fix: prevent reasoning box from rendering 3x during tool-calling loops - #3405
Merged
Conversation
Two independent bugs caused the reasoning box to appear three times when the model produced reasoning + tool_calls: Bug A: _build_assistant_message() re-fired reasoning_callback with the full reasoning text even when streaming had already displayed it. The original guard only checked structured reasoning_content deltas, but reasoning also arrives via content tag extraction (<REASONING_SCRATCHPAD>/<think> tags in delta.content), which went through _fire_stream_delta not _fire_reasoning_delta. Fix: skip the callback entirely when streaming is active — both paths display reasoning during the stream. Any reasoning not shown during streaming is caught by the CLI post-response fallback. Bug B: The post-response reasoning display checked _reasoning_stream_started, but that flag was reset by _reset_stream_state() during intermediate turn boundaries (when stream_delta_callback(None) fires between tool calls). Introduced _reasoning_shown_this_turn flag that persists across the tool loop and is only reset at the start of each user turn. Live-tested in PTY: reasoning now shows exactly once per API call, no duplicates across tool-calling loops.
teknium1
force-pushed
the
hermes/hermes-caea5692
branch
from
March 27, 2026 16:54
9238b2f to
499c734
Compare
angelburgosrosado
pushed a commit
to angelburgosrosado/hermes-agent
that referenced
this pull request
Apr 27, 2026
NousResearch#3405) Two independent bugs caused the reasoning box to appear three times when the model produced reasoning + tool_calls: Bug A: _build_assistant_message() re-fired reasoning_callback with the full reasoning text even when streaming had already displayed it. The original guard only checked structured reasoning_content deltas, but reasoning also arrives via content tag extraction (<REASONING_SCRATCHPAD>/<think> tags in delta.content), which went through _fire_stream_delta not _fire_reasoning_delta. Fix: skip the callback entirely when streaming is active — both paths display reasoning during the stream. Any reasoning not shown during streaming is caught by the CLI post-response fallback. Bug B: The post-response reasoning display checked _reasoning_stream_started, but that flag was reset by _reset_stream_state() during intermediate turn boundaries (when stream_delta_callback(None) fires between tool calls). Introduced _reasoning_shown_this_turn flag that persists across the tool loop and is only reset at the start of each user turn. Live-tested in PTY: reasoning now shows exactly once per API call, no duplicates across tool-calling loops.
02356abc
pushed a commit
to 02356abc/hermes-agent
that referenced
this pull request
May 14, 2026
NousResearch#3405) Two independent bugs caused the reasoning box to appear three times when the model produced reasoning + tool_calls: Bug A: _build_assistant_message() re-fired reasoning_callback with the full reasoning text even when streaming had already displayed it. The original guard only checked structured reasoning_content deltas, but reasoning also arrives via content tag extraction (<REASONING_SCRATCHPAD>/<think> tags in delta.content), which went through _fire_stream_delta not _fire_reasoning_delta. Fix: skip the callback entirely when streaming is active — both paths display reasoning during the stream. Any reasoning not shown during streaming is caught by the CLI post-response fallback. Bug B: The post-response reasoning display checked _reasoning_stream_started, but that flag was reset by _reset_stream_state() during intermediate turn boundaries (when stream_delta_callback(None) fires between tool calls). Introduced _reasoning_shown_this_turn flag that persists across the tool loop and is only reset at the start of each user turn. Live-tested in PTY: reasoning now shows exactly once per API call, no duplicates across tool-calling loops.
gweeteve
pushed a commit
to gweeteve/hermes-agent
that referenced
this pull request
Jun 2, 2026
NousResearch#3405) Two independent bugs caused the reasoning box to appear three times when the model produced reasoning + tool_calls: Bug A: _build_assistant_message() re-fired reasoning_callback with the full reasoning text even when streaming had already displayed it. The original guard only checked structured reasoning_content deltas, but reasoning also arrives via content tag extraction (<REASONING_SCRATCHPAD>/<think> tags in delta.content), which went through _fire_stream_delta not _fire_reasoning_delta. Fix: skip the callback entirely when streaming is active — both paths display reasoning during the stream. Any reasoning not shown during streaming is caught by the CLI post-response fallback. Bug B: The post-response reasoning display checked _reasoning_stream_started, but that flag was reset by _reset_stream_state() during intermediate turn boundaries (when stream_delta_callback(None) fires between tool calls). Introduced _reasoning_shown_this_turn flag that persists across the tool loop and is only reset at the start of each user turn. Live-tested in PTY: reasoning now shows exactly once per API call, no duplicates across tool-calling loops.
waefrebeorn
pushed a commit
to waefrebeorn/slermes
that referenced
this pull request
Jul 2, 2026
NousResearch#3405) Two independent bugs caused the reasoning box to appear three times when the model produced reasoning + tool_calls: Bug A: _build_assistant_message() re-fired reasoning_callback with the full reasoning text even when streaming had already displayed it. The original guard only checked structured reasoning_content deltas, but reasoning also arrives via content tag extraction (<REASONING_SCRATCHPAD>/<think> tags in delta.content), which went through _fire_stream_delta not _fire_reasoning_delta. Fix: skip the callback entirely when streaming is active — both paths display reasoning during the stream. Any reasoning not shown during streaming is caught by the CLI post-response fallback. Bug B: The post-response reasoning display checked _reasoning_stream_started, but that flag was reset by _reset_stream_state() during intermediate turn boundaries (when stream_delta_callback(None) fires between tool calls). Introduced _reasoning_shown_this_turn flag that persists across the tool loop and is only reset at the start of each user turn. Live-tested in PTY: reasoning now shows exactly once per API call, no duplicates across tool-calling loops.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
When the model produces reasoning + tool_calls (e.g.
Okay merge that→ model reasons → calls terminal), the reasoning box rendered three times with identical content:_build_assistant_messagere-firing the callbackRoot Cause
Bug A:
_build_assistant_message()(run_agent.py:4637) always re-firedreasoning_callbackfor structured reasoning (reasoning_contentfield), even when streaming had already displayed it via_fire_reasoning_delta().Bug B: The post-response reasoning display (cli.py:5742) checked
_reasoning_stream_started, but that flag was reset toFalseby_reset_stream_state()during the intermediate turn boundary (whenstream_delta_callback(None)fires between tool calls). So after the tool loop completed, the guard was ineffective.Fix
Bug A: Added
_reasoning_deltas_firedflag on the agent that tracks whether reasoning was already displayed during the current streaming API call. Reset at the start of each streaming call (chat completions, Codex, Anthropic paths)._build_assistant_messagechecks this flag and skips the callback when reasoning was already streamed.Bug B: Introduced
_reasoning_shown_this_turnflag on the CLI that persists across_reset_stream_state()calls (tool loop iterations). Only reset at the start of each user turn. Post-response code checks this instead of the per-stream-call flag.Tests
Added 7 new tests in
test_reasoning_command.py:TestReasoningDeltasFiredFlag: verifies the flag is set, callback is skipped/fired correctlyTestReasoningShownThisTurnFlag: verifies the turn-persistent flag survives_reset_stream_stateFull suite: 6455 passed, 0 failed.