Skip to content

Agent Zero AI framework

Skip unused Chat replay preparation and share input conversion

Alessandro028d3bec1ae5
diff --git a/agent.py b/agent.py
index 31a240b92..2e558bd15 100644
--- a/agent.py
+++ b/agent.py
@@ -961,7 +961,7 @@ class Agent:
model,
history_counter,
)
- call_data["responses_local_input_items"] = self._responses_prompt_input_items(
+ call_data["responses_local_input_items"] = ResponsesTransport.input_from_model_messages(
model,
messages,
)
@@ -1051,18 +1051,7 @@ class Agent:
output = message.output()
langchain_messages = history.output_langchain(output)
- if hasattr(model, "_convert_messages"):
- converted = model._convert_messages(langchain_messages)
- return ResponsesTransport.input_from_messages(converted)
- return []
-
- def _responses_prompt_input_items(
- self, model: Any, messages: list[BaseMessage]
- ) -> list[dict[str, Any]]:
- if not hasattr(model, "_convert_messages"):
- return []
- converted = model._convert_messages(messages)
- return ResponsesTransport.input_from_messages(converted)
+ return ResponsesTransport.input_from_model_messages(model, langchain_messages)
def _remember_llm_result_state(
self, llm_result: LLMResult, history_message: history.Message
diff --git a/helpers/litellm_transport.py b/helpers/litellm_transport.py
index 5c0594680..fa402f5ec 100644
--- a/helpers/litellm_transport.py
+++ b/helpers/litellm_transport.py
@@ -28,6 +28,10 @@ class TransportMode(Enum):
RESPONSES = "responses"
CHAT_COMPLETIONS = "chat_completions"
+ @classmethod
+ def from_value(cls, value: Any) -> "TransportMode":
+ return cls.RESPONSES if str(value or "").lower().strip() in RESPONSES_ALIASES else cls.CHAT_COMPLETIONS
+
class TransportRecovery(Enum):
RAISE = "raise"
@@ -89,7 +93,7 @@ class TransportPolicy:
kwargs: dict[str, Any],
messages: list[dict[str, Any]] | None = None,
) -> "TransportPolicy":
- mode = cls._pop_mode(kwargs)
+ mode = TransportMode.from_value(kwargs.pop("a0_api_mode", None))
allow_fallback = _coerce_bool(
kwargs.pop("a0_responses_fallback", True), default=True
)
@@ -138,13 +142,6 @@ class TransportPolicy:
state=state,
)
- @staticmethod
- def _pop_mode(kwargs: dict[str, Any]) -> TransportMode:
- value = str(kwargs.pop("a0_api_mode", "") or "").lower().strip()
- if value in RESPONSES_ALIASES:
- return TransportMode.RESPONSES
- return TransportMode.CHAT_COMPLETIONS
-
@property
def using_responses(self) -> bool:
return self.mode is TransportMode.RESPONSES
@@ -750,6 +747,12 @@ class ChatCompletionsStreamParser:
class ResponsesTransport:
+ @staticmethod
+ def input_from_model_messages(model: Any, messages: list) -> list[dict[str, Any]]:
+ if not hasattr(model, "_convert_messages"):
+ return []
+ return ResponsesTransport.input_from_messages(model._convert_messages(messages))
+
@classmethod
def from_chat(
cls,
diff --git a/helpers/litellm_transport.py.dox.md b/helpers/litellm_transport.py.dox.md
index 4367fb0f2..826bd0656 100644
--- a/helpers/litellm_transport.py.dox.md
+++ b/helpers/litellm_transport.py.dox.md
@@ -49,6 +49,8 @@
- Treat fallback behavior as a shared transport contract, not a provider registry.
- Keep tool conversion symmetric between Chat Completions and Responses requests.
+- TransportMode.from_value owns API-mode interpretation; input_from_model_messages owns model-message conversion shared with Agent and replay preparation.
+
## Verification
- Run `pytest tests/test_stream_tool_early_stop.py tests/test_responses_architecture.py -q` after changing transport normalization or fallback behavior.
diff --git a/helpers/responses_history.py b/helpers/responses_history.py
index 56cae50fa..6851ff87d 100644
--- a/helpers/responses_history.py
+++ b/helpers/responses_history.py
@@ -28,6 +28,7 @@ def remember_prompt(loop_data: Any, text: str, system_message: Any, protocol: li
def prepare_call(agent: Any, call_data: dict) -> dict:
from langchain_core.prompts import ChatPromptTemplate
from helpers import history
+ from helpers.litellm_transport import ResponsesTransport, TransportMode
from helpers.secrets import get_secrets_manager
model, messages = call_data["model"], call_data["messages"]
@@ -37,21 +38,22 @@ def prepare_call(agent: Any, call_data: dict) -> dict:
)}
prepared = params.get("responses_history", {})
if (
- not prepared or not hasattr(model, "_convert_messages")
+ TransportMode.from_value(getattr(model, "kwargs", {}).get("a0_api_mode")) is not TransportMode.RESPONSES
+ or not prepared or not hasattr(model, "_convert_messages")
or any(message.additional_kwargs or getattr(message, "tool_calls", None) for message in messages)
or ChatPromptTemplate.from_messages(messages).format() != prepared["text"]
):
return kwargs
def render(record):
- rendered = agent._responses_prompt_input_items(
+ rendered = ResponsesTransport.input_from_model_messages(
model, history.output_langchain([{**record, "ai": False}]),
)
return rendered[0].get("content") if len(rendered) == 1 else None
kwargs["responses_history_context"] = {
- "prompt": agent._responses_prompt_input_items(model, messages),
- "prefix": agent._responses_prompt_input_items(model, prepared["prefix"]),
+ "prompt": ResponsesTransport.input_from_model_messages(model, messages),
+ "prefix": ResponsesTransport.input_from_model_messages(model, prepared["prefix"]),
"groups": prepare_groups(
agent.loop_data.history_output, render,
get_secrets_manager(agent.context).mask_values,
diff --git a/helpers/responses_history.py.dox.md b/helpers/responses_history.py.dox.md
index 6df15b473..586ddd703 100644
--- a/helpers/responses_history.py.dox.md
+++ b/helpers/responses_history.py.dox.md
@@ -16,6 +16,8 @@
- Preserve summaries, unmatched/custom layouts, attachments and all remaining prepared input. No execution or policy authorization occurs here.
- Internal history context is removed before every provider request; Chat and fallback messages remain untouched.
+- Skip replay preparation for Chat mode after model-call hooks. Reuse the transport conversion primitive rather than private Agent methods; the actual request retains final fallback and input-validation checks.
+
## Verification
- Run native history, Responses architecture/transport, prompt protocol, history and tool-policy tests. Verify named Codex and Venice presets live after integration changes.
diff --git a/tests/test_responses_history.py b/tests/test_responses_history.py
index 6a7032662..3e64f8d72 100644
--- a/tests/test_responses_history.py
+++ b/tests/test_responses_history.py
@@ -85,7 +85,7 @@ def test_prepared_context_lifecycle_and_post_hook_changes(monkeypatch):
agent = object.__new__(Agent)
agent.context = SimpleNamespace()
agent.loop_data = LoopData()
- model = LiteLLMChatWrapper(model="test", provider="openai", model_config=None)
+ model = LiteLLMChatWrapper(model="test", provider="openai", model_config=None, a0_api_mode="responses")
prompt = [SystemMessage("Rules"), HumanMessage("Question")]
monkeypatch.setattr(secrets, "get_secrets_manager", lambda _: SimpleNamespace(mask_values=lambda text: text))
history.remember_prompt(agent.loop_data, ChatPromptTemplate.from_messages(prompt).format(), prompt[0], [])
@@ -93,6 +93,12 @@ def test_prepared_context_lifecycle_and_post_hook_changes(monkeypatch):
prepared = history.prepare_call(agent, call)
assert prepared["responses_history_context"]["prompt"] == BASE
assert prepared["responses_prompt_replacements"] == {"Rules": "Override"}
+ model.kwargs["a0_api_mode"] = "chat"
+ with monkeypatch.context() as chat:
+ chat.setattr(model, "_convert_messages", lambda _: pytest.fail("Chat must not prepare replay input"))
+ chat.setattr(secrets, "get_secrets_manager", lambda _: pytest.fail("Chat must not prepare replay masking"))
+ assert "responses_history_context" not in history.prepare_call(agent, call)
+ model.kwargs["a0_api_mode"] = "responses"
for changed in (
[SystemMessage("Changed by hook"), prompt[1]],
[SystemMessage("Rules", additional_kwargs={"provider_control": True}), prompt[1]],