[lia-by-librechat]

Summary

• Preserve Responses cache-write TTL subsets as cache_creation_12h and cache_creation_30m in normalized usage.
• Keep aggregate cache creation and token totals unchanged, retain user metadata, and remove internal bridge keys.
• Restore custom TTL details after chunk concatenation so final messages and handleLLMEnd callbacks agree.
• At the Langfuse boundary, keep the aggregate cache bucket without overlapping TTL subsets. Public usage retains its TTL details; uncached input is no longer subtracted twice.

Verification

• npx tsc --noEmit: passed.
• Touched-file ESLint and import-order checks: passed.
• Cache usage, summarization usage, stream metadata, provider smoke, Langfuse, masking, and deterministic trace ID tests: 332 passed across 17 suites.
• npm run build: passed, including declaration generation.

No version bump or release. Cyber integration remains blocked until a maintainer publishes the required dependency.

Title emoji: no ⏲️ occurrences across 1,915 agents main commit subjects.

[lia-by-librechat]

Verified head: 882b144c20b099e0ade0e462304f216a6710d740.

Normalized usage exposes input_token_details.cache_creation_12h and cache_creation_30m. OpenAI and Azure mock-fetch invocation tests cover final messages and actual handleLLMEnd callbacks for streaming and non-streaming calls, explicit streams, callback-preferred streaming, zero counts, and missing optional usage. TTL subsets do not change aggregate cache creation or token totals.

Local checks passed: npx tsc --noEmit; touched-file ESLint and import-order checks; npm run build including declarations; 115 focused tests; 205 tracing tests. Prettier applied to changed code, with unrelated baseline formatting preserved.

Not run locally: full library suite or live-provider API suites. Independent review is being coordinated by the parent task. CI status will be reported separately.

No version bump, merge, publish, or release.

[danny-avila]

@codex review the latest head, final review

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-10T17:35:49.685610Z	882b144	Manual request

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Delightful!

Reviewed commit: 882b144c20

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".

[lia-by-librechat]

Head: 176b34408e66956142070b07e9cce716cbe4d82c. Fixes review finding SDK-1 (P2): Langfuse now receives disjoint cache usage without changing public TTL details or the source messages. Regression tests exercise actual Langfuse callbacks for messages/chunks, partial/zero TTLs and masked generations. All 332 focused SDK/tracing tests, tsc --noEmit, scoped ESLint/import checks and the declaration build pass. A fresh independent review is running for this exact head. No SDK version bump or release.

[lia-by-librechat]

Independent review completed for 176b34408e66956142070b07e9cce716cbe4d82c: no findings. SDK-1 (P2, overlapping Langfuse TTL buckets) is confirmed fixed. Public TTL details are preserved while the export boundary avoids double counting. Local verification: 332 focused tests, typecheck, scoped lint/import checks and declaration build passed. No version bump or release was performed.