Code Context

A collection of OpenCode plugins that enable multiple AI agents to collaborate on software development tasks.

Plugins

Plugin	Agents	Approach	Best For	Docs
Commander	4	Single orchestrator, fast iteration, Coder↔Tester fix loops	Most development tasks	README
Emperor	11	Three Departments & Six Ministries (三省六部), governance with checks & balances	Tasks requiring rigorous multi-stage review	README
Hive	Dynamic	Domain auto-discovery, EventBus coordination, autonomous execution	Large multi-domain projects	README
Superpower	Dynamic	Loads agents, skills, and commands from superpowers repo	Extensible agent capabilities	README
Stock	5	Multi-dimensional A-share analysis (fundamental, technical, sentiment, flow, sector)	Chinese A-share stock analysis	README
Triage	6	Jira ticket triage (Bug/Feature), code investigation, solution design	Bug triage and feature specification	-

Commander — Adaptive 4-Agent Team

User Task → Lead (analyze & plan) → Coder (implement) ↔ Tester (verify) → [Reviewer] → Report

• Lead explores the codebase and creates plans with adaptive complexity classification (trivial / simple / standard / complex)
• Coder↔Tester fix loop: Tester fails → Coder fixes in the same session (context accumulates) → Tester re-verifies → up to N rounds
• Reviewer only engages on complex tasks
• Parallel wave dispatch for independent subtasks

Emperor — 11-Agent Governance System (三省六部)

Edict → Crown Prince → Jinyiwei Recon → Zhongshu (plan) → Menxia (review/veto) → Shangshu → Six Ministries (parallel) → Memorial

• Full governance model: planning, review/veto, parallel execution, post-verification
• Jinyiwei reconnaissance with git-hash caching
• Mandatory department participation (Hubu testing enforced)
• Sensitive operation detection with manual confirmation

Hive — Dynamic Domain-Agent Coordination

Startup → Scan project → Discover domains → Create per-domain Agents + Queen → EventBus coordination → Autonomous execution

• Domain Discovery: Static project scanning + LLM enrichment + user config merge
• EventBus: Pub/sub event system for agent-to-agent communication
• Queen Coordinator: Broadcasts requirements, negotiates interfaces, dispatches parallel tasks
• Autonomous Execution: Domain agents can self-adapt to breaking changes from dependencies

Stock — Multi-Dimensional A-Share Analysis

Query → Fundamentalist (基本面) + Technician (技术面) + Sentiment (情绪) + Flow (资金流) + Industry (行业) → HTML Report

• 5 并发 Agent: 基本面、技术面、情绪面、资金流、行业分析
• 智能评分: 预设权重（保守/平衡/激进），自定义权重
• HTML 报告生成: 可视化分析结果

Triage — Jira Ticket Analysis & Implementation

Jira Ticket → Triage (分类) → Scout (探索) → [Detective (Bug调查) | Architect (Feature方案)] → [Auto-fix | Awaiting Confirm]

• Bug 路径: Scout 探索 → Detective 根因调查 → 自动修复
• Feature 路径: Scout 探索 → Architect 方案设计 → 等待确认后实现
• 6 Agent 协作: triage, scout, detective, architect, coder, tester

Superpower — Extensible Agent Capabilities

Load agents, skills, and commands from external superpowers repository

• Dynamic Agent Loading: Loads agent definitions from ~/.superpowers/agents/
• Skill Registry: Registers skills from ~/.superpowers/skills/
• Command Templates: Registers command templates from ~/.superpowers/commands/
• Extends OpenCode with custom agents and capabilities from the superpowers repo

Quick Start

1. Register plugins

In .opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    "./plugins/commander/index.ts",
    "./plugins/emperor/index.ts",
    "./plugins/hive/index.ts",
    "./plugins/superpower/index.ts",
    "./plugins/stock/index.ts",
    "./plugins/triage/index.ts"
  ]
}

Pick one or more.

2. Use

Commander:

@lead Implement user authentication with JWT and role-based access control

Emperor:

@taizi Implement user authentication with JWT and role-based access control

Hive:

@queen Implement user authentication with JWT and role-based access control

Superpower:

@<agent-name> Your task description here

The Superpower plugin loads agents from the superpowers repository. Use any agent defined there.

3. Configure (optional)

• Commander: .opencode/commander.json — config docs
• Emperor: .opencode/emperor.json — config docs
• Hive: .opencode/hive.json — design docs
• Superpower: No config needed (loads from ~/.superpowers/)
• Stock: .opencode/stock.json — config docs
• Triage: .opencode/triage.json — config docs

Project Structure

.
├── plugins/                              # Plugin directory
│   ├── commander/                       # Commander plugin
│   │   ├── index.ts                     # Entry point
│   │   ├── agents/                      # 4 agent definitions
│   │   ├── engine/                      # Pipeline, classifier, dispatcher
│   │   └── tools/                       # cmd_task, cmd_status, cmd_halt
│   ├── emperor/                         # Emperor plugin
│   │   ├── index.ts                     # Entry point
│   │   ├── agents/                      # 11 agent definitions
│   │   ├── engine/                      # Pipeline, recon, reviewer, dispatcher
│   │   ├── tools/                       # edict, memorial, halt
│   │   └── skills/                      # Built-in skills
│   ├── hive/                            # Hive plugin
│   │   ├── index.ts                     # Entry point
│   │   ├── discovery/                   # Domain auto-discovery
│   │   ├── agents/                      # Dynamic agent generator
│   │   ├── eventbus/                    # Event pub/sub system
│   │   ├── tools/                       # hive_emit, hive_status, hive_broadcast, etc.
│   │   └── hooks/                       # config, system-transform, file-watcher, autonomy
│   ├── superpower/                      # Superpower plugin
│   │   └── index.ts                     # Entry point (loads agents/skills/commands from ~/.superpowers/)
│   ├── stock/                            # Stock plugin
│   │   ├── index.ts                     # Entry point
│   │   ├── agents/                      # 5 agent definitions
│   │   └── skills/                      # Built-in skills
│   └── triage/                          # Triage plugin
│       ├── index.ts                     # Entry point
│       ├── agents/                      # 6 agent definitions
│       ├── engine/                      # Pipeline, classifier, dispatcher
│       └── tools/                       # jira_analyze, jira_implement, jira_status
├── package.json                         # Build tooling (private)
├── tsconfig.json                        # TypeScript config
├── .hive/                               # Hive runtime data (events.json, domains.json)
└── .github/workflows/
    ├── ci.yml                           # Type check + tests
    └── npm-publish.yml                  # Tag-based selective publish

Development

# Install dependencies
bun install

# Type check
bun run build

# Run tests
bun test

Publishing

Each plugin is published automatically to npm when changes are pushed to the main branch. The workflow detects which plugins have changed and publishes them with auto-incremented version numbers.

Configuration

Plugin publishing is configured in publish.config.json:

{
  "$schema": "./schemas/publish.config.json",
  "plugins": {
    "commander": { "name": "opencode-plugin-commander", "path": "plugins/commander" },
    "emperor": { "name": "opencode-plugin-emperor", "path": "plugins/emperor" },
    ...
  },
  "publish": { "autoBumpVersion": true, "versionBumpType": "patch" }
}

To add a new plugin:

1. Add the plugin directory under plugins/
2. Add its config to publish.config.json
3. Push to main — it will be auto-published

Manual Trigger

You can also manually trigger a publish via GitHub Actions workflow dispatch.

Package	npm
opencode-plugin-commander	Commander plugin
opencode-plugin-emperor	Emperor plugin
opencode-plugin-hive	Hive plugin
opencode-plugin-superpower	Superpower plugin
opencode-plugin-stock	Stock plugin
opencode-plugin-triage	Triage plugin

Tech Stack

• Runtime: Bun
• Language: TypeScript (strict mode)
• Plugin SDK: @opencode-ai/plugin
• Storage: JSON file persistence

License

MIT

---

中文版