From 2b728e12748e3a30273acdbef36ecad15a04f2b9 Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Tue, 28 Apr 2026 03:50:51 -0700
Subject: [PATCH] fix(agent): drop thinking-only assistant turns before
 provider call (#16959)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Adds a pre-call sanitizer that detects assistant messages containing only
reasoning (reasoning / reasoning_content, no visible content, no
tool_calls) and drops them from the API copy. Adjacent user messages
left behind are merged so role alternation is preserved for the
provider.

Mirrors Claude Code's approach in src/utils/messages.ts
(filterOrphanedThinkingOnlyMessages + mergeAdjacentUserMessages). We
drop the whole turn rather than fabricate stub text (the '.' /
'(continued)' pattern from contributor PRs #11098, #13010, #16842 that
were rejected because they put words in the model's mouth).

The stored conversation history (self.messages) is never mutated — only
the per-call api_messages copy. Users still see the reasoning block in
the CLI/gateway transcript; only the wire copy is cleaned. Session
persistence keeps the full trace.

Two call sites covered:
- Main agent loop, after _sanitize_api_messages (catches every turn).
- Iteration-limit-summary fallback path.

Tests: tests/run_agent/test_thinking_only_sanitizer.py — 25 cases
covering detection (string/list content, whitespace-only, tool_calls,
reasoning_details list form), drop behavior, adjacent-user merge
(string+string, list+list, mixed), non-mutation of input dicts, and
system-message handling.

E2E live-tested against 5 providers with a poisoned history (empty
assistant message + reasoning_content): OpenRouter→Anthropic/OpenAI/
DeepSeek-R1/Qwen, native Gemini. All 5 accepted the cleaned request.
Happy-path regression (5/5) confirms the sanitizer is a noop when no
thinking-only turn exists.

Related: #16823 (wontfix — stub-text approach rejected).

Co-authored-by: teknium1 <teknium@users.noreply.github.com>
---
 run_agent.py                                  | 153 +++++++++++
 .../run_agent/test_thinking_only_sanitizer.py | 249 ++++++++++++++++++
 2 files changed, 402 insertions(+)
 create mode 100644 tests/run_agent/test_thinking_only_sanitizer.py

diff --git a/run_agent.py b/run_agent.py
index 568a7f7ac0..387a6d00ed 100644
--- a/run_agent.py
+++ b/run_agent.py
@@ -4793,6 +4793,145 @@ class AIAgent:
             )
         return messages
 
+    @staticmethod
+    def _is_thinking_only_assistant(msg: Dict[str, Any]) -> bool:
+        """Return True if ``msg`` is an assistant turn whose only payload is reasoning.
+
+        "Thinking-only" means the model emitted reasoning (``reasoning`` or
+        ``reasoning_content``) but no visible text and no tool_calls. When sent
+        back to providers that convert reasoning into thinking blocks (native
+        Anthropic, OpenRouter Anthropic, third-party Anthropic-compatible
+        gateways), the resulting message has only thinking blocks — which
+        Anthropic rejects with HTTP 400 "The final block in an assistant
+        message cannot be `thinking`."
+
+        Symmetric with Claude Code's ``filterOrphanedThinkingOnlyMessages``
+        (src/utils/messages.ts). We drop the whole turn from the API copy
+        rather than fabricating stub text — the message log (UI transcript)
+        keeps the reasoning block; only the wire copy is cleaned.
+        """
+        if not isinstance(msg, dict) or msg.get("role") != "assistant":
+            return False
+        if msg.get("tool_calls"):
+            return False
+        # Does it have any actual output?
+        content = msg.get("content")
+        if isinstance(content, str):
+            if content.strip():
+                return False
+        elif isinstance(content, list):
+            for block in content:
+                if not isinstance(block, dict):
+                    if block:  # non-empty non-dict string etc.
+                        return False
+                    continue
+                btype = block.get("type")
+                if btype in ("thinking", "redacted_thinking"):
+                    continue
+                if btype == "text":
+                    text = block.get("text", "")
+                    if isinstance(text, str) and text.strip():
+                        return False
+                    continue
+                # tool_use, image, document, etc. — real payload
+                return False
+        elif content is not None and content != "":
+            return False
+        # Content is empty-ish. Is there reasoning to make it thinking-only?
+        reasoning = msg.get("reasoning_content") or msg.get("reasoning")
+        if isinstance(reasoning, str) and reasoning.strip():
+            return True
+        # reasoning_details list form
+        rd = msg.get("reasoning_details")
+        if isinstance(rd, list) and rd:
+            return True
+        return False
+
+    @staticmethod
+    def _drop_thinking_only_and_merge_users(
+        messages: List[Dict[str, Any]],
+    ) -> List[Dict[str, Any]]:
+        """Drop thinking-only assistant turns; merge any adjacent user messages left behind.
+
+        Runs on the per-call ``api_messages`` copy only. The stored
+        conversation history (``self.messages``) is never mutated, so the
+        user still sees the thinking block in the CLI/gateway transcript and
+        session persistence keeps the full trace. Only the wire copy sent to
+        the provider is cleaned.
+
+        Why drop-and-merge rather than inject stub text:
+        - Fabricating ``"."`` / ``"(continued)"`` text lies in the history
+          and makes future turns see model output the model didn't emit.
+        - Dropping the turn preserves honesty; merging adjacent user messages
+          preserves the provider's role-alternation invariant.
+        - This is the pattern used by Claude Code's ``normalizeMessagesForAPI``
+          (filterOrphanedThinkingOnlyMessages + mergeAdjacentUserMessages).
+        """
+        if not messages:
+            return messages
+
+        # Pass 1: drop thinking-only assistant turns.
+        kept = [m for m in messages if not AIAgent._is_thinking_only_assistant(m)]
+        dropped = len(messages) - len(kept)
+        if dropped == 0:
+            return messages
+
+        # Pass 2: merge any newly-adjacent user messages.
+        merged: List[Dict[str, Any]] = []
+        merges = 0
+        for m in kept:
+            prev = merged[-1] if merged else None
+            if (
+                prev is not None
+                and prev.get("role") == "user"
+                and m.get("role") == "user"
+            ):
+                prev_content = prev.get("content", "")
+                cur_content = m.get("content", "")
+                # Work on a copy of ``prev`` so the caller's input dicts are
+                # never mutated. ``_sanitize_api_messages`` upstream already
+                # hands us per-call copies, but staying pure here means we
+                # can be called safely from anywhere (tests, other loops).
+                prev_copy = dict(prev)
+                # Only string-content merge is meaningful for role-alternation
+                # purposes. If either side is a list (multimodal), append as a
+                # separate block rather than collapsing.
+                if isinstance(prev_content, str) and isinstance(cur_content, str):
+                    sep = "\n\n" if prev_content and cur_content else ""
+                    prev_copy["content"] = prev_content + sep + cur_content
+                elif isinstance(prev_content, list) and isinstance(cur_content, list):
+                    prev_copy["content"] = list(prev_content) + list(cur_content)
+                elif isinstance(prev_content, list) and isinstance(cur_content, str):
+                    if cur_content:
+                        prev_copy["content"] = list(prev_content) + [
+                            {"type": "text", "text": cur_content}
+                        ]
+                    else:
+                        prev_copy["content"] = list(prev_content)
+                elif isinstance(prev_content, str) and isinstance(cur_content, list):
+                    new_blocks: List[Dict[str, Any]] = []
+                    if prev_content:
+                        new_blocks.append({"type": "text", "text": prev_content})
+                    new_blocks.extend(cur_content)
+                    prev_copy["content"] = new_blocks
+                else:
+                    # Unknown content shape — fall back to appending separately
+                    # (violates alternation, but safer than raising in a hot path).
+                    merged.append(m)
+                    continue
+                merged[-1] = prev_copy
+                merges += 1
+            else:
+                merged.append(m)
+
+        logger.debug(
+            "Pre-call sanitizer: dropped %d thinking-only assistant turn(s), "
+            "merged %d adjacent user message(s)",
+            dropped,
+            merges,
+        )
+        return merged
+
     @staticmethod
     def _cap_delegate_task_calls(tool_calls: list) -> list:
         """Truncate excess delegate_task calls to max_concurrent_children.
@@ -9498,6 +9637,10 @@ class AIAgent:
                 for idx, pfm in enumerate(self.prefill_messages):
                     api_messages.insert(sys_offset + idx, pfm.copy())
 
+            # Same safety net as the main loop: drop thinking-only assistant
+            # turns so Anthropic-family providers don't 400 the summary call.
+            api_messages = self._drop_thinking_only_and_merge_users(api_messages)
+
             summary_extra_body = {}
             try:
                 from agent.auxiliary_client import _fixed_temperature_for_model, OMIT_TEMPERATURE as _OMIT_TEMP
@@ -10208,6 +10351,16 @@ class AIAgent:
             # manual message manipulation are always caught.
             api_messages = self._sanitize_api_messages(api_messages)
 
+            # Drop thinking-only assistant turns (reasoning but no visible
+            # output and no tool_calls) and merge any adjacent user messages
+            # left behind. Prevents Anthropic 400s ("The final block in an
+            # assistant message cannot be `thinking`.") and equivalent errors
+            # from third-party Anthropic-compatible gateways that can't replay
+            # a thinking-only turn. Runs on the per-call copy only — the
+            # stored conversation history keeps the reasoning block for the
+            # UI transcript and session persistence.
+            api_messages = self._drop_thinking_only_and_merge_users(api_messages)
+
             # Normalize message whitespace and tool-call JSON for consistent
             # prefix matching.  Ensures bit-perfect prefixes across turns,
             # which enables KV cache reuse on local inference servers
diff --git a/tests/run_agent/test_thinking_only_sanitizer.py b/tests/run_agent/test_thinking_only_sanitizer.py
new file mode 100644
index 0000000000..83cf35f6d1
--- /dev/null
+++ b/tests/run_agent/test_thinking_only_sanitizer.py
@@ -0,0 +1,249 @@
+"""Tests for the thinking-only assistant message sanitizer.
+
+Covers _is_thinking_only_assistant() + _drop_thinking_only_and_merge_users()
+in run_agent.py. The sanitizer runs on the per-call api_messages copy and
+drops assistant turns that contain only reasoning (no visible content, no
+tool_calls). Adjacent user messages left behind are merged so role
+alternation is preserved for the provider.
+
+Claude Code uses this exact pattern (filterOrphanedThinkingOnlyMessages +
+mergeAdjacentUserMessages in src/utils/messages.ts). See #16823 for the
+backstory on why the alternative — fabricating "." stub text — was rejected.
+"""
+
+from run_agent import AIAgent
+
+
+# ---------------------------------------------------------------------------
+# _is_thinking_only_assistant — detection
+# ---------------------------------------------------------------------------
+
+
+class TestIsThinkingOnlyAssistant:
+
+    def test_plain_assistant_reply_is_not_thinking_only(self):
+        msg = {"role": "assistant", "content": "Hello there"}
+        assert not AIAgent._is_thinking_only_assistant(msg)
+
+    def test_assistant_with_tool_calls_is_not_thinking_only(self):
+        msg = {
+            "role": "assistant",
+            "content": "",
+            "reasoning": "let me use a tool",
+            "tool_calls": [{"id": "c1", "function": {"name": "terminal", "arguments": "{}"}}],
+        }
+        assert not AIAgent._is_thinking_only_assistant(msg)
+
+    def test_empty_content_plus_reasoning_is_thinking_only(self):
+        msg = {"role": "assistant", "content": "", "reasoning": "thinking..."}
+        assert AIAgent._is_thinking_only_assistant(msg)
+
+    def test_none_content_plus_reasoning_content_is_thinking_only(self):
+        msg = {"role": "assistant", "content": None, "reasoning_content": "thinking..."}
+        assert AIAgent._is_thinking_only_assistant(msg)
+
+    def test_whitespace_only_content_plus_reasoning_is_thinking_only(self):
+        msg = {"role": "assistant", "content": "   \n\n  ", "reasoning": "r"}
+        assert AIAgent._is_thinking_only_assistant(msg)
+
+    def test_empty_content_no_reasoning_is_not_thinking_only(self):
+        # If there's no reasoning either, this is just an empty turn — let
+        # other sanitizers handle it (orphan-tool-pair, etc.). We only care
+        # about the specific thinking-only case.
+        msg = {"role": "assistant", "content": ""}
+        assert not AIAgent._is_thinking_only_assistant(msg)
+
+    def test_list_content_all_thinking_blocks_is_thinking_only(self):
+        # Anthropic-native shape
+        msg = {
+            "role": "assistant",
+            "content": [
+                {"type": "thinking", "thinking": "...", "signature": "sig"},
+            ],
+            "reasoning": "...",
+        }
+        assert AIAgent._is_thinking_only_assistant(msg)
+
+    def test_list_content_with_real_text_is_not_thinking_only(self):
+        msg = {
+            "role": "assistant",
+            "content": [
+                {"type": "thinking", "thinking": "..."},
+                {"type": "text", "text": "Hi there"},
+            ],
+            "reasoning": "...",
+        }
+        assert not AIAgent._is_thinking_only_assistant(msg)
+
+    def test_list_content_with_tool_use_block_is_not_thinking_only(self):
+        msg = {
+            "role": "assistant",
+            "content": [
+                {"type": "thinking", "thinking": "..."},
+                {"type": "tool_use", "id": "tu1", "name": "terminal", "input": {}},
+            ],
+        }
+        assert not AIAgent._is_thinking_only_assistant(msg)
+
+    def test_list_content_thinking_plus_whitespace_text_is_thinking_only(self):
+        msg = {
+            "role": "assistant",
+            "content": [
+                {"type": "thinking", "thinking": "..."},
+                {"type": "text", "text": "   "},
+            ],
+            "reasoning": "...",
+        }
+        assert AIAgent._is_thinking_only_assistant(msg)
+
+    def test_reasoning_details_list_form_detected(self):
+        msg = {
+            "role": "assistant",
+            "content": "",
+            "reasoning_details": [{"type": "thinking", "text": "..."}],
+        }
+        assert AIAgent._is_thinking_only_assistant(msg)
+
+    def test_user_message_never_thinking_only(self):
+        assert not AIAgent._is_thinking_only_assistant({"role": "user", "content": ""})
+
+    def test_tool_message_never_thinking_only(self):
+        assert not AIAgent._is_thinking_only_assistant(
+            {"role": "tool", "content": "", "tool_call_id": "x"}
+        )
+
+    def test_non_dict_returns_false(self):
+        assert not AIAgent._is_thinking_only_assistant(None)
+        assert not AIAgent._is_thinking_only_assistant("hello")
+
+
+# ---------------------------------------------------------------------------
+# _drop_thinking_only_and_merge_users — the full pass
+# ---------------------------------------------------------------------------
+
+
+class TestDropThinkingOnlyAndMergeUsers:
+
+    def test_empty_list_passthrough(self):
+        assert AIAgent._drop_thinking_only_and_merge_users([]) == []
+
+    def test_no_thinking_only_messages_is_noop_identity(self):
+        msgs = [
+            {"role": "user", "content": "hi"},
+            {"role": "assistant", "content": "hello"},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        # Should return the original list untouched (identity) when no changes.
+        assert out is msgs
+
+    def test_drops_thinking_only_between_user_messages_and_merges(self):
+        msgs = [
+            {"role": "user", "content": "help me with X"},
+            {"role": "assistant", "content": "", "reasoning": "let me think"},
+            {"role": "user", "content": "ok continue"},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert len(out) == 1
+        assert out[0]["role"] == "user"
+        assert out[0]["content"] == "help me with X\n\nok continue"
+
+    def test_preserves_alternation_after_drop(self):
+        msgs = [
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "content": "", "reasoning": "..."},
+            {"role": "user", "content": "u2"},
+            {"role": "assistant", "content": "real reply"},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        roles = [m["role"] for m in out]
+        assert roles == ["user", "assistant"]
+        assert out[0]["content"] == "u1\n\nu2"
+        assert out[1]["content"] == "real reply"
+
+    def test_does_not_merge_when_drop_leaves_non_adjacent_users(self):
+        # Thinking-only at end of conversation — no trailing user to merge
+        msgs = [
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "content": "reply"},
+            {"role": "user", "content": "u2"},
+            {"role": "assistant", "content": "", "reasoning": "..."},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert [m["role"] for m in out] == ["user", "assistant", "user"]
+
+    def test_multiple_thinking_only_in_sequence_collapses(self):
+        msgs = [
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "content": "", "reasoning": "r1"},
+            {"role": "assistant", "content": "", "reasoning": "r2"},
+            {"role": "user", "content": "u2"},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert len(out) == 1
+        assert out[0]["content"] == "u1\n\nu2"
+
+    def test_does_not_touch_stored_messages_original_list_unmutated(self):
+        original_first_user = {"role": "user", "content": "u1"}
+        original_assistant = {"role": "assistant", "content": "", "reasoning": "..."}
+        original_second_user = {"role": "user", "content": "u2"}
+        msgs = [original_first_user, original_assistant, original_second_user]
+        AIAgent._drop_thinking_only_and_merge_users(msgs)
+        # Caller passes in a per-call copy already, but the sanitizer itself
+        # must not rewrite the dicts it was handed on the drop path.
+        # (It CAN mutate merged dicts — those come from the caller's copy.)
+        assert original_first_user["content"] == "u1"
+        assert original_second_user["content"] == "u2"
+
+    def test_tool_result_between_user_and_thinking_preserved(self):
+        # Tool results shouldn't block a drop — but they do block the merge
+        # (user/tool are different roles). This scenario shouldn't happen in
+        # practice because a thinking-only turn won't have tool_calls, but if
+        # it did somehow, the surrounding tool result stays put.
+        msgs = [
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "tool_calls": [{"id": "c1", "function": {"name": "t", "arguments": "{}"}}]},
+            {"role": "tool", "tool_call_id": "c1", "content": "ok"},
+            {"role": "assistant", "content": "", "reasoning": "..."},
+            {"role": "user", "content": "u2"},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert [m["role"] for m in out] == ["user", "assistant", "tool", "user"]
+
+    def test_merge_concatenates_list_content_user_messages(self):
+        msgs = [
+            {"role": "user", "content": [{"type": "text", "text": "first"}]},
+            {"role": "assistant", "content": "", "reasoning": "..."},
+            {"role": "user", "content": [{"type": "text", "text": "second"}]},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert len(out) == 1
+        assert out[0]["content"] == [
+            {"type": "text", "text": "first"},
+            {"type": "text", "text": "second"},
+        ]
+
+    def test_merge_mixed_string_and_list_content(self):
+        msgs = [
+            {"role": "user", "content": "plain text"},
+            {"role": "assistant", "content": "", "reasoning": "..."},
+            {"role": "user", "content": [{"type": "text", "text": "block text"}]},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert len(out) == 1
+        assert out[0]["content"] == [
+            {"type": "text", "text": "plain text"},
+            {"type": "text", "text": "block text"},
+        ]
+
+    def test_system_messages_ignored_by_pass(self):
+        msgs = [
+            {"role": "system", "content": "sys prompt"},
+            {"role": "user", "content": "u1"},
+            {"role": "assistant", "content": "", "reasoning": "..."},
+            {"role": "user", "content": "u2"},
+        ]
+        out = AIAgent._drop_thinking_only_and_merge_users(msgs)
+        assert len(out) == 2
+        assert out[0]["role"] == "system"
+        assert out[1]["role"] == "user"
+        assert out[1]["content"] == "u1\n\nu2"