Applies to:
- Plan -
- Deployment -
Summary
Runs can fail withArgument list too long when the CLI passes the entire remote eval request to a fresh runner via a single environment variable.
What is happening
Thebt eval launcher hands the full remote eval request to a fresh runner process in one environment variable. Linux imposes a per-environment-variable size cap of roughly 128 KB. If your request or config exceeds that cap, the runner process fails to start and the run shows Argument list too long or a similar error.
Fix or suggestion
Use the braintrust SDK (fastest workaround)
- Run your remote eval with the
braintrust eval --devcommand - Verify runs complete without the startup error
How to confirm it worked
- Run the same remote eval that previously failed and confirm it completes without
Argument list too longerrors. - Verify the process exit code is zero and results are produced as expected.