deployment Runtime
                          │
              ┌───────────┴────────────┐
              ↓                        ↓
        Release / Artifact        Environment
              │                        │
              └───────────┬────────────┘
                          ↓
                   Deployment Plan
                          │
                          ↓
                    Policy / Approval
                          │
                          ↓
                    Task / Job / Queue
                          │
                          ↓
                        Worker
                          │
                          ↓
                  Provider Runtime
                          │
             ┌────────────┼────────────┐
             ↓            ↓            ↓
          Docker         SSH       Kubernetes
             │            │            │
             └────────────┼────────────┘
                          ↓
                  Runtime Instance
                          │
                          ↓
                       Health
                          │
                ┌─────────┼──────────┐
                ↓         ↓          ↓
              Event     Logs      Metrics
                │         │          │
                └─────────┼──────────┘
                          ↓
                  Observability
                          │
                          ↓
                  Notification / UI

---

232. DevOS 完整生产闭环

经过 33 + 34，DevOS 已经形成：

Requirement
    ↓
Agent
    ↓
Project Manifest
    ↓
Project Factory
    ↓
Capability Resolver
    ↓
Module / Plugin
    ↓
Generated Project
    ↓
Build
    ↓
Artifact
    ↓
Release
    ↓
Deployment Plan
    ↓
Environment
    ↓
Provider
    ↓
Infrastructure
    ↓
Runtime Instance
    ↓
Health
    ↓
Observability
    ↓
Event
    ↓
Notification

发生故障：

Health Failure
     ↓
Event
     ↓
Agent / Workflow
     ↓
Diagnosis
     ↓
Rollback Plan
     ↓
Approval
     ↓
Rollback

---

233. DevOS 最终职责边界

Project Factory
    ↓
“我要什么项目？”

Build Runtime
    ↓
“如何产生 Artifact？”

Release Runtime
    ↓
“这个版本包含什么？”

Deployment Runtime
    ↓
“这个版本部署到哪里？”

Provider Runtime
    ↓
“如何操作外部基础设施？”

Environment Runtime
    ↓
“在哪个隔离环境运行？”

Health Runtime
    ↓
“它是否正常？”

Observability Runtime
    ↓
“发生了什么？”

Event Runtime
    ↓
“发生之后通知谁/触发什么？”

Agent Runtime
    ↓
“下一步应该做什么？”

---

234. 最终 Golden Rule

Deployment Runtime 负责描述和执行“把哪个 Release 部署到哪个 Environment”，Infrastructure Provider 负责“如何操作实际基础设施”；任何 Docker、SSH、Cloud、Kubernetes 操作都必须经过统一的 Security + Credential + Provider + Action + Task Runtime。

最终链路：

PROJECT
   ↓
RELEASE
   ↓
ARTIFACT
   ↓
DEPLOYMENT PLAN
   ↓
ENVIRONMENT
   ↓
APPROVAL / POLICY
   ↓
TASK / JOB
   ↓
PROVIDER
   ↓
INFRASTRUCTURE
   ↓
RUNTIME INSTANCE
   ↓
HEALTH
   ↓
OBSERVABILITY
   ↓
EVENT
   ↓
NOTIFICATION / AGENT / ROLLBACK

DevOS 不应该成为另一个 Kubernetes，也不应该成为另一个 Terraform。

它真正应该做的是：

把 Project、Artifact、Environment、Provider、Deployment、Task、Security、Observability 统一到一个可被 Human、Agent、Plugin、MCP、Workflow、CLI、API 共同使用的 Runtime 中。