fix: react agent (#200)

Summary:
1. update prompt to match the json response object
2. call tools with tool_args when present (e.g. vector_db_ids for rag
tool)

Test Plan:
Run react agent with rag tool
```

react_agent_with_vector_db = ReActAgent(
    client=client,
    model=model_id,
    tools=[
        {
            "name": "builtin::rag",
            "args": {"vector_db_ids": [vector_db_id]},
        }
    ],
)
react_agent_with_vector_db_config = react_agent_with_vector_db.agent_config
react_tool_parser = react_agent_with_vector_db.tool_parser
react_agent_client_tools = react_agent_with_vector_db.client_tools

results_vector_db_agent = evaluator_no_context.run(
    react_agent_with_vector_db_config,
    client,
    num_examples=100,
    client_tools=[],
    tool_parser=react_tool_parser,
    json_response_format=True,
    # answer_parsing_fn=lambda x: json.loads(x)["answer"],
)
```

Author: ehhuang

src/llama_stack_client/lib/agents/agent.py

@@ -188,8 +188,8 @@ def initialize(self) -> None:
         )
         self.agent_id = agentic_system_create_response.agent_id
         for tg in self.agent_config["toolgroups"]:
-            for tool in self.client.tools.list(toolgroup_id=tg):
-                self.builtin_tools[tool.identifier] = tool
+            for tool in self.client.tools.list(toolgroup_id=tg if isinstance(tg, str) else tg.get("name")):
+                self.builtin_tools[tool.identifier] = tg.get("args", {}) if isinstance(tg, dict) else {}
 
     def create_session(self, session_name: str) -> str:
         agentic_system_create_session_response = self.client.agents.session.create(
@@ -225,7 +225,7 @@ def _run_tool(self, tool_calls: List[ToolCall]) -> ToolResponseParam:
         if tool_call.tool_name in self.builtin_tools:
             tool_result = self.client.tool_runtime.invoke_tool(
                 tool_name=tool_call.tool_name,
-                kwargs=tool_call.arguments,
+                kwargs={**tool_call.arguments, **self.builtin_tools[tool_call.tool_name]},
             )
             tool_response = ToolResponseParam(
                 call_id=tool_call.call_id,
@@ -411,7 +411,7 @@ async def initialize(self) -> None:
         self._agent_id = agentic_system_create_response.agent_id
         for tg in self.agent_config["toolgroups"]:
             for tool in await self.client.tools.list(toolgroup_id=tg):
-                self.builtin_tools[tool.identifier] = tool
+                self.builtin_tools[tool.identifier] = tg.get("args", {}) if isinstance(tg, dict) else {}
 
     async def create_session(self, session_name: str) -> str:
         await self.initialize()
@@ -462,7 +462,7 @@ async def _run_tool(self, tool_calls: List[ToolCall]) -> ToolResponseParam:
         if tool_call.tool_name in self.builtin_tools:
             tool_result = await self.client.tool_runtime.invoke_tool(
                 tool_name=tool_call.tool_name,
-                kwargs=tool_call.arguments,
+                kwargs={**tool_call.arguments, **self.builtin_tools[tool_call.tool_name]},
             )
             tool_response = ToolResponseParam(
                 call_id=tool_call.call_id,

src/llama_stack_client/lib/agents/react/prompts.py

@@ -22,7 +22,7 @@
 
 The `action` key should specify the $TOOL_NAME the name of the tool to use and the `tool_params` key should specify the parameters key as input to the tool.
 
-Make sure to have the $TOOL_PARAMS as a dictionary in the right format for the tool you are using, and do not put variable names as input if you can find the right values.
+Make sure to have the $TOOL_PARAMS as a list of dictionaries in the right format for the tool you are using, and do not put variable names as input if you can find the right values.
 
 You should always think about one action to take, and have the `thought` key contain your thought process about this action.
 If the tool responds, the tool will return an observation containing result of the action. 
@@ -37,7 +37,7 @@
     "thought": "I need to transform the image that I received in the previous observation to make it green.",
     "action": {
         "tool_name": "image_transformer",
-        "tool_params": {"image": "image_1.jpg"}
+        "tool_params": [{"name": "image"}, {"value": "image_1.jpg"}]
     },
     "answer": null
 }
@@ -61,7 +61,7 @@
     "thought": "I will proceed step by step and use the following tools: `document_qa` to find the oldest person in the document, then `image_generator` to generate an image according to the answer.",
     "action": {
         "tool_name": "document_qa",
-        "tool_params": {"document": "document.pdf", "question": "Who is the oldest person mentioned?"}
+        "tool_params": [{"name": "document"}, {"value": "document.pdf"}, {"name": "question"}, {"value": "Who is the oldest person mentioned?"}]
     },
     "answer": null
 }
@@ -73,7 +73,7 @@
     "thought": "I will now generate an image showcasing the oldest person.",
     "action": {
         "tool_name": "image_generator",
-        "tool_params": {"prompt": "A portrait of John Doe, a 55-year-old man living in Canada."}
+        "tool_params": [{"name": "prompt"}, {"value": "A portrait of John Doe, a 55-year-old man living in Canada."}]
     },
     "answer": null
 }
@@ -93,7 +93,7 @@
     "thought": "I will use python code evaluator to compute the result of the operation and then return the final answer using the `final_answer` tool",
     "action": {
         "tool_name": "python_interpreter",
-        "tool_params": {"code": "5 + 3 + 1294.678"}
+        "tool_params": [{"name": "code"}, {"value": "5 + 3 + 1294.678"}]
     },
     "answer": null
 }
@@ -113,7 +113,7 @@
     "thought": "I need to get the populations for both cities and compare them: I will use the tool `search` to get the population of both cities.",
     "action": {
         "tool_name": "search",
-        "tool_params": {"query": "Population Guangzhou"}
+        "tool_params": [{"name": "query"}, {"value": "Population Guangzhou"}]
     },
     "answer": null
 }
@@ -124,7 +124,7 @@
     "thought": "Now let's get the population of Shanghai using the tool 'search'.",
     "action": {
         "tool_name": "search",
-        "tool_params": {"query": "Population Shanghai"}
+        "tool_params": [{"name": "query"}, {"value": "Population Shanghai"}]
     },
     "answer": null
 }