RepoJournal
OpenAI OpenAI
123 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 2026-09-10
stories 129
RSS Archive Reviews Topics

© 2026 RepoJournal Home Showcase Explore How it works Privacy

$ the-wire · showcase

Codex 0.154.0 adds GPT-6-Astra and experimental worktrees, Python SDK stops crashing on null output

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

copyberry apcha-oai openai-sdks[bot] Shaurya Singh github-actions[bot] Abhinav rohitdeppi38
9 people shipped this
copyberry @copyberry 3 cited apcha-oai @apcha-oai 3 cited openai-sdks[bot] @openai-sdks[bot] 2 cited Shaurya Singh @LeSingh1 2 cited github-actions[bot] @github-actions[bot] 1 cited Abhinav @abhinav-oai 1 cited rohitdeppi38 @rohitdeppi38 1 cited HAYDEN-OAI @HAYDEN-OAI dependabot @dependabot

Codex 0.154.0 ships GPT-6-Astra in the model picker and Amazon Bedrock catalogs, experimental `--worktree` isolated checkouts, and inline question answering, while openai-python fixes a Responses parsing crash on `output: null` and openai-node adds API key expiration controls.

Codex 0.154.0 lands GPT-6-Astra in the model picker and Amazon Bedrock catalogs, and introduces experimental worktree support: `--worktree` or `/worktree` creates isolated checkouts for new or forked sessions that you can browse and resume later [1]. Sessions can now also answer questions inline while Codex keeps working, using suggested choices or custom text without losing the main draft [1]. Windows sessions can share a background something the release notes truncate, so treat that line as incomplete [1].

Session-start hooks get a correctness fix. Forked threads used to report `startup`, which caused startup hooks to run again even when their context was inherited from the parent, and resuming with supplied history also reported `startup` instead of `resume` [2]. The commit adds `fork` as a `SessionStart` source and exposes it in the hook input schema, reporting `fork` for histories with a fork parent and `resume` for supplied histories without one [2]. If you match on `SessionStart` source values, add the new case. This is a breaking change for hook logic that assumed only `startup` and `resume`.

On the Python side, `openai-python` 3.11.0 adds expiration controls for service account keys [3]. The pipeline commit behind it fixes a real crash: Responses parsing raised `TypeError` when a response contained `output: null` [4]. Null output is now treated as empty for standalone parsing, and finalized `response.output_item.done` items are recovered when a streaming completion has null or missing output, preserving item order, final statuses, annotations, refusals, structured output, tool arguments, and completion metadata [4][5]. Explicit completion output, including `[]`, remains authoritative [5]. A separate fix handles bare `dict` and `list` annotations without type arguments, where `transform()` and `construct_type()` indexed unconditionally into `get_args()` and raised on the empty tuple [6].

`openai-node` 7.13.0 mirrors the same feature, adding API key expiration controls [7][8]: `expires_in_seconds` on service-account creation parameters for the initial key lifetime, plus optional `expires_at` timestamps on project API keys and newly created service-account keys [8]. CI also got attention on both SDKs after the dependency-security checker landed. The Node test jobs had grown from 1m32s in the August 12 reference run to about 5m38s on current main, with the regular test step going from 32s to 224 to 228s [9]; the fix compiles example, ecosystem CLI, and SDK sources once per Vitest invocation so each subprocess test runs prebuilt JavaScript in a fresh Node process [9]. The Python test job had jumped from 2m24s to 7m00s on Python 3.14 after the policy regression matrix landed, and is back to 2m41s, with Pydantic v2 pytest going from 192.88s to 62.75s [10]. On the Codex orchestration side, synchronous Guardian review moved into `codex-guardian-reviewer`, with core supplying adapters for evidence capture, authorization checks, session creation, and event publication [11], and remote compaction now always routes through the streamed v2 implementation, retiring the `remote_compaction_v2` feature toggle while still advertising it in the model client beta features header [12]. The Windows `/sandbox-add-read-dir` slash command is gone, along with its event handlers and non-elevated read-root grant helper [13].

Action items

References

  1. [1] 0.154.0 ↗ openai/codex
  2. [2] Distinguish forked sessions in session-start hooks (#44349) ↗ openai/codex
  3. [3] v3.11.0 ↗ openai/openai-python
  4. [4] fix: preserve finalized output on null response completion (#3345) ↗ openai/openai-python
  5. [5] fix: preserve finalized output on null response completion ↗ openai/openai-python
  6. [6] fix: handle bare `dict` and `list` annotations without type arguments ↗ openai/openai-python
  7. [7] v7.13.0 ↗ openai/openai-node
  8. [8] feat(api): Add API key expiration controls ↗ openai/openai-node
  9. [9] test: reduce CI subprocess and validation overhead ↗ openai/openai-node
  10. [10] test: restore CI test runtime after dependency-policy regression (#3830) ↗ openai/openai-python
  11. [11] Move synchronous Guardian orchestration into the reviewer extension ↗ openai/codex
  12. [12] Always use streamed remote compaction for supported providers ↗ openai/codex
  13. [13] Remove the Windows `/sandbox-add-read-dir` slash command ↗ openai/codex

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

the week in review openai-python 3.14.0 and WebSocket sessions land the month in review August: Codex hardens, SDKs consolidate security, GPT-5.5 lands
Topics: AI / ML Full archive → Weekly reviews →

Quick answers

What shipped in OpenAI on September 10, 2026?
Codex 0.154.0 ships GPT-6-Astra in the model picker and Amazon Bedrock catalogs, experimental `--worktree` isolated checkouts, and inline question answering, while openai-python fixes a Responses parsing crash on `output: null` and openai-node adds API key expiration controls. In total, 60 commits, 63 pull requests, and 6 releases landed.
Who contributed to OpenAI on September 10, 2026?
9 developers shipped this update, including github-actions[bot], Abhinav, copyberry, openai-sdks[bot], apcha-oai, HAYDEN-OAI, dependabot, and Shaurya Singh, and 1 more.
What were the notable OpenAI updates?
0.154.0, Distinguish forked sessions in session-start hooks (#44349), and v3.11.0.
♥

$ log OpenAI --recent

Codex consolidates PTY spawning, openai-node restores options-only GET calls 87 shipped Codex grows a backend-independent agent control contract as the SDKs land vault credentials and deprecate MCP connector_id 140 shipped openai-node and openai-python ship managed Responses WebSocket sessions 124 shipped Codex enforces managed residency, openai-node fixes streaming and abort paths 142 shipped

$ log --all # elsewhere on the wire

@openclaw Cold-start Gateway diagnostics stop crying wolf, doctor repairs plugin drift Agentic Coding Cline desktop bundles shrink, goose unblocks its own clicks

$ ls showcases/

Rails Django Laravel Node.js Go Vercel Vue.js Kubernetes HashiCorp Arch Linux Linux Anthropic

$ 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.