RepoJournal
OpenAI OpenAI
150 wires and counting

$ follow OpenAI

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.

$ status

wire 2026-05-21
stories 103
RSS Archive Reviews

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

SANDBOX ENFORCEMENT STACK LANDS: Windows permissions now runtime-resolved across three major migrations

By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology

bolinfest evawong-oai abhinav-oai caseychow-oai
4 people shipped this
bolinfest @bolinfest 2 cited evawong-oai @evawong-oai 1 cited abhinav-oai @abhinav-oai 1 cited caseychow-oai @caseychow-oai 1 cited

Codex shipped the final piece of its Windows sandbox permission architecture, moving from legacy enum questions to resolved runtime capability checks across setup, spawn, and enforcement layers.

The sandbox `SandboxPolicy` to `PermissionProfile` migration reached critical mass overnight with three interconnected PRs landing in sequence [1] [2] [3]. PR #22923 completes the setup/spawn helper migration, replacing questions like 'is this WorkspaceWrite?' with 'does this profile require write roots?' [1]. Meanwhile, MITM hook enforcement wired into the request path [2], enforcing hooked HTTPS hosts to require MITM, evaluating inner requests post-CONNECT, and blocking unmatched hooks. On the lifecycle side, `SessionStart` hooks now support compaction by queuing pending states alongside compact rewrites, letting durable context re-inject after conversation history replacement [3]. Plugin creator tooling gained a dedicated personal-marketplace update flow for iterating on existing local plugins [4], keeping scaffold paths intact while making the development loop explicit. Finally, `codex exec-server` now accepts `--strict-config` validation [5], closing the gap left when earlier commands got fast-fail support for misspelled keys.

Action items

References

  1. [1] windows-sandbox: drive write roots from resolved permissions ↗ openai/codex
  2. [2] Wire MITM hooks into runtime enforcement ↗ openai/codex
  3. [3] Support compact SessionStart hooks ↗ openai/codex
  4. [4] [skills] Create a personal update flow for plugin creator ↗ openai/codex
  5. [5] cli: add strict config to exec-server ↗ openai/codex

$ ls openai/reviews/ # the step-back reads

the week in review Codex hands model defaults to the server the month in review Agents API lands in both OpenAI SDKs
Full archive → Weekly reviews →

Quick answers

What shipped in OpenAI on May 21, 2026?
Codex shipped the final piece of its Windows sandbox permission architecture, moving from legacy enum questions to resolved runtime capability checks across setup, spawn, and enforcement layers. In total, 50 commits, 50 pull requests, and 3 releases landed.
Who contributed to OpenAI on May 21, 2026?
4 developers shipped this update, including bolinfest, evawong-oai, abhinav-oai, and caseychow-oai.
What were the notable OpenAI updates?
windows-sandbox: drive write roots from resolved permissions, Wire MITM hooks into runtime enforcement, and Support compact SessionStart hooks.
♥

$ log OpenAI --recent

Safety identifiers reach images, plugin approvals bind to contents 93 shipped Credential masking lands in Codex, Live gains Decisions across both SDKs 100 shipped Codex stabilizes environment tools, python adds residency options 135 shipped openai-python restores streaming connection reuse, browser and computer toolsets land in both SDKs 123 shipped

$ log --all # elsewhere on the wire

Alignment & Post-training verl rejects conflicting vLLM engine_kwargs, open-instruct adds MILES OPD behind a Megatron learner Interpretability nnsight 0.8.0 cuts import time from 7.6s to 1.6s

$ ls showcases/

Safety Evals AI Red Teaming Agent Security Interpretability Alignment & Post-training

$ init your-desk

RepoJournal can write one of these for any repo you care about. Sign in with GitHub, pick your repos and sources, and your desk gets a wire like this one every morning.