[codex] Run user shell commands in explicit remote environments

[rasmusrygaard]

Why

A turn can select a remote environment, but user shell commands could only run on the local host. Callers that know which selected environment the user intends to target need a way to execute the command there without silently falling back to local execution.

What changed

• Adds an optional environment_id to RunUserShellCommand; omitting it preserves the existing local app-server and TUI behavior.
• Resolves explicit IDs against the selected environments for the turn and requires the target to be remote.
• Runs the command through the remote environment exec backend using its working directory, shell, and configured shell environment policy.
• Preserves user-shell lifecycle behavior, including begin/end events, streamed output, active-turn reuse, standalone turns, persisted output, cancellation, and the existing timeout.
• Bounds remote read sizes, emitted deltas, and retained stdout, stderr, and aggregate output. Remote processes are terminated on cancellation, timeout, read errors, and backend failures.
• Records selection and setup failures in command history instead of falling back to local execution.

Environment selection policy remains with the caller: core honors an explicit ID but does not infer a default when multiple environments are selected.

Test coverage

• Adds an integration test that selects local and remote environments and verifies an explicitly targeted user shell command runs in the remote working directory.
• Adds coverage confirming an unknown environment ID records a failure and never executes the command locally.
• Existing local user-shell call sites and tests continue to use environment_id: None.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0968f33680

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Use remote shell info without local path probing

For remotes on another OS or with a shell path absent locally, this host-side detection can fall back to /bin/sh and send that argv remotely, so explicit remote ! commands fail despite valid remote shell info. Use ShellInfo/turn_environment.shell directly. guidance

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Decode remote output with the normal exec decoder

For remote Windows shells or commands emitting non-UTF-8, these lossy conversions bypass normal exec's smart decoder, so final stdout/stderr/history show replacement characters for supported code pages. Convert through StreamOutput<Vec<u8>>::from_utf8_lossy(). guidance

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Advance remote reads to the last consumed sequence

When remote output spans multiple reads, setting after_seq to read.next_seq skips the next chunk: the exec-server read API returns chunks with seq > after_seq, so a later chunk whose sequence equals the previous next_seq is filtered out. This drops stdout/stderr and streamed deltas for longer-running or chatty remote ! commands; store the last consumed chunk seq instead (or next_seq - 1).

Useful? React with 👍 / 👎.