RepoJournal
OpenAI

OpenAI

Codex, the SDKs, and the engine behind ChatGPT

Pick a date

  1. Mon 25 may
  2. Tue 26 may
  3. Wed 27 may
  4. Thu 28 may
  5. Fri 29 may
  6. Sat 30 may
  7. Sun 31 may
  8. Mon 1 jun
  9. Mon 8 jun
  10. Tue 9 jun
  11. Wed 10 jun
  12. Thu 11 jun
  13. Fri 12 jun
  14. Sat 13 jun
  15. Sun 14 jun
  16. Mon 15 jun
  17. Tue 16 jun
  18. Wed 17 jun
  19. Thu 18 jun
  20. Fri 19 jun
  21. Sat 27 jun
  22. Sun 28 jun
  23. Mon 29 jun
  24. Tue 30 jun
  25. Wed 1 jul
  26. Thu 2 jul
  27. Fri 3 jul
  28. Sat 4 jul
  29. Sun 5 jul
  30. Today 7 jul

The Wire · Showcase

CODEX FIXES PROXY ATTRIBUTION BUG, NODE SDK PATCHES STREAM CORRUPTION

By RepoJournal · Filed · About OpenAI

Concurrent network requests in Codex conversations were getting routed to the wrong executors, and the Node SDK was silently corrupting streamed assistant arrays.

The Codex team shipped a critical fix [1] that attributes network requests to the exact executor on Linux, solving a proxy routing problem where concurrent curl calls within the same conversation couldn't be distinguished by the Guardian proxy. This surfaces a broader theme across both desks: data integrity under concurrent conditions. The Node SDK hit the same wall [6] when streaming assistant array deltas, where out-of-order indices were being appended instead of assigned, landing entries at wrong positions and corrupting state. Both fixes are small but ship against the kind of bugs that only surface under load, so they matter. On the Codex side, the team also moved Apps guidance into persistent world state [2], ensuring the model gets proper tool instructions even when the MCP recovers mid-turn, and shipped canonical command execution items [3] to standardize the event model across app-server v2. The Node SDK backfilled two more safety patches: Next.js got upgraded to 15.5.16 to close a flagged CVE in examples [5], and the streaming layer now treats keepalive events as no-ops instead of crashing the accumulator [7]. Autocomplete insertion in Codex also got stricter token range handling [4] to prevent cursor position ambiguity from replacing the wrong token.

Action items

References

  1. [1] fix: attribut network requests to the exact exec on linux (#29697) openai/codex
  2. [2] Make Apps guidance react to MCP availability (#30226) openai/codex
  3. [3] feat(core): emit canonical command execution items ↗ openai/codex
  4. [4] Use popup token ranges for autocomplete insertion ↗ openai/codex
  5. [5] fix: upgrade next to 15.5.16 in examples ↗ openai/openai-node
  6. [6] fix(assistants): place array delta entries by index instead of appending ↗ openai/openai-node
  7. [7] Handle keepalive events in Responses streams ↗ openai/openai-node

FAQ

What changed in OpenAI on July 7, 2026?
Concurrent network requests in Codex conversations were getting routed to the wrong executors, and the Node SDK was silently corrupting streamed assistant arrays.
What should OpenAI teams do about it?
Pull Codex fix for proxy attribution if you run multi-exec conversations with network calls • Upgrade openai-node to get the assistant streaming delta fix before processing long arrays • Review Apps MCP availability handling in your Codex integrations - behavior changed
Which OpenAI repositories shipped on July 7, 2026?
openai/codex, openai/openai-node

Related across the cluster

For your repos

The showcase is a teaser.
Your wire is the product.

Same engine. Different stack. Below: what changes when the wire is yours.

Showcase wire

  • •14 famous open source orgs
  • •One wire per day
  • •Public, generic
  • •Read on the web, when you remember

Your wire

  • →Up to 1,500 of your repos - orgs, deps, vendors
  • →Morning and evening briefs
  • →Action items routed to your team
  • →Slack delivery, email, breaking-news CVE alerts

Want a hands-on demo first? Ask a current user for an invite link.