Resource suggestion: ORCH — CLI runtime for coordinating multi-agent AI engineering teams

[oxgeneral]

Turn Claude, Cursor, and Codex into one team.

I built ORCH — an open-source CLI runtime that orchestrates AI coding agents (Claude Code, OpenCode, Codex, Cursor) as a typed engineering team with a formal state machine, auto-retry, and inter-agent messaging.

Relevance to prompt engineering practitioners:

Multi-agent prompt engineering requires more than just crafting individual agent prompts — it requires coordination. How do you ensure agents hand off context reliably? How do you prevent silent failures? ORCH solves this at the infrastructure layer.

ORCH implements proven patterns for multi-agent prompt engineering:

• Shared context store: orch context set key value. Agents read upstream prompt results without re-prompting
• Inter-agent messaging: orch msg send agent-id. Structured handoffs between specialist agents
• Mandatory review gate: no task completes without going through a review agent
• State machine: todo -> in_progress -> review -> done. Every agent output is tracked

This means you can build prompt chains where:

1. A researcher agent gathers data
2. An analyst agent processes it
3. A writer agent synthesizes results
4. A reviewer agent validates output

All coordinated automatically without manual orchestration.

npm install -g @oxgeneral/orch

1493 tests, TypeScript strict, MIT license.

GitHub: https://github.com/oxgeneral/ORCH