OPC — многопроцессная оркестрация агентов Claude Code

orchestration

opc-architecture is a Claude Code skill that documents the architecture of OPC (Orchestrated Parallel Claude), a system that extends Claude Code through hooks, skills, and scripts without replacing it. Orchestration is layered on top of Claude Code's CLI engine: hooks intercept lifecycle events such as PreToolUse and PostToolUse, Python scripts handle coordination logic, and PostgreSQL persists state across Claude Code instances. Each agent is simply a separate `claude -p` process spawned via `subprocess.Popen`, with the parent instance tracking child PIDs in the database and exchanging results through files written to `.claude/cache/agents/`. Key directories include `.claude/hooks/`, `.claude/skills/`, and `opc/scripts/`, while infrastructure runs PostgreSQL, Redis, and PgBouncer via docker-compose. This skill is aimed at developers building parallel multi-agent workflows on top of Claude Code who need a clear mental model of how inter-process coordination actually works.