Fix Docker sandbox filesystem tool fallback

[penso]

Summary

• Add regression coverage for sandboxed Read/Write/Edit/MultiEdit on /home/sandbox and workspace/data paths
• Fall back from translated Docker host paths to container copy when the gateway process cannot access the host mount
• Improve empty docker/OCI copy failure diagnostics and clarify sandbox path guidance in fs tool descriptions
• Closes [Bug]: Read/Write/Edit tools don't work in Docker #1096

Validation

Completed

• cargo test -p moltis-tools fs::contract_tests::sandbox_
• cargo test -p moltis-tools sandbox::tests::core::test_docker_write_file_falls_back_to_container_copy_when_host_mount_is_inaccessible
• cargo test -p moltis-tools sandbox::tests::core::test_docker_read_file_uses_mounted_workspace_path
• cargo test -p moltis-tools sandbox::tests::core::test_docker_write_file_uses_mounted_workspace_path
• cargo test -p moltis-tools sandbox::file_system::tests::oci_copy_failure_detail_is_not_empty_when_stderr_is_empty
• cargo fmt --all -- --check

Remaining

• Full just lint / just test not run locally

Manual QA

• Not run locally; Docker-in-Docker behavior is covered with a fake OCI CLI regression test that forces host mount access failure and verifies fallback to container copy.

[greptile-apps]

Greptile Summary

This PR fixes the Docker sandbox filesystem fallback so that write_file and list_files catch host-mount access errors and retry via container copy, and updates read_file to fall back on PermissionDenied/NotRegularFile host results in addition to NotFound. It also improves OCI copy error messages when stderr is empty and adds sandbox path guidance to all four fs tool descriptions.

• Write/list fallback: native_host_write_file and native_host_list_files errors are now caught and logged; execution falls through to the container path instead of propagating. A fake-CLI test verifies the write path; list_files has no equivalent regression test yet.
• Read fallback condition: Condition changed from !NotFound → return to Ok | TooLarge → return, so PermissionDenied and NotRegularFile host results also fall through to the container read.
• Diagnostics: New container_copy_failure_detail helper produces a non-empty message when the OCI CLI exits with no stderr output, fixing silent/blank error strings.

Confidence Score: 4/5

Safe to merge; the fallback logic is correct and the new tests cover the primary regression cases.

The write and list fallbacks are well-implemented and the write path has a focused regression test. The read path relies implicitly on native_host_read_file never returning Err for file-access failures, creating a silent coupling that differs from the write/list pattern. The list_files fallback branch is also untested.

crates/tools/src/sandbox/docker.rs — the read_file error-propagation asymmetry and the untested list_files fallback branch both warrant a second look.

Important Files Changed

Filename	Overview
crates/tools/src/sandbox/docker.rs	Adds error-catching fallback to write_file and list_files when host-mount access fails; changes read_file condition from !NotFound to explicit Ok
crates/tools/src/sandbox/file_system.rs	Adds container_copy_failure_detail helper to produce non-empty error messages when OCI cp exits with no stderr; logic and test are correct.
crates/tools/src/sandbox/tests/core.rs	Adds write-fallback regression test using a fake CLI script; no equivalent test for list_files fallback.
crates/tools/src/fs/contract_tests.rs	Adds four new sandbox contract tests covering /home/sandbox/ and workspace paths for Write, Read+Write, Edit, and MultiEdit through MockSandbox.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[fs operation] --> B{mounted_host_path?}
    B -- None --> G[container op]
    B -- Some --> C[native_host_* call]
    C -- read Ok/TooLarge --> D[return host result]
    C -- read NotFound/PermissionDenied/NotRegularFile --> F[debug log, fallback]
    C -- read Err via ? --> E[return Err, no fallback]
    C -- write Ok --> D
    C -- write Err --> F
    C -- list Ok --> H[remap and return]
    C -- list Err --> F
    F --> G
    G --> I[return container result]
    style E fill:#f99,stroke:#c00
    style F fill:#ffe,stroke:#cc0

Loading

Comments Outside Diff (1)

1. crates/tools/src/sandbox/docker.rs, line 766-772 (link)

   Read-path error not caught by fallback

   write_file and list_files now explicitly catch errors from their respective native_host_* helpers and fall back to container operations. read_file still uses .await?, so any Err from native_host_read_file propagates immediately without attempting the container fallback. This works today because native_host_read_file converts every IO error into an Ok(SandboxReadResult::…) variant rather than returning Err. If that mapping ever gains an error path, the read fallback will silently break while write and list continue to work.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

_{Reviews (1): Last reviewed commit: "fix(sandbox): route docker fs tools thro..." | Re-trigger Greptile}

[greptile-apps]

No test for list_files host-mount fallback

The analogous write_file fallback has a dedicated regression test (test_docker_write_file_falls_back_to_container_copy_when_host_mount_is_inaccessible), but the list_files fallback added here has no corresponding test. native_host_list_files returns Err on non-NotFound IO errors (e.g., permission denied on the directory), so a similar fake-CLI test that sets host_data_dir to a non-existent path and asserts the container list_files call is reached would cover the new branch.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 39 untouched benchmarks
⏩ 5 skipped benchmarks¹

---

_{Comparing shining-roof (36be45e) with main (a568a3c)}

Footnotes

1. 5 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩