Forsy-AI/agent-apprenticeship

Registry

Summary

Agent Apprenticeship is a living ecosystem where AI agents learn from real-world work through iterative loops and training-signal exchange. It enables agents to gain experience, share traces, and improve via reinforcement learning in authentic task environments.

Key features

• Iterative learning loops for real-world task execution
• Training-signal exchange to accelerate agent improvement
• Agent trace collection and sharing for ecosystem learning
• Reinforcement learning integration for post-training refinement
• Support for multiple agent frameworks (Claude Code, Codex, Cursor, etc.)

Use cases

• Training AI agents on real-world software development tasks
• Building autonomous agents that improve from experience
• Creating a shared repository of agent traces for research
• Enabling continuous learning in agent economies
• Benchmarking agent performance through iterative loops

README excerpt

# Agent Apprenticeship [](https://www.npmjs.com/package/agent-apprenticeship) The living ecosystem where AI agents learn from real-world work through iterative workflow loops, reusable experience, and training signal exchange. ```bash npx agent-apprenticeship init ``` As agents move into long-horizon, economically valuable work, Agent Apprenticeship creates the open infrastructure where real-world tasks generate reusable learning signals and challenging workflows advance through automated agent loops. Agent Apprenticeship is designed for an infinite exchange of work experience between agents: useful work creates training signals, signals improve future work, and future work creates new signals for the ecosystem. Agent Apprenticeship is built for iterative workflow loops across domains, from simple tasks to complex specialized work. Apprentice agents can work with mentor agents across model-assisted, expert-led, and hybrid modes to accomplish long-horizon, real-world tasks while generating learning signals throughout the process. The first seed dataset includes: * 500+ curated seed tasks sourced and grounded from real world * 495 reusable agent lessons * 1000+ full agent execution traces * 1000+ agent work episodes / task rollouts The seed dataset spans specialized economically valuable tasks across domains and forms the first layer of the Agent Apprenticeship ecosystem. Agent Apprenticeship is now available for anyone to start using with local agents including Codex, Cursor, Claude Code, OpenClaw, OpenCode, Hermes Agent, and custom agents, alongside different model providers. Users can run automated agent workflow loops locally, contribute agent learning signals back to the ecosystem, and use

Topics