fix(runner): add thinking config to allowed forwarded props

Address CodeRabbit review feedback on PR #985: add the new `thinking`
config (adaptive/enabled/disabled) to ALLOWED_FORWARDED_PROPS so
clients can use the preferred API alongside the legacy
`max_thinking_tokens` option.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: jeremyeder

components/runners/ambient-runner/ag_ui_claude_sdk/adapter.py

@@ -137,7 +137,7 @@ class ClaudeAgentAdapter:
     Forwarded Props Support:
         Per-run overrides for execution control without changing agent identity.
         Whitelisted keys include: resume, fork_session, model, temperature, max_tokens,
-        max_thinking_tokens, max_turns, max_budget_usd, output_format, etc.
+        thinking, max_thinking_tokens, max_turns, max_budget_usd, output_format, etc.
 
         Example:
             RunAgentInput(
@@ -770,12 +770,16 @@ def flush_pending_msg():
                             current_reasoning_id = str(uuid.uuid4())
                             ts = now_ms()
                             yield ReasoningStartEvent(
-                                threadId=thread_id, runId=run_id,
-                                messageId=current_reasoning_id, timestamp=ts,
+                                threadId=thread_id,
+                                runId=run_id,
+                                messageId=current_reasoning_id,
+                                timestamp=ts,
                             )
                             yield ReasoningMessageStartEvent(
-                                threadId=thread_id, runId=run_id,
-                                messageId=current_reasoning_id, timestamp=ts,
+                                threadId=thread_id,
+                                runId=run_id,
+                                messageId=current_reasoning_id,
+                                timestamp=ts,
                             )
                         elif block_type == "tool_use":
                             # Tool call starting - emit TOOL_CALL_START
@@ -807,12 +811,16 @@ def flush_pending_msg():
                             in_thinking_block = False
                             ts = now_ms()
                             yield ReasoningMessageEndEvent(
-                                threadId=thread_id, runId=run_id,
-                                messageId=current_reasoning_id, timestamp=ts,
+                                threadId=thread_id,
+                                runId=run_id,
+                                messageId=current_reasoning_id,
+                                timestamp=ts,
                             )
                             yield ReasoningEndEvent(
-                                threadId=thread_id, runId=run_id,
-                                messageId=current_reasoning_id, timestamp=ts,
+                                threadId=thread_id,
+                                runId=run_id,
+                                messageId=current_reasoning_id,
+                                timestamp=ts,
                             )
 
                             # Persist thinking content as ReasoningMessage per AG-UI spec.
@@ -1126,9 +1134,7 @@ def flush_pending_msg():
                 yield ReasoningMessageEndEvent(
                     threadId=thread_id, runId=run_id, timestamp=ts
                 )
-                yield ReasoningEndEvent(
-                    threadId=thread_id, runId=run_id, timestamp=ts
-                )
+                yield ReasoningEndEvent(threadId=thread_id, runId=run_id, timestamp=ts)
                 in_thinking_block = False
 
             if has_streamed_text and current_message_id:

components/runners/ambient-runner/ag_ui_claude_sdk/config.py

@@ -16,7 +16,8 @@
     "fallback_model",  # Fallback if primary fails
     "temperature",  # Sampling temperature
     "max_tokens",  # Response length limit
-    "max_thinking_tokens",  # Reasoning depth limit
+    "max_thinking_tokens",  # Reasoning depth limit (legacy, prefer thinking)
+    "thinking",  # Thinking config: {"type": "adaptive"} | {"type": "enabled", "budget_tokens": N} | {"type": "disabled"}
     "max_turns",  # Conversation turn limit
     "max_budget_usd",  # Cost limit per run
     # Output control