LangChain

Review a tool-calling agent for loops, bad arguments, auth mistakes, and irreversible side effects, then add stops the runtime can actually enforce.

Design an offline and CI eval harness for an LLM feature with task metrics, slice bugs, and a promotion gate that prompt diffs cannot skip.