[rollout] feat: improve error messages for malformed tool calls

[xiefan46]

Split the catch-all exception handler in _call_tool() into specific error cases so the LLM receives actionable feedback for self-correction:

• Unknown function name: lists available tools
• Invalid JSON arguments: reports parse error with tool name
• Tool execution failure: includes tool name in error message

What does this PR do?

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, veomni, sglang, vllm, vllm_omni, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward, fully_async, one_step_off
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

Added CPU unit tests that call the real _call_tool method via mock.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)
• If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

[gemini-code-assist]

Code Review

This pull request improves the error handling in the ToolAgentLoop._call_tool method by adding explicit validation for tool names and JSON arguments, providing more descriptive error messages to the agent. It also includes a new set of unit tests to verify these error paths. A potential issue was identified where an exception during the tool release process in the finally block could mask the original execution error; it is recommended to wrap the cleanup logic in its own try-except block to ensure robust error reporting.

[gemini-code-assist]

The except Exception as e block catches all exceptions during tool creation and execution. However, the finally block at line 470, which calls await tool.release(instance_id), is executed before the return statement in the except block. If tool.release itself raises an exception, it will propagate and suppress the original error message being returned to the LLM. It is recommended to wrap the tool.release call in its own try-except block to ensure that cleanup failures do not mask tool execution errors.

Suggested change

	except Exception as e:
	logger.warning(f"Error when executing tool: {e}")
	return (
	ToolResponse(
	text=f"Error when executing tool: {e}",
	),
	0.0,
	{},
	)
	logger.warning(f"Error executing tool '{tool_name}': {e}")
	return ToolResponse(text=f"Error executing tool '{tool_name}': {e}"), 0.0, {}
	finally:
	if tool and instance_id:
	except Exception as e:
	logger.warning(f"Error executing tool '{tool_name}': {e}")
	return ToolResponse(text=f"Error executing tool '{tool_name}': {e}"), 0.0, {}
	finally:
	if tool and instance_id:
	try:
	await tool.release(instance_id)
	except Exception as release_error:
	logger.error(f"Failed to release tool '{tool_name}': {release_error}")