100 wires and counting

$ follow OpenAI

Keep up with OpenAI in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good morning, this isn't your newsletter.

One email a day. Unsubscribe in one click.

$ status

wire 2026-08-28
stories 132

© 2026 RepoJournal Home Showcase Explore How it works Privacy

$ the-wire · showcase

OpenAI's codex cuts locking, Guardian survives long sessions

By RepoJournal · Filed · About OpenAI

Codex rolled over its Guardian context to keep long reviews alive, while the Node SDK finally lets browsers import natively.

Codex is getting smarter about long-running reviews: Guardian sessions now compact before follow-up reviews, preventing context exhaustion on long-lived sessions [1]. That's a quiet but critical fix for anyone running code review agents over large PRs. Meanwhile, trusted skill collection dropped its mutex, storing paths in a BTreeSet for cleaner per-turn handling [2]. The deny-read matcher now aligns with executor path semantics, so URI-based policies and filesystem enumeration enforce the same rules [3]. Over on the Node SDK, native browser ESM imports are restored, and the private X.509 state package now resolves to relative imports for direct browser loading [4]. The SDK also hardens the credential-free ecosystem checks: webpack bundles, direct browser imports, and keyless checks are now required on release PRs [5][6][7]. For Python, the Agents SDK downstream check is gone, matching the Node repo's move to keep SDK CI unblocked [8]. And the auth layer hardened X.509 workload identity across sync and async clients, with "Workload-identity regression coverage" [9]. Dependabot also lands for Python and Actions with security-aware config, Monday UTC updates, and an eight-day cooldown [10].

Action items

References

  1. [1] Roll over Guardian context before follow-up reviews ↗ openai/codex
  2. [2] Remove locking from trusted skill collection (#41150) openai/codex
  3. [3] Align deny-read matching with executor path semantics ↗ openai/codex
  4. [4] fix: restore native browser ESM imports (#2495) openai/openai-node
  5. [5] ci: require keyless ecosystem checks on release PRs (#2497) openai/openai-node
  6. [6] test: execute webpack browser bundle without live credentials (#2498) openai/openai-node
  7. [7] test: exercise direct browser imports without credentials (#2501) openai/openai-node
  8. [8] ci: remove the Agents SDK downstream check (#3750) openai/openai-python
  9. [9] fix(auth): harden X.509 workload identity integration ↗ openai/openai-python
  10. [10] ci: add security-aware Dependabot updates for Python and GitHub Actions (#3641) openai/openai-python

Quick answers

What shipped in OpenAI on August 28, 2026?
Codex rolled over its Guardian context to keep long reviews alive, while the Node SDK finally lets browsers import natively. In total, 66 commits, 61 pull requests, and 5 releases landed.
Who contributed to OpenAI on August 28, 2026?
6 developers shipped this update, including rka-oai, copyberry, jif, Justin Beckwith, Hayden, and Alex Chang.
What were the notable OpenAI updates?
Roll over Guardian context before follow-up reviews, Remove locking from trusted skill collection (#41150), and Align deny-read matching with executor path semantics.