[sernote]

The Bedrock OpenAI summary calls the limit 4 checkpoints, while AWS defines up to four cache writes per request. Replace only that clause with 4 writes max, preserving the distinction from Claude's checkpoint limit. AWS prompt-caching contract.

Value: remove a literal contradiction that could cause unsupported marker-count remediation in a five-marker GPT-5.6 audit. The compact wording keeps the existing budget when combined with PR #42: 381259 characters / 95315 estimated tokens, within the unchanged ceiling.

Validation on 8ad87fe3451f41790c767ac75c50a21a3172c007:

• 260 unit tests, package validator, 40 trigger cases, Python syntax, whitespace and bytecode-absence checks passed.
• Independent targeted final review: PASS; byte comparison confirms exactly one clause changed.
• Static before/final comparison: score 59 unchanged, all token-budget deltas zero, no new failures; existing static findings remain.
• Two frozen advisory cases once before/after retained the correct OpenAI writes versus Claude checkpoints distinction. Baseline was already correct, so no behavioral uplift is claimed. The final shorter equivalent was checked separately by source inspection and independent review without replaying correct scenarios. Ancillary advisory hypotheses do not establish runtime acceptance, writes, hits or savings.

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-08T17:06:18.414666Z	8ad87fe	PR opened

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.