Vouch request: kezouke

[kezouke]

What do you want to work on?

I’d like to work on #4046: #4046

The CLI checks that upload paths exist before creating the sandbox, but Git filtering and upload planning happen after it reaches Ready. If Git filtering rejects an upload, the sandbox remains running, and earlier uploads may already have completed.

I’d like to move planning for all uploads before sandbox creation and add regression tests, including a valid first upload followed by an invalid second one. I’d preserve the unfiltered upload option and symlink handling. Actual transfer failures after creation would still leave the sandbox available for retry.

Is this issue available, and would you accept that scope?

Why this change?

I'm new to Rust and haven't used OpenShell yet, but I've built AI agents for banks and Yandex and contributed to vLLM. What pulls me in is that "Open Agent Safety Platform" uses OpenShell for sandboxing while monitoring happens outside the agent itself, and I want to understand how that boundary really works under the hood. It also just feels cool to help with the trust layer for agents, and it would strengthen my portfolio too.

I code mostly in Python, with C++ and SQL for backend and data work. I've built production agents for banking and worked on recommendation and memory systems. I also contributed to vLLM, adding prompt-token support in Logits Processors. For testing and evaluation, I've built LLM-as-a-Judge frameworks and set up monitoring with Prometheus and Grafana.

Checklist

• I wrote this request myself (not AI-generated)
• I have read CONTRIBUTING.md