OTel execute_tool bash span does not expose nonzero process exit outcome

Describe the observability gap

When the shell tool successfully launches the synthetic command false, the resulting execute_tool bash span has status code 0 and no exit-code, error, result, or outcome attribute. Consumers cannot distinguish exit 0 from a nonzero command outcome in metadata-only telemetry.

This is distinct from a tool invocation failure: reading a guaranteed nonexistent synthetic path correctly produced an execute_tool view span with error status, error.type=failure, and an exception event.

Affected version

GitHub Copilot CLI 1.0.83 on macOS.

Steps to reproduce

1. Enable metadata-only OTel file export.
2. Run a synthetic prompt that directs the shell tool to execute exactly false once.
3. Inspect the execute_tool bash span.

The tool execution output reports completion, while the span contains no metadata representing the subprocess's nonzero exit.

Expected behavior

It is reasonable for the tool invocation itself to be considered operationally successful, but metadata-only telemetry should expose a bounded, content-free command outcome such as process.exit.code=1 and/or a Copilot tool outcome attribute. The span status policy can remain separate from the subprocess exit status.

Privacy

The command is synthetic and harmless. Content capture was disabled; no command output, paths, repository data, or credentials are included.

Start by reproducing the metadata-only OTel export with the shell tool running the synthetic command false, then inspect the resulting execute_tool bash span. Done means the span exposes a bounded, content-free nonzero subprocess outcome such as process.exit.code=1 or an equivalent tool outcome attribute without capturing command content.