PR #15830: fix(reasoning): promote reasoning field before empty-pad for DeepSeek/Kimi (#15812)

• Repository: NousResearch/hermes-agent
• Author: briandevans
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/15830

Description (problem / solution / changelog)

Summary

• _copy_reasoning_content_for_api had a step-ordering bug introduced by PR #15749: the tool-call empty-string injection (old step 2) ran before the reasoning field promotion (old step 3)
• A tool-call message with a valid reasoning field but no reasoning_content would hit the empty-string path, get reasoning_content="" injected, and return early — the reasoning value was silently discarded
• Fix: move the reasoning → reasoning_content promotion to step 2, before the empty-string fallbacks

The bug

run_agent.py — _copy_reasoning_content_for_api (pre-fix):

# Step 1: explicit reasoning_content → preserve ✓
# Step 2: tool_calls + DeepSeek/Kimi → inject ""  ← fires here
#         RETURNS EARLY
# Step 3: reasoning field → promote               ← never reached

A message like {"role": "assistant", "reasoning": "thought trace", "tool_calls": [...]} on a DeepSeek session would return reasoning_content="" instead of reasoning_content="thought trace".

The fix

Reorder steps 2 and 3 so reasoning promotion runs first:

# Step 1: explicit reasoning_content → preserve
# Step 2: reasoning field present → promote to reasoning_content  ← moved up
# Step 3: tool_calls + DeepSeek/Kimi + no reasoning → inject ""
# Step 4: all other DeepSeek/Kimi + no reasoning → inject ""
# Step 5: non-string cleanup

Empty-string injection now only fires when neither reasoning_content nor reasoning is present — the poisoned-history case it was designed for.

Test plan

• Before: test_deepseek_reasoning_field_promoted → AssertionError: assert '' == 'thought trace'
• After: all 21 tests in test_deepseek_reasoning_content_echo.py pass
• Regression guard: reverted run_agent.py → test_deepseek_reasoning_field_promoted failed with '' != 'thought trace'; restored → 21 passing
• Broader suite: tests/run_agent/ — 1048 passed, 1 pre-existing baseline failure (test_inf_stays_string_for_integer_only reproduces on clean origin/main)

Related

• Fixes #15812
• Regression introduced by #15749
• Related: #15748 (same code path)

🤖 Generated with Claude Code

Changed files

• run_agent.py (modified, +14/-12)

PR #15883: fix(run_agent): prevent reasoning_content regression in DeepSeek/Kimi tool-call replay (#15812)

• Repository: NousResearch/hermes-agent
• Author: yes999zc
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/15883

Description (problem / solution / changelog)

Problem

PR #15478 fixed missing reasoning_content for DeepSeek API calls, but introduced a regression in the logic order of _copy_reasoning_content_for_api.

Regression Detail

When a message contains:

• role: assistant
• tool_calls (non-empty)
• reasoning: "some genuine reasoning" (from a prior provider that uses the internal reasoning key)
• No reasoning_content

The old (buggy) order was:

1. Preserve explicit reasoning_content → skip (not present)
2. DeepSeek/Kimi tool-call check → inject "" → return → reasoning never promoted
3. Promote reasoning field → never reached

Result: genuine reasoning content is lost, replaced by empty string.

Fix

Swap steps 2 and 3 so promotion happens before the empty-string fallback:

1. Preserve explicit reasoning_content
2. Promote 'reasoning' field (MOVED UP) ← fix
3. DeepSeek/Kimi tool-call empty-string fallback (MOVED DOWN)
4. Non-thinking provider cleanup

Verification

pytest tests/run_agent/test_deepseek_reasoning_content_echo.py -v: 21/21 passed

Key regression test: test_deepseek_reasoning_field_promoted → PASSED

References

• Fixes #15812
• Relates #15749, #15478

Checklist

• Bug fix (non-breaking change)
• Tests pass (21/21)
• Only run_agent.py modified (single-file change)
• Comments updated to explain ordering rationale

Changed files

• run_agent.py (modified, +11/-8)