fix(langchain/createAgent): don't modify messages in PromptCachingMiddleware, return new objects

[christian-bromann]

Prompt caching is only possible for 5 messages at the same time. The current version of the middleware updates the message object in directly rather than creating a new object. This causes messages to stay cached over multiple hook iterations and eventually failing due to the limit of 5.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 463650d

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
langchainjs-docs	Ready	Preview	Comment	Oct 1, 2025 4:03am

1 Skipped Deployment

Project	Deployment	Preview	Comments	Updated (UTC)
langchainjs-api-refs	Ignored			Oct 1, 2025 4:03am

[hntrl]

Do we know if the cache_control breakpoint only caches the content contained within a single message? Anthropic docs make it a little ambiguous..

The system automatically checks for cache hits at all previous content block boundaries

If we're only adding the cache_control param to the last message, and given that cache breakpoints are contained within a single message, then we aren't caching much of anything

[christian-bromann]

Do we know if the cache_control breakpoint only caches the content contained within a single message?

As far as I understood it caches the message flagged with the cache property and all previous ones.

then we aren't caching much of anything

In Python we do the same here. The integration test also checks that we actually see cached tokens.