From 8b58f9f68f01a96f101366b6b9a98dbd341db301 Mon Sep 17 00:00:00 2001
From: kshitij <82637225+kshitijk4poor@users.noreply.github.com>
Date: Sat, 15 Aug 2026 12:43:49 +0530
Subject: [PATCH] test(bedrock): pin stream-path cap omission; document
 truthiness edge

Self-review follow-up: cover call_converse_stream's max_tokens=None path
(same builder, previously unpinned) and document why the shim reads the
caller cap with truthiness rather than 'is None' (parity with the
Anthropic shim's reading).
---
 agent/auxiliary_client.py           |  3 +++
 tests/agent/test_bedrock_adapter.py | 21 +++++++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/agent/auxiliary_client.py b/agent/auxiliary_client.py
index 0f163f38ef..d49a797ed8 100644
--- a/agent/auxiliary_client.py
+++ b/agent/auxiliary_client.py
@@ -2218,6 +2218,9 @@ class _BedrockCompletionsAdapter:
             # allowed output, matching the no-cap-by-default policy every
             # other aux wire already follows (#10809: vision descriptions
             # stayed capped at the shim's old hardcoded 4096 on Bedrock).
+            # Truthiness (not `is None`) is deliberate — it matches the
+            # sibling Anthropic shim's reading of max_tokens above, so a
+            # nonsense explicit 0 is treated as "no cap" on both wires.
             max_tokens=int(max_tokens) if max_tokens else None,
             temperature=kwargs.get("temperature"),
             top_p=kwargs.get("top_p"),
diff --git a/tests/agent/test_bedrock_adapter.py b/tests/agent/test_bedrock_adapter.py
index 8d6acf2ede..e6c2c3c3c4 100644
--- a/tests/agent/test_bedrock_adapter.py
+++ b/tests/agent/test_bedrock_adapter.py
@@ -423,6 +423,27 @@ class TestBuildConverseKwargs:
         )
         assert "inferenceConfig" not in kwargs
 
+    def test_call_converse_stream_omits_cap_for_none(self):
+        """The streaming entry point funnels through the same builder — pin
+        that max_tokens=None omits the cap there too."""
+        from unittest.mock import MagicMock, patch as mock_patch
+        from agent.bedrock_adapter import call_converse_stream
+        boto3_client = MagicMock()
+        boto3_client.converse_stream.return_value = {"stream": []}
+        with mock_patch(
+            "agent.bedrock_adapter._get_bedrock_runtime_client",
+            return_value=boto3_client,
+        ):
+            call_converse_stream(
+                region="us-east-1",
+                model="test-model",
+                messages=[{"role": "user", "content": "Hi"}],
+                max_tokens=None,
+                temperature=0.2,
+            )
+        wire_kwargs = boto3_client.converse_stream.call_args.kwargs
+        assert "maxTokens" not in wire_kwargs.get("inferenceConfig", {})
+