RepoJournal

$ cat openai/week/2026-08-31.log

OpenAI

OpenAI

the week in review · Aug 31 – Sep 6, 2026

Codex hardens sandbox and auth across platforms

By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology

♥

Two SDKs shipped 3.7.0 and 7.9.0 with usage API updates, plus Codex release 0.152.1.

374 commits 386 PRs merged 25 releases 7 briefings covered

all openai reviews →

What shipped: openai-python 3.7.0 and openai-node 7.9.0 both updated usage APIs and documentation [1][2], while openai-node fixed streaming hangs. Codex release 0.152.1 bundles several breaking changes, including update_plan becoming opt-in [3] and Guardian REPL policy sourcing from model metadata [4]. Windows authenticated sandbox provisioning went live [5], alongside MXC availability now recorded [6]. macOS sandbox was hardened against terminal input injection, and the SDKs gained canonical security models.

Why it matters: codex behavior shifts. update_plan tool is no longer called automatically; you must enable it explicitly. Guardian authorization and user answers now survive history compaction [7][8], but Guardian policy now comes from model metadata, which could change REPL decisions. If you run on Windows, sandbox provisioning is enabled and authenticated [5], and deferred environments recover after provisioning failure [9].

Under the hood: Codex added package-style MCP server names [10][11], scoped session MCP approvals to app account links [12], bound Git root discovery for metadata enrichment [13], and exposed loaded thread environments in app-server responses [14]. Agent navigation restores after TUI reconnects [15], and actionable rate-limit banners now appear in the TUI [16].

Also: Guardian compacted answers preserve user answers [17], Windows MXC availability recorded [6], and v3.7.0 python usage API updates [1], plus Codex release 0.152.1 [18].

References

  1. [1] v3.7.0 ↗ openai/openai-python
  2. [2] v7.9.0 ↗ openai/openai-node
  3. [3] Make the update_plan tool opt-in ↗ openai/codex
  4. [4] Source Guardian REPL policy from model metadata (#41919) ↗ openai/codex
  5. [5] Enable authenticated Windows sandbox provisioning (#42351) ↗ openai/codex
  6. [6] Record Windows MXC availability (#42094) ↗ openai/codex
  7. [7] Preserve Guardian authorization across history compaction (#41660) ↗ openai/codex
  8. [8] Preserve Guardian user answers across compaction (#41852) ↗ openai/codex
  9. [9] Recover deferred environments after provisioning failure ↗ openai/codex
  10. [10] Support package-style MCP server names (#41700) ↗ openai/codex
  11. [11] Support package-style MCP server names ↗ openai/codex
  12. [12] Scope session MCP approvals to app account links ↗ openai/codex
  13. [13] Bound Git root discovery for metadata enrichment ↗ openai/codex
  14. [14] Expose loaded thread environments in app-server responses ↗ openai/codex
  15. [15] Restore agent navigation after TUI reconnects ↗ openai/codex
  16. [16] Show actionable rate-limit banners in the TUI ↗ openai/codex
  17. [17] feat(api): update usage APIs and documentation ↗ openai/openai-node
  18. [18] 0.152.1 ↗ openai/codex

$ ls openai/week/ # the briefings behind this review

Keep up with OpenAI in about 3 minutes: 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 read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

all openai reviews →