Reported evidence
LangChain → changes → Invalid tool calls
LangChain 1.4.3 repairs invalid tool calls in create_agent; langchain-anthropic 1.7.5 changes their representation during replay. These are changes at different processing points, not evidence that resulting actions are correct or authorized.
Read the evidence
Two LangChain changes provide a useful contrast: one repairs invalid tool calls in create_agent, while the other adjusts the message representation of invalid calls during replay.
[S9] LangChain [S10] LangChain
Read the full research