Bug: Malformed +config_paths surfaces raw Pydantic ValidationError stack #1490

Summary

Found by our virtual developer review agent.

When a user passes a malformed +config_paths value (e.g. a scalar string instead of a list), NeMo Gym raises a raw Pydantic ValidationError with a full traceback instead of a clear, actionable error message.

Location

nemo_gym/global_config.py — config parsing (~line 412)

ta = TypeAdapter(List[str])
config_paths = merged_config_for_config_paths.get(CONFIG_PATHS_KEY_NAME) or []
config_paths = ta.validate_python(config_paths)

Current behavior

Passing a scalar instead of a list produces a library-internal error with a full traceback:

pydantic_core.ValidationError: 1 validation error for list[str]
  Input should be a valid list

The message is technically correct but reads as an internal Pydantic failure (including a pydantic.dev URL in the traceback) and gives no guidance on the expected Hydra list syntax.

Expected behavior

A malformed +config_paths value should fail fast with a concise, user-facing message explaining the expected format — no traceback.

Suggested fix

Wrap validate_python in a try/except ValidationError block and re-raise a clean SystemExit with guidance on the expected Hydra list syntax:

try:
    config_paths = ta.validate_python(config_paths)
except ValidationError:
    raise SystemExit(
        "+config_paths must be a list, e.g. "
        '+config_paths=[a.yaml,b.yaml]'
    )

Catch this in the CLI entrypoint so a traceback is not printed.

Repro

# Scalar instead of list — exact Hydra syntax may vary, but any non-list value triggers this:
ng_run "+config_paths=resources_servers/example_single_tool_call/configs/example_single_tool_call.yaml"