… thinking mode

DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:

    The reasoning content in the thinking mode must be passed back to the API.

run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.

Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.

Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).

Refs #15250, #17400.
Closes #17341.

… thinking mode

DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:

    The reasoning content in the thinking mode must be passed back to the API.

run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.

Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.

Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).

Refs NousResearch#15250, NousResearch#17400.
Closes NousResearch#17341.

… thinking mode

DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:

    The reasoning content in the thinking mode must be passed back to the API.

run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.

Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.

Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).

Refs NousResearch#15250, NousResearch#17400.
Closes NousResearch#17341.

DeepSeek thinking mode requires reasoning_content echoed back on every
assistant message with tool_calls. The checklist evaluator in /goal
(appends assistant msgs to judge history via raw dict construction)
was dropping reasoning_content, causing HTTP 400 on the next API call:

  The reasoning_content in the thinking mode must be passed back to the API.

Fixes the only unprotected code path — the main agent loop already has
two defenses (_build_assistant_message + _copy_reasoning_content_for_api)
but the goal judge uses a separate OpenAI client directly.

Refs NousResearch#17341

…provides it

Previous fix injected reasoning_content unconditionally (fallback ' ') for
every provider, adding a spurious DeepSeek/Kimi-specific field to Claude,
Gemini, and other providers.  Now the key is only added when the SDK
response carries it — absent on non-thinking providers, always present
(possibly empty) on DeepSeek/Kimi.

Empty string upgraded to ' ' for DeepSeek V4 Pro which rejects empty
reasoning_content with HTTP 400.

Refs NousResearch#17341

OpenRouter uses 'reasoning' + 'reasoning_details' (in model_extra) for
DeepSeek thinking tokens — different field names from the direct-API
'reasoning_content'.  The OpenRouter docs require echoing these back
for multi-turn reasoning continuity.

Now preserves three fields when present in the SDK response:
  - reasoning_content  — direct DeepSeek/Kimi API (HTTP 400 if missing)
  - reasoning          — OpenRouter normalized thinking text
  - reasoning_details  — OpenRouter structured reasoning (list)

All three are absent on Claude/Gemini — no spurious keys injected.

Refs NousResearch#17341

…mpty

DeepSeek V4 Pro (and other reasoning models) put their chain of thought
into reasoning tokens rather than the structured tool-call arguments,
often leaving the 'reason' field empty in update_checklist calls.

Now when reason is empty, we synthesize it from the verdict counts
(e.g. 'Judge verdict: 3 completed, 1 impossible').  Falls back to
'no reason provided (judge returned empty reason field)' only when
truly no updates were made — which is itself diagnostic.

Refs NousResearch#17341

… thinking mode

DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:

    The reasoning content in the thinking mode must be passed back to the API.

run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.

Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.

Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).

Refs NousResearch#15250, NousResearch#17400.
Closes NousResearch#17341.

… thinking mode

DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:

    The reasoning content in the thinking mode must be passed back to the API.

run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.

Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.

Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).

Refs NousResearch#15250, NousResearch#17400.
Closes NousResearch#17341.

… thinking mode

DeepSeek V4 Pro tightened thinking-mode validation and rejects empty-string
reasoning_content with HTTP 400:

    The reasoning content in the thinking mode must be passed back to the API.

run_agent.py injected "" at three fallback sites — the tool-call pad in
_build_assistant_message and both injection branches of
_copy_reasoning_content_for_api (cross-provider poison guard + unconditional
thinking pad). All three now emit " " (single space), which satisfies the
non-empty check on V4 Pro without leaking fabricated reasoning.

Also upgrades stale empty-string placeholders on replay: sessions persisted
before this change have reasoning_content="" pinned at creation time; when
the active provider enforces thinking-mode echo, the replay path now rewrites
"" -> " " so existing users don't 400 on their first V4 Pro turn after
updating. Non-thinking providers still round-trip "" verbatim.

Updates 9 existing assertions + adds 2 regression tests (stale-placeholder
upgrade, non-thinking verbatim preservation).

Refs NousResearch#15250, NousResearch#17400.
Closes NousResearch#17341.