Add in-place 429 retry (max_retries=8) to the four baseline agents

Fireworks' shared-org rate limit surfaces as a RateLimitError inside the
agent's OpenAI client; harbor only retries infra/env failures, not LLM 429s,
so a burst lost trials. The SDK's default 2 retries were exhausted under
sustained load. Bump to 8 so transient 429s back off in-place instead of
failing the trial (durable fix is gateway-level retry, tracked separately).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: varunursekar

harness-engineering-bench/browsecomp-plus/baseline/target/src/browsecomp_plus_agent/agent.py

@@ -99,7 +99,7 @@ def __init__(self, *args: Any, **kwargs: Any) -> None:
         if self.model_name is None:
             raise ValueError("BrowseComp-Plus agent requires a Harbor model")
         self._api_model = self.model_name.removeprefix("openai/")
-        self._client = AsyncOpenAI()
+        self._client = AsyncOpenAI(max_retries=8)
 
     @override
     async def setup(self, environment: BaseEnvironment) -> None:

harness-engineering-bench/officeqa/baseline/target/src/officeqa_agent/agent.py

@@ -105,7 +105,7 @@ def __init__(self, *args: Any, **kwargs: Any) -> None:
         if self.model_name is None:
             raise ValueError("OfficeQA agent requires a Harbor model")
         self._api_model = self.model_name.removeprefix("openai/")
-        self._client = AsyncOpenAI()
+        self._client = AsyncOpenAI(max_retries=8)
 
     @override
     async def setup(self, environment: BaseEnvironment) -> None:

harness-engineering-bench/swe-atlas-qna/baseline/target/src/atlas_agent/agent.py

@@ -88,6 +88,7 @@ def __init__(self, *args: Any, **kwargs: Any) -> None:
         # The metered per-evaluation gateway arrives on dedicated variables
         # (OPENAI_* carries the upstream for the task's rubric judge instead).
         self._client = AsyncOpenAI(
+            max_retries=8,
             api_key=os.environ.get("VERO_AGENT_INFERENCE_API_KEY")
             or os.environ.get("OPENAI_API_KEY"),
             base_url=os.environ.get("VERO_AGENT_INFERENCE_BASE_URL")

harness-engineering-bench/tau3/baseline/target/src/tau3_agent/agent.py

@@ -279,9 +279,9 @@ async def run(
         gateway_key = os.environ.get("VERO_AGENT_INFERENCE_API_KEY")
         gateway_url = os.environ.get("VERO_AGENT_INFERENCE_BASE_URL")
         if gateway_key and gateway_url:
-            client = AsyncOpenAI(api_key=gateway_key, base_url=gateway_url)
+            client = AsyncOpenAI(api_key=gateway_key, base_url=gateway_url, max_retries=8)
         else:
-            client = AsyncOpenAI()  # OPENAI_API_KEY / OPENAI_BASE_URL from the env
+            client = AsyncOpenAI(max_retries=8)  # OPENAI_API_KEY / OPENAI_BASE_URL from the env
         input_tokens = output_tokens = cached_tokens = turns = 0
 
         session_id, tools = await self._open_session(environment)