Removed question argument from gen_answer, since it's dead code

paperqa/agents/main.py

@@ -163,7 +163,7 @@ async def _run_with_timeout_failure(
         generate_answer_tool = next(
             filter(lambda x: x.info.name == GenerateAnswer.TOOL_FN_NAME, env.tools)
         )
-        await generate_answer_tool._tool_fn(question=query.query, state=env.state)
+        await generate_answer_tool._tool_fn(state=env.state)
     return env.state.session, status
 
 
@@ -216,7 +216,7 @@ async def rollout() -> AgentStatus:
         ):
             await step(search_tool, query=search, min_year=None, max_year=None)
         await step(gather_evidence_tool, question=question)
-        await step(generate_answer_tool, question=question)
+        await step(generate_answer_tool)
         return AgentStatus.SUCCESS
 
     return await _run_with_timeout_failure(rollout, query, env)

paperqa/agents/tools.py

@@ -268,19 +268,18 @@ class GenerateAnswer(NamedTool):
     summary_llm_model: LiteLLMModel
     embedding_model: EmbeddingModel
 
-    async def gen_answer(self, question: str, state: EnvironmentState) -> str:
+    async def gen_answer(self, state: EnvironmentState) -> str:
         """
-        Ask a model to propose an answer using current evidence.
+        Generate an answer using current evidence.
 
         The tool may fail, indicating that better or different evidence should be found.
         Aim for at least five pieces of evidence from multiple sources before invoking this tool.
         Feel free to invoke this tool in parallel with other tools, but do not call this tool in parallel with itself.
 
         Args:
-            question: Question to be answered.
             state: Current state.
         """
-        logger.info(f"Generating answer for '{question}'.")
+        logger.info(f"Generating answer for '{state.session.question}'.")
 
         if f"{self.TOOL_FN_NAME}_initialized" in self.settings.agent.callbacks:
             await asyncio.gather(
@@ -292,8 +291,6 @@ async def gen_answer(self, question: str, state: EnvironmentState) -> str:
                 )
             )
 
-        # TODO: Should we allow the agent to change the question?
-        # self.answer.question = query
         state.session = await state.docs.aquery(
             query=state.session,
             settings=self.settings,

tests/test_agents.py

@@ -404,7 +404,7 @@ async def test_gather_evidence_rejects_empty_docs(
 ) -> None:
 
     @wraps(GenerateAnswer.gen_answer)
-    async def gen_answer(self, question: str, state) -> str:  # noqa: ARG001
+    async def gen_answer(self, state) -> str:  # noqa: ARG001
         return "I cannot answer."
 
     # Patch GenerateAnswerTool.gen_answer so that if this tool is chosen first,
@@ -525,7 +525,7 @@ async def test_agent_sharing_state(
             summary_llm_model=summary_llm_model,
             embedding_model=embedding_model,
         )
-        result = await generate_answer_tool.gen_answer(answer.question, state=env_state)
+        result = await generate_answer_tool.gen_answer(state=env_state)
 
         if callback_type == "async":
             gen_answer_initialized_callback.assert_awaited_once_with(env_state)
@@ -655,24 +655,14 @@ def test_tool_schema(agent_test_settings: Settings) -> None:
             "info": {
                 "name": "gen_answer",
                 "description": (
-                    "Ask a model to propose an answer using current"
-                    " evidence.\n\nThe tool may fail, indicating that better or"
-                    " different evidence should be found.\nAim for at least five"
-                    " pieces of evidence from multiple sources before invoking this"
-                    " tool.\nFeel free to invoke this tool in parallel with other"
-                    " tools, but do not call this tool in parallel with itself."
+                    "Generate an answer using current evidence.\n\nThe tool may fail,"
+                    " indicating that better or different evidence should be"
+                    " found.\nAim for at least five pieces of evidence from multiple"
+                    " sources before invoking this tool.\nFeel free to invoke this tool"
+                    " in parallel with other tools, but do not call this tool in"
+                    " parallel with itself."
                 ),
-                "parameters": {
-                    "type": "object",
-                    "properties": {
-                        "question": {
-                            "type": "string",
-                            "description": "Question to be answered.",
-                            "title": "Question",
-                        }
-                    },
-                    "required": ["question"],
-                },
+                "parameters": {"type": "object", "properties": {}, "required": []},
             },
         },
     ]
@@ -752,7 +742,7 @@ async def test_deepcopy_env(self, agent_test_settings: Settings) -> None:
 
         # 3. Generate an answer for both, and confirm they are identical
         gen_answer_action = ToolRequestMessage(
-            tool_calls=[ToolCall.from_name("gen_answer", question=question)]
+            tool_calls=[ToolCall.from_name("gen_answer")]
         )
         _, _, done, _ = await env.step(gen_answer_action)
         assert done