Skip to content

About

Orchestrating Agents: From Hard Guards to Soft Constraints — A systematic guide to AI Agent orchestration

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Orchestrating Agents: From Hard Guards to Soft Constraints

中文 | English

GitHub Pages License: CC BY-SA 4.0 GitHub stars

A systematic guide to AI Agent orchestration — from infrastructure-level hard orchestration to prompt-level soft orchestration.

What Makes This Book Different?

This is not a project analysis report. It's a systematically integrated book that extracts key insights from multiple open-source orchestrator projects and production systems, organizes them by theme, and provides deep cross-project comparisons.

Part I: Hard Orchestration

Using infrastructure to keep agents running continuously:

  • Process Daemons: systemd, tmux, supervisor — keeping agents alive 24/7
  • Scheduling: cron, self-scheduling loops, bracket-paste injection — driving agent work cycles
  • Fault Recovery: double-layer Watchdog, progressive recovery, rapid crash protection — surviving failures
  • Communication: send-keys, bracket-paste, structured mail, file coordination — inter-agent messaging
  • Concurrency Isolation: git worktree, process isolation, file locks — preventing conflicts

Part II: Soft Orchestration

Using prompts, skills, and process specifications to shape agent behavior:

  • Prompt Engineering: rule blocks, mission injection, sprint-driven workflows
  • Skill Systems: reusable capability templates, MCP tool integration
  • Pipeline Orchestration: multi-stage flows with quality gates and escalation
  • Antipatterns: loop detection, laziness prevention, rule self-deletion guards

Part III: Practice & Evolution

  • Hands-on: Build your own orchestrator from scratch
  • Evolution Roadmap: From "barely running" to "intelligent self-healing"

Source Projects

Project Tech Stack Key Feature Link
Tmux-Orchestrator Bash + tmux + Claude CLI Minimal self-scheduling, hierarchical roles GitHub
agency-agents-zh Markdown Prompt + MCP 50+ agent multi-stage pipeline GitHub
Composio agent-orchestrator TypeScript + pnpm monorepo Orchestrator-Worker, git worktree GitHub
Overstory TypeScript/Bun + 11 runtimes Layered Watchdog, structured mail GitHub

Read Online

🌐 https://puzhre.github.io/agent-orchestration-book

Contributing

Contributions welcome! Please:

  1. Fork the repository
  2. Create a feature branch (git checkout -b feature/your-topic)
  3. Commit your changes with clear, atomic messages
  4. Open a Pull Request

Please ensure all cited content includes proper source links.

License

This book is licensed under CC BY-SA 4.0. You are free to share and adapt the material with attribution.


⭐ If this book helps you, please give it a star!

About

Orchestrating Agents: From Hard Guards to Soft Constraints — A systematic guide to AI Agent orchestration

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages