v0.3.0

Major Version Upgrade. The multi-agent foundation framework upgraded its abilities, and preliminary completion of the self-closed-loop on AWorld.

Important Changed

• (Add) The train module provided a unified API for agent training, running in code pattern.
• (Add) Supported agent to the agent server, such as Google A2A and Anthropic MCP, so endorsed streaming task, and provided new streaming task API
• (Add) Anthropic Skills ability is supported
• (Update) Context ability upgrade, supported context engineering

What's Changed

• AWorld training with VeRL, AReaL by @yuchengyue @yaoyueduzhen @rainsonGain in #517 #586 #590 @tallate in #528 @wangchunf in #543 #572
• AWorld agent server (experimental), supported A2A by @rainsonGain in #539 #570 @wangchunf in #576, supported to MCP @wangchunf in #560
• Skills ability support by @SzekiHou in #541 #585 @ahgpt in #542 @wuhulala in #573
• Context implement upgrade by @ahgpt in #525 #526 #532 @xiudongxu in #546 @wuhulala in #567 #571
• Streaming response API supported by @SzekiHou in #556
• Workspace support files up-download by @ahgpt in #561 #562 #564 #566
• Sandbox and MCP improve by @kevinfeng030 in #511 #536 #553 #581
• Trajectory improve by @SzekiHou in #550 #552 #577
• Trace improve by @wangchunf in #519 #533 asyncio monitor in #554 #582
• Agent improve by @rainsonGain in #551 @SzekiHou in #569
• Eventbus improve by @rainsonGain in #535
• Examples IMO trajectories and results by @JasonTse1989 in #514 #515 #516
• Examples add the visualwebarena example folder by @kevinhikali in #521
• Examples add Aworld tutorials of quick start by @rainsonGain in #559
• Some fix #512 #513 #518 #522 #524 #529 #530 #547 #574 #575 #584 #587

New Contributors

• @tomokinakamaru made their first contribution in #518
• @kevinhikali made their first contribution in #521
• @wuhulala made their first contribution in #567

Full Changelog: v0.2.8...v0.3.0