[hwolff99]

What does this PR do?

This PR fixes README instructions that caused a launch command to fail configuration validation after changes in using
environment servers.

The old command manually bound the resources server. The new command uses
--agent-type langchain_deepagents_agent.

Checklist

• I have read the contributing guidelines.
• The change is focused; unrelated "drive-by" edits are tracked as separate issues/PRs.
• Tests added or updated and pass locally, or N/A for docs-only / non-code changes (so CI unit/server checks pass when applicable).
• Documentation added or updated, or N/A with justification.
• Pre-commit checks pass locally (pre-commit run --all-files) (so CI lint/format checks pass).
• New source files include the required Apache-2.0 SPDX header.
• All commits have DCO sign-off (git commit -s) (so the DCO check passes).

[qiaochuz-nv]

POR Test Summary — DeepAgents README Launch

Verdict: PASS — documented --agent-type flow validated end to end

The command form documented by this PR was exercised against a clean Gym checkout at a7ca33b:

gym env start --resources-server reasoning_gym --model-type inference_provider --agent-type langchain_deepagents_agent

Observed behavior:

• The command resolved a healthy 4-server graph: reasoning_gym, the composed DeepAgents instance, its legacy environment server, and policy_model.

• The bundled simple agent was removed; the existing environment server was repointed to the composed agent.

• The generic config's resources-server placeholder resolved to reasoning_gym, and model calls routed through policy_model.

• Static launch validation was clean and emitted no deprecation result.

• The former --config override was rejected or deprecated, and it failed under strict environment-server enforcement.

• All 5 original Reasoning Gym inputs produced real trajectories; deterministic verifier replay/controls passed, aggregate mean/reward was 1.0, capture artifacts were complete, and all servers stopped cleanly.

• QA testcase/check: langchain_deepagents_agent / README --agent-type composition

• QA run: 20261006T012157814144Z-c47a4502

• Targeted check duration: 1m 53s

• Execution: real inference-provider calls; no model stubs, source repair, or reduced dataset

This specifically validates that the README correction selects the intended generic agent composition instead of the stale config-override path.

[github-actions]

🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA-NeMo/Gym/actions/runs/37517405059

[github-actions]

🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA-NeMo/Gym/actions/runs/37518504361