fix(tool-executor): set tool_name on sequential per-iteration interrupt skip messages

[EloquentBrush0x]

Summary

execute_tool_calls_sequential checks _interrupt_requested at the top of each loop iteration before starting the next tool. When triggered, it builds skip messages manually — without tool_name. PR #28914 fixed every other interrupt path in the file to use make_tool_result_message, but this per-iteration block was missed.

Before (manual dict, tool_name absent):

skip_msg = {
    "role": "tool",
    "name": skipped_name,
    "content": f"[Tool execution cancelled — {skipped_name} was skipped due to user interrupt]",
    "tool_call_id": skipped_tc.id,
}
messages.append(skip_msg)

After (symmetric with every other interrupt path):

messages.append(make_tool_result_message(
    skipped_name,
    f"[Tool execution cancelled — {skipped_name} was skipped due to user interrupt]",
    skipped_tc.id,
))

Coverage of all interrupt skip paths in tool_executor.py after this fix:

Path	Fixed in
execute_tool_calls_concurrent pre-flight	#28914
execute_tool_calls_sequential end-of-loop	#28914
execute_tool_calls_sequential per-iteration ←	this PR

Test plan

• New regression test TestPreToolCheck::test_sequential_per_iteration_interrupt_sets_tool_name asserts tool_name == name on all skip messages produced by the per-iteration check
• All existing interrupt tests pass (tests/tools/test_interrupt.py — 7 passed)
• Existing test_tool_name_db_persistence.py still passes (1 passed)
• make_tool_result_message was already imported; no new imports needed
• Zero remaining manual {"role": "tool", ...} constructions in tool_executor.py

[teknium1]

Thanks for the focused fix. I verified the premise against current main, and this still looks like a real gap.

The top-of-loop sequential interrupt path in agent/tool_executor.py:782 still manually constructs the skipped tool result without tool_name. The helper this PR switches to, make_tool_result_message, sets both name and tool_name at agent/tool_dispatch_helpers.py:337, matching the sibling interrupt paths already using it at agent/tool_executor.py:253 and agent/tool_executor.py:1396.

I do not see blockers in the implementation or the regression test. The change is narrow, fits the existing helper pattern introduced by #28914, and does not add new core surface or cache-sensitive behavior.

This is an automated hermes-sweeper review.

[teknium1]

Thanks for the focused regression report and test coverage.

This is an automated hermes-sweeper review. Current main already provides the requested guarantee:

• agent/tool_executor.py:1036 builds the sequential per-iteration interrupt skip result with make_tool_result_message(...).
• agent/tool_dispatch_helpers.py:389-395 sets both name and tool_name on that message.
• tests/run_agent/test_tool_name_db_persistence.py:28-45 already verifies helper-created tool results persist tool_name through the SessionDB flush path.
• The current implementation arrived with d682f320b35a13084371a541a835e1d988c982b8; the PR now conflicts with main.

Closing as implemented on main.