After a human edits a tool call, the agent runs the original call anyway

With HumanInTheLoopMiddleware, choosing 'edit' runs the edited tool call, but on the next step the agent issues the original, unedited call again.

Difficulty

Easy

Time

~5 min

Steps

Fixed in

langchain==1.4.2

Last checked

Oct 17, 2026

Verified fix: PR #40463 merged on 2026-09-18. Fixed in langchain==1.4.2 (first stable release containing the merge commit). Last checked 2026-10-17.

Symptom

With HumanInTheLoopMiddleware, choosing 'edit' runs the edited tool call, but on the next step the agent issues the original, unedited call again.

Evidence what you'll see

The edited call executes first
A new AI message then repeats the original tool call
Happens with the edit decision, not approve

Cause

The edit rewrote the model's own tool call in history, so the model saw a call it didn't make and a result that didn't match the request, and tried again.

Fix steps

0 of 1 done

Step 1Upgrade langchain to a release that includes PR #40463, which keeps the model's original tool call in history and adds a note to the tool result explaining the edit.

Your check marks stay in this browser only.

Verify the fix worked

Edit a tool call in the middleware and confirm the agent continues instead of re-issuing the original call.

Sources and credit

Summarized in our own words from these public reports. Credit goes to their authors.

Provided as is, without warranty. Try changes in a non-production environment and back up first. Security-related steps: verify against the official docs. Terms · Request removal or attribution change

Hit an error we don't cover? Submit it