Forward output_jsonl_gcs to evaluation job

[simonrosenberg]

Summary

• add output_jsonl_gcs input to run-eval workflow
• forward to evaluation dispatch payload

Testing

• gh workflow run run-eval.yml --ref issue-236-output-jsonl -f benchmark=gaia -f eval_limit=1 -f sdk_ref=issue-236-output-jsonl -f benchmarks_branch=issue-236-output-jsonl -f eval_branch=issue-236-output-jsonl -f output_jsonl_gcs=gs://openhands-evaluation-results/eval-20622411875-claude-son_litellm_proxy-claude-sonnet-4-5-20250929_25-12-31-16-36.tar.gz -f reason="rerun latest gaia 1-image"
• gh workflow run run-eval.yml --ref issue-236-output-jsonl -f benchmark=commit0 -f eval_limit=1 -f sdk_ref=issue-236-output-jsonl -f benchmarks_branch=issue-236-output-jsonl -f eval_branch=issue-236-output-jsonl -f output_jsonl_gcs=gs://openhands-evaluation-results/eval-20622412428-claude-son_litellm_proxy-claude-sonnet-4-5-20250929_25-12-31-18-13.tar.gz -f reason="rerun latest commit0 1-image"
• gh workflow run run-eval.yml --ref issue-236-output-jsonl -f benchmark=swebench -f eval_limit=1 -f sdk_ref=issue-236-output-jsonl -f benchmarks_branch=issue-236-output-jsonl -f eval_branch=issue-236-output-jsonl -f output_jsonl_gcs=gs://openhands-evaluation-results/eval-20600678832-claude-son_litellm_proxy-claude-sonnet-4-5-20250929_25-12-30-16-22.tar.gz -f reason="eval-only swebench reuse eval-20600678832 (fix)"
• gh workflow run run-eval.yml --ref issue-236-output-jsonl -f benchmark=swebench -f eval_limit=1 -f sdk_ref=issue-236-output-jsonl -f benchmarks_branch=issue-236-output-jsonl -f eval_branch=issue-236-output-jsonl -f reason="infer+eval swebench smoke (pydantic 2.12)"

[enyst]

@simonrosenberg out of curiosity, what version of GPT-5 are you using, and is it with oh via SDK or with good ole’ V0?

[all-hands-bot]

[Automatic Post]: I have assigned @raymyers as a reviewer based on git blame information. Thanks in advance for the help!

[all-hands-bot]

[Automatic Post]: It has been a while since there was any activity on this PR. @simonrosenberg, are you still working on it? If so, please go ahead, if not then please request review, close it, or request that someone else follow up.

[enyst]

@simonrosenberg out of curiosity, what version of GPT-5 are you using, and is it with oh via SDK or with good ole’ V0?

Just to clarify for everyone stumbling upon this, Simon answered in private, it was GPT-5.2 medium reasoning.

I think we have some phrase telling the LLM to not use \n\n, but it's obviously not working. (the PR description had \n\n stuff all over)