Add a new flag to disable checks for system configuration

[benglewis]

Today with AI agentic tools like Claude Code, Cursor's Agents, OpenAI's Codex and Gemini's CLI, there are lots of tools that will run Pixi sandboxed such that it can only access the current directory. The current Pixi implementation doesn't work well with this since it always tries to read the system configuration which causes a crash due to it trying to escape the sandbox. Some kind of flag / environment variable to disable this behavior would be very helpful

[baszalmstra]

Do you have a suggestion?

[benglewis]

Maybe something like SANDBOX_PIXI. It should be relatively easy to tell an agent via AGENTS.md to always pass in this environment variable. Another alternative would be to have a configuration that you could place in a pixi.toml / pyproject.toml to skip reading the system configuration, but that might be less desirable. I actually wonder how tools like node and uv don't encounter this problem... I may need to do some more digging

[benglewis]

This is what ChatGPT had to say about how node and uv handle things:
https://chatgpt.com/share/69305fc1-8f58-800c-a204-52566f95c40a

[benglewis]

This seems to actually be related to this ticket: #5173