Revert "refactor: remove unused and methods from (#5172)"

[lucasgomide]

This reverts commit bb9bcd6.

The AgentExecutor is a Flow — its execution is driven by state transitions defined via @router decorators. Here's exactly what happens:

The deleted check_native_todo_completion was the router for the execute_native_tool state:

@router(execute_native_tool)          # ← routes FROM this state
def check_native_todo_completion(
    self,
) -> Literal["todo_satisfied", "todo_not_satisfied"]:

When the agent uses native tool calling (which gpt-5.4-mini does), the execution path is:

call_llm_native_tools → "native_tool_calls" → execute_native_tool → ???

With check_native_todo_completion present, the flow continues:

execute_native_tool → check_native_todo_completion → "todo_satisfied"/"todo_not_satisfied" → ... → eventually  AgentFinish

With it deleted, the Flow has no router registered for the execute_native_tool output. The flow simply stops — it has nowhere to go. At that point, state.current_answer is still an AgentAction (a tool call), not an AgentFinish, so the check at line 2591 raises:

if not isinstance(formatted_answer, AgentFinish):
    raise RuntimeError("Agent execution ended without reaching a final answer.")

The _extract_tool_name helper was also deleted — it's used by other routing logic to figure out which tool was called. Without it, any code path that tries to identify tool calls by name would also break.

In short: deleting these methods broke the Flow's state machine by removing a critical routing node. The agent successfully called a tool, but the Flow had no transition defined for what to do after the tool executed, so it silently exited the loop without a final answer.

---

Note

Medium Risk
Touches AgentExecutor flow routing for the native tool-calling path; mistakes here could cause executions to stop early or mark todos incorrectly, but the change is small and covered by a targeted unit test.

Overview
Restores missing AgentExecutor Flow nodes for the native tool-calling execution path by re-adding _extract_tool_name and a @router(execute_native_tool) method (check_native_todo_completion).

This ensures that after execute_native_tool runs, the Flow can transition to todo_satisfied/todo_not_satisfied (instead of dead-ending) when todos are active. Adds a regression test validating check_native_todo_completion behavior with and without an active todo.

^{Written by Cursor Bugbot for commit f121f79. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Reverted method _extract_tool_name is unused dead code

Low Severity

The _extract_tool_name method is defined but never called anywhere in the codebase — not in production code, not in tests. The similar utility extract_tool_call_info (imported and used throughout the file) already handles tool call info extraction. This is dead code that adds maintenance burden. The original PR (#5172) correctly identified it as unused; reverting it reintroduces the same dead code.