Package Exports
- opencode-orchestrator
Readme
OpenCode Orchestrator 🎯
Autonomous Multi-Agent Plugin for OpenCode
Why I Built This 🤔
I was frustrated coding with DeepSeek and Z.AI.
I wanted to achieve the same quality of work as premium models like Gemini 3 Pro and Claude Opus, but with affordable models. For developers who chose budget-friendly subscriptions instead of expensive plans, I built a multi-agent system that can autonomously complete complex engineering tasks even with mid-tier LLMs.
"Intelligence is a resource. Orchestrate it."
What is this?
A 5-agent autonomous architecture designed to solve complex engineering tasks with high reliability, even on mid-range LLMs.
Core Philosophy: Intelligence is a resource. We orchestrate that resource through Phase-based Workflows and Mandatory Environment Scans to ensure code always fits the project's infrastructure.
Key Features
- 🎯 Autonomous Loop — Commander runs relentlessly until the mission is complete.
- 🔍 Environment Scan — Mandatory analysis of Infra (Docker/OS), Stack, and Domain before any code change.
- 🔨 Smart Implementation — Builder matches existing codebase patterns exactly.
- 🛡️ Rigorous Audit — Inspector proves success with environment-specific evidence (Builds/Tests/Logs).
- 💾 Persistent Context — Recorder saves session state to disk, enabling resume at any time.
- 🏗️ Parallel Tasking — Architect splits work into concurrent DAG groups.
Installation
npm install -g opencode-orchestratorRestart OpenCode after installation.
Usage
🚀 Method 1: Select Commander via Tab Key (Recommended)
In OpenCode, press Tab to open the Agent selection menu. Select Commander and type your mission!
Press Tab to select Commander
"Fix the login bug in the docker-compose environment"The Commander will:
- Survey: Scan the Docker environment and codebase.
- Plan: Break the fix into steps.
- Execute: Call Builder to fix while matching patterns.
- Verify: Run builds/tests to prove the fix works.
- Complete: Report results with concrete evidence.
📋 Method 2: Use /task Command
/task "Implement user authentication with JWT"💡 Tip: Using the
/taskcommand makes Commander mode run 2x longer. Use/taskfor complex tasks that need extended processing!
Agents (5-Agent Architecture)
| Agent | Emoji | Role | Responsibility |
|---|---|---|---|
| Commander | 🎯 | Orchestrator | Autonomous mission control & delegation |
| Architect | 🏗️ | Planner | Task decomposition & strategy correction |
| Builder | 🔨 | Developer | Full-stack implementation (Logic + UI) |
| Inspector | 🔍 | Quality | 5-point audit & automatic bug fixing |
| Recorder | 💾 | Context | Persistent environment & progress tracking |
The Workflow (Progressive Phases)
- Phase 0: Triage (Smart)
- Fast Track 🟢: Simple fixes → Execute instantly (Skip heavy scans).
- Deep Track 🔴: Complex features → Full Environment Scan & Plan.
- Phase 1: Environment Scan
- Mandatory for Deep Track: Infra/Domain/Stack analysis.
- Phase 2: Parallel Planning
- Architect creates a DAG of atomic tasks (Scalable Planning).
- Phase 3: Execution & Audit
- Builder writes code ↔ Inspector verifies with evidence.
- Phase 4: Completion
- Mission Complete reported with proof of build/test success.
Uninstall
npm run dev:uninstall
# OR
npm uninstall -g opencode-orchestratorDocs
- Architecture & Design — Detailed system design and agent protocols
- Plugin Troubleshooting — Setup and common issues
License
MIT License. LICENSE
Reliability over slop. Environment over assumptions.