feat(langchain): rejection_response override for HumanInTheLoopMiddleware

[Mason Daugherty (mdrxy)]

Warning

This feature is alpha and subject to change or removal without notice.

Let InterruptOnConfig callers customize the message that gets sent back to the model when a human rejects a tool call. The default reply uses status="error", which some providers might treat as a transient failure and immediately retry, looping the interrupt. This lets you swap in your own copy and status without forking the middleware.

Changes

• New rejection_response option on InterruptOnConfig. Pass either a ToolMessage to reuse across tools (the middleware stamps the id and name for you) or a function that builds the message from the original tool call. Marked alpha.
• Default behavior is unchanged when the option is left unset.
• A factory that returns the wrong type or a ToolMessage with a mismatched id or name raises clearly, with the tool name in the error — so a typo surfaces immediately instead of confusing the model later.
• Catches a common mistake at setup: setting rejection_response without including "reject" in allowed_decisions now raises, instead of silently ignoring the override.
• Exports RejectionResponse and RejectionResponseFactory type aliases for users who want to type-annotate their own factories.

[codspeed-hq]

Merging this PR will not alter performance

✅ 2 untouched benchmarks
⏩ 13 skipped benchmarks¹

---

_{Comparing mdrxy/langchain-hitl-override (40ed1e8) with master (5a9b1ec)²}

Footnotes

1. 13 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on master (1c08d47) during the generation of this report, so 5a9b1ec was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩