fix(openai): preserve managed request metadata

[usnavy13]

Summary

Preserve the managed OpenAI request metadata that LibreChat needs for reliable explicit caching and tier-aware accounting.

• Forward the server-generated prompt cache key through OpenAI and Azure Chat Completions and Responses requests.
• Keep the instruction breakpoint and the two latest prior-history breakpoints, within OpenAI's marker limit.
• Preserve provider-reported service_tier on non-streaming Chat Completions messages.
• Keep the current user turn unmarked.
• Cover Chat Completions and Responses rollover behavior.

Why

Live sequential-request testing showed that replacing the prior history marker on every turn could produce another cache write instead of a cache read. LibreChat's generated prompt cache key also needed to be represented in the shared managed-request options so it reaches all four OpenAI/Azure transport paths. Non-streaming Chat Completions dropped the response-level service tier during LangChain conversion, forcing downstream accounting to infer the requested tier instead of using the provider-reported tier.

Change Type

• Bug fix (non-breaking change which fixes an issue)

Testing

• npx jest src/llm/openai/managedRequests.test.ts src/llm/openai/llm.spec.ts --runInBand — 71 tests passed.
• npx tsc --noEmit — passed.
• npm run build — passed.
• ESLint on all changed files — passed.
• git diff --check — passed.

Test Configuration

• Rebased onto current Agents main at v3.7.1.
• Node.js 24.x.

Checklist

• My code adheres to this project's style guidelines
• I have performed a self-review of my own code
• My changes do not introduce new warnings
• I have written tests demonstrating that my changes are effective
• Local unit tests and build pass with my changes

[usnavy13]

Marked ready for review as the small follow-up to danny-avila/agents#303.

It fixes provider-tier metadata and cache-breakpoint retention needed by LibreChat model support, independently of native PTC. Because the Agents CI is scoped to PRs targeting main or dev, this stacked PR will receive its full CI after #303 lands and this branch is rebased onto main. It will then be included in the first published Agents version.

[danny-avila]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4826739647

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[usnavy13]

Follow-up update: this branch is now rebased directly onto current main (v3.2.65). The valid cache-breakpoint rollover review finding is fixed, with an exact regression test, and I replied on the inline thread.

Local validation on the rebased single-commit head: 59 focused tests, TypeScript, ESLint for the changed files, and the full Agents build all pass.

[danny-avila]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 05f80e379f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[usnavy13]

Addressed the latest review in b399e88 after rebasing onto main v3.2.67.

Cache-breakpoint selection now exits as soon as it has the two newest assistant/tool completion candidates. Added a long-history regression confirming the selected breakpoints.

Validation: focused tests, TypeScript, targeted ESLint, and the full production build all pass.

[usnavy13]

Rebased this focused follow-up onto current Agents main (v3.7.1). The patch is unchanged in substance and remains cleanly mergeable. Revalidated: 71 focused tests, TypeScript, production build, changed-file ESLint, and diff whitespace checks all pass.