Skip to content

The agent engineering platform.

docs(langchain): clarify create_tool_calling_agent system_prompt formatting and add troubleshooting (#33679)

Shanto Mathew2d30ebb53b2d
diff --git a/libs/langchain/langchain_classic/agents/tool_calling_agent/base.py b/libs/langchain/langchain_classic/agents/tool_calling_agent/base.py
index 1c8b23ae66..d23ccb68cf 100644
--- a/libs/langchain/langchain_classic/agents/tool_calling_agent/base.py
+++ b/libs/langchain/langchain_classic/agents/tool_calling_agent/base.py
@@ -83,11 +83,15 @@ def create_tool_calling_agent(
```
Prompt:
-
The agent prompt must have an `agent_scratchpad` key that is a
`MessagesPlaceholder`. Intermediate agent actions and tool output
messages will be passed in here.
+ Troubleshooting:
+ - If you encounter `invalid_tool_calls` errors, ensure that your tool
+ functions return properly formatted responses. Tool outputs should be
+ serializable to JSON. For custom objects, implement proper __str__ or
+ to_dict methods.
"""
missing_vars = {"agent_scratchpad"}.difference(
prompt.input_variables + list(prompt.partial_variables),