Enterprise Multi-Agent Automation — Production Harness with Denbun, Retry & Self-Healing

What it does

This skill provides a production-grade orchestration framework for managing multiple AI agents working in parallel on a single codebase. It implements the "Denbun Protocol," a robust handoff system that externalizes agent state to prevent context loss during restarts or token compaction. It manages the entire lifecycle of an agentic pipeline, including task routing, error handling, and automated recovery.

Why use this skill

Building multi-agent systems is easy in a demo but hard in production. This skill solves common failure modes like rate-limiting, agent stalls, and messy handoffs. Instead of manually prompting agents, you get a repeatable harness that ensures reliability. It’s better than manual prompting because it adds a "self-healing" layer that detects if an agent is stuck and automatically resumes it, saving hours of manual monitoring.

Supported tools

• Claude Code: Primary orchestration and reasoning engine.
• Local LLMs (Gemma/Codex): For cost-efficient subtask routing.
• Tmux: For persistent session management and event-driven communication.
• Python/Bash: For the core logic of retries, guards, and logging.

Expected Output

The framework produces machine-readable YAML task files, versioned Markdown handoff documents (Denbun), and structured JSON log files that track token usage, latencies, and success rates across your fleet.