fix: rebuild bwrap sandbox after cwd change

[Ray0907]

Summary

• key the cached Bubblewrap sandbox by its resolved workspace
• rebuild the jail after /cwd switches to a different workspace
• clean up the retired sandbox and cover reuse, replacement, and cleanup with a regression test

Background

_get_bwrap_sandbox() previously cached one process-wide jail. After /cwd changed the session workspace, later shell commands reused a jail whose writable bind still pointed at the original directory.

Validation

• uv run pytest -q — 1722 passed
• uv run ruff check frontier_agent/ apodex/ benchmarks/ workflows/ plugins/ deploy/ tools/ scripts/
• uv run pyright — 0 errors
• git diff --check

[zhanghanduo]

The fix correctly rebuilds the cached Bubblewrap sandbox after /cwd switches to a different workspace, while preserving reuse for equivalent paths and cleaning up the retired sandbox. The regression test covers the expected behavior well. LGTM!

[Copilot]

Pull request overview

Fixes an issue where the Bubblewrap (bwrap) sandbox instance was cached process-wide and incorrectly reused after the session working directory changed, causing writable binds to point at the wrong workspace.

Changes:

• Key the cached bwrap sandbox by the resolved cwd and rebuild it when the resolved workspace changes.
• Dispose of the previously cached sandbox when replacing it.
• Add a regression test covering reuse for equivalent paths, replacement on workspace change, and cleanup behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
apodex/sandbox.py	Cache bwrap sandbox by resolved workspace and rebuild/cleanup when cwd switches.
apodex/tests/test_native.py	Add regression test ensuring sandbox reuse for the same workspace and rebuild on workspace change.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.