Extract _prepare_trimmed_messages() to prevent create_trimmed/acreate_trimmed drift

[dhruvbatra]

Summary

• create_trimmed and acreate_trimmed both called trimmed_messages_to_fit with identical arguments and discarded the same return values — if the trim logic ever needs updating, both functions need the same edit
• Extracted the shared trim step into _prepare_trimmed_messages(), a private pure function that both callers delegate to
• No behavioral change — identical on-wire messages, same parameters

Why it's safe

Pure extraction of identical code. Both functions produce the same trimmed_messages list as before. No public API changes.

Test plan

• Verify _prepare_trimmed_messages returns the same list as the previous inline trimmed_messages_to_fit call
• create_trimmed and acreate_trimmed still pass trimmed messages to completions.create

---

Generated by Claude Code

---

Note

Low Risk
Pure refactor with identical trim parameters and outputs; no public API or runtime behavior changes.

Overview
Refactors create_trimmed and acreate_trimmed in loop.py so they no longer each inline the same trimmed_messages_to_fit call.

A new private helper _prepare_trimmed_messages runs that trim step once (same max_bytes / keep_recent defaults) and returns only the message list; both completion helpers pass its result to completions.create. No public API or on-wire behavior change—only deduplication to avoid sync/async drift when trim logic changes.

^{Reviewed by Cursor Bugbot for commit 35d7b12. Bugbot is set up for automated code reviews on this repo. Configure here.}