ORCH — CLI runtime I built for orchestrating long-running coding agents with state machine + retries

[oxgeneral]

What I built

ORCH — a TypeScript CLI runtime for orchestrating teams of AI coding agents.

Context for PromptFlow users

PromptFlow excels at building and testing LLM application flows with DAGs. ORCH fills a different gap: runtime orchestration for coding agents that execute as long-running OS processes (Claude Code, Cursor IDE, OpenCode).

The core problem: coding agents are stateful, long-running, and fail unpredictably. PromptFlow's DAG model works well for prompt chains — ORCH handles the messier reality of external agent processes.

What ORCH adds

PID-based process tracking — knows when agents stall vs just running slowly
Auto-retry with state — crashed agent → retrying, restarted automatically
Mandatory review gate — in_progress → review → done, can't skip
Shared context — all agents read from the same typed key-value store

npm install -g @oxgeneral/orch
orch agent add "Code Generator" --adapter claude
orch agent add "Code Reviewer" --adapter claude  
orch task add "Add pagination to /users endpoint" --assignee agt_codegen
orch tui  # live dashboard

Details

• 1647 tests, TypeScript strict, MIT
• Adapters: Claude Code, OpenCode, Cursor, shell, Codex

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