Opencode Agentic Workflows

Multi-Agent OpenCode Workflows

**Turn OpenCode into a disciplined multi-agent engineering system.**

This repo gives you a ready-to-adapt control plane for OpenCode agents:

• one user-facing orchestrator instead of agent chaos
• structured planning with tracks, epics, readiness gates, and plan deltas
• hidden specialist workers for coding, review, debugging, and discovery
• independent acceptance verification after review
• multi-model consensus review
• reusable internal skills instead of bloated prompts
• template-based durable memory that downstream projects can safely adopt

**Clone it, adapt it, and use it as a serious foundation for agentic delivery — not a demo prompt pack.**

Pattern

This is a **supervisor-worker** multi-agent system, not a peer-to-peer hive.

• One primary agent (orchestrator) is user-facing — the main entrypoint.
• Thirteen subagents are invoked programmatically by the orchestrator via the Task tool.
• planner is a hidden subagent — invoked by orchestrator via Task tool for planning tasks.
• All other subagents are hidden and cannot call each other; flow is strictly hierarchical: orchestrator → {planner, coder, designer, debugger, explore, reviewer-a/b/c, multi-reviewer, verifier, improvement-analyst, improvement-evaluator}.

What this is **not**:

• Not a decentralized hive-mind. There is one control plane.
• Not peer-to-peer. Agents do not negotiate with each other.
• Not self-organizing. The orchestrator explicitly routes by task type and planning track.

What this **is**:

• A governed delegation tree with structured planning, independent review, and objective verification gates.
• Multi-model consensus for code review (reviewer-a + reviewer-b + reviewer-c → multi-reviewer).

---

Based on:

• burkeholland gist
• simkeyur/vscode-agents
• [github/awesome-copilot](https://github.com/github/awesom