bytedance / bytedance/agentkit-samples
agentkit沙箱工具(skill_sandbox)的疑问
- Dominant language
- Python
- Stars
- 454
- Forks
- 96
- PR merge metrics
- No merged PRs in 30d
Description
看后台对接的aio sandbox,现有几个疑问:
1. aio sandbox原生是具备 bash、code 运行接口的,然而 agentkit的沙箱工具(无论是aio sandbox 还是 skill_sandbox),都只配了run_code工具,无法执行bash,这显然不合理,如果我skill中有命令行交互怎么办? 我的理解是,aio sandbox原来接口咋样,agentkit的原生工具就该咋样
2. 我采用skill_sanbox 配合 execute_skill工具,看传参形式像是execute_skill内部也搞了个智能体,这就很奇怪,不应该是沙箱暴露接口为tool,并且外部挂载skill到沙箱目录下,主智能体直接交互吗?为什么又是在沙箱内部又要搞一个智能体,还需要与主智能体分开配置模型key?十分不合理
3. 我想实现skill规约下的agentic链路,现在想用veadk开发框架+agentkit相关工具资源实现,那到底该如何配置?
希望协助回答下以上几个问题呢
Contributor guide
Research direction
Start by reviewing the aio sandbox, skill_sandbox, run_code, bash, and execute_skill entry points, then compare their interfaces and model configuration with the VeADK and AgentKit usage described here. Done means providing a documented explanation of the sandbox tool behavior and a concrete configuration path for an agentic skill workflow.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- ai, tooling
- Issue type
- Documentation
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100