Who contributed to OpenAI on August 5, 2026?

5 developers shipped this update, including efrazer-oai, Eric Traut, copyberry, HAYDEN-OAI, and Justin Beckwith.

What were the notable OpenAI updates?

Prompt before trusting local project directories (#36960), Accept user input when starting idle turns (#36947), and Keep image resize notices attached during remote compaction.

OpenAI

Codex, the SDKs, and the engine behind ChatGPT

Pick a date

Topics: AI / ML Full archive →

The Wire · Showcase

CODEX HARDENS SECURITY, SDKS MODERNIZE TOOLCHAINS

By RepoJournal · Filed 06:58 UTC on August 5, 2026 · About OpenAI

5 people shipped this

HAYDEN-OAI @HAYDEN-OAI 3 cited

efrazer-oai @efrazer-oai 1 cited

Eric Traut @etraut-openai 1 cited

copyberry @copyberry 1 cited

Justin Beckwith @jbeckwith-oai 1 cited

Codex now prompts users before trusting local directories, closing a prompt injection exposure, while both Node and Python SDKs shift to Oxlint and Oxfmt tooling.

The biggest move on the Codex desk is a security hardening that makes explicit trust mandatory for project directories [1]. Previously, Codex would automatically trust unset directories, which exposed users to prompt injection through project-local config, hooks, and exec policies. The fix adds a directory-trust step to TUI onboarding with a clear choice: trust and continue or quit. Separately, Codex now accepts user input when starting idle turns [2], generalizing the `CodexThread::try_start_turn_if_idle` method to preserve user content and client IDs in the first model request. The platform also ships improved handling for image resize notices during remote compaction [3], ensuring notices stay grouped with their source history items when trimming to a token budget.

On the SDK front, both openai-node and openai-python have completed migrations away from the ESLint/Prettier stack. The Node SDK upgraded TypeScript to 6.0.3 and replaced deprecated `moduleResolution: "node"` with `"bundler"` [4], then adopted Oxlint 1.75.0 and Oxfmt 0.60.0 while preserving all existing formatting rules and exclusions [5]. It layered Ultracite 7.8.4 on top as a shared workflow configuration [6]. The Python SDK followed the same Release Please pattern as Node [7]: both now mint repository-scoped installation tokens for the openai-sdks GitHub App instead of relying on GITHUB_TOKEN, which reduces exposure when handling release automation and pull-request events. These changes land across 55 commits and 55 PRs across the three repos over the last 24 hours.

Action items

→ Review and merge the directory-trust PR for Codex deployments openai/codex [immediate]
→ Validate TypeScript 6.0.3 and Oxlint/Oxfmt behavior in Node SDK tests openai/openai-node [plan]
→ Monitor Release Please GitHub App token rollout for both SDKs openai/openai-node [monitor]

References

[1] Prompt before trusting local project directories (#36960) openai/codex
[2] Accept user input when starting idle turns (#36947) openai/codex
[3] Keep image resize notices attached during remote compaction ↗ openai/codex
[4] [SDK-201] Replace deprecated module resolution settings ↗ openai/openai-node
[5] [SDK-198] Migrate ESLint and Prettier to Oxlint and Oxfmt ↗ openai/openai-node
[6] [SDK-223] Adopt Ultracite for Oxlint and Oxfmt (#2057) openai/openai-node
[7] ci: use GitHub App for Release Please ↗ openai/openai-python

Quick answers

What shipped in OpenAI on August 5, 2026?: Codex now prompts users before trusting local directories, closing a prompt injection exposure, while both Node and Python SDKs shift to Oxlint and Oxfmt tooling. In total, 55 commits, 55 pull requests, and 4 releases landed.
Who contributed to OpenAI on August 5, 2026?: 5 developers shipped this update, including efrazer-oai, Eric Traut, copyberry, HAYDEN-OAI, and Justin Beckwith.
What were the notable OpenAI updates?: Prompt before trusting local project directories (#36960), Accept user input when starting idle turns (#36947), and Keep image resize notices attached during remote compaction.

CODEX HARDENS SKILL SELECTION AND SESSION LOOKUP, NODE SDK CLOSES LINTER GAPS

Codex decouples explicit skill selection into a dedicated crate while tightening TUI session validation, as the Node SDK completes its oxlint cleanup stack across five rule categories.

ai 118 shipped

OpenAI Aug 4

GPT-5.5 LANDS ACROSS NODE AND PYTHON SDKS, CODEX HARDENS MCP CONFORMANCE

OpenAI shipped gpt-5.5 model support and tool metadata across both JavaScript and Python SDKs while Codex locked down a full MCP client conformance regression suite to prevent protocol slippage.

ai 115 shipped

OpenAI Aug 3

CODEX HARDENS LOGIN FLOW WITH ONBOARDING HINTS AND METADATA PROTECTION

Three critical fixes shipped across auth, data persistence, and usage tracking to prevent thread metadata loss and secure OAuth callback handling.

ai 6 shipped

OpenAI Aug 2

CODEX SHIPS AGENT PLUGIN OVERHAUL AND TUI PERFORMANCE FIXES

Codex now handles portable Agent Plugins with schema-declared manifests and cuts terminal redraw overhead by caching screen dimensions instead of querying on every frame.

ai 12 shipped

Elsewhere on the wire

AI Agents Aug 6

LANGCHAIN HARDENS ANTHROPIC TOOL CALLING AND GIT CREDENTIALS AS OPENHANDS SHIPS 1.10.0

LangChain fixed a critical credential leak in evaluation runs while Anthropic tool schemas get proper fallback handling, as three major agent frameworks push releases across the ecosystem.

ai-agents 54 shipped

Local LLMs Aug 6

OLLAMA FIXES KV CACHE CORRUPTION, LLAMA.CPP HARDENS FILE SEARCH, VLLM KILLS MOE LEGACY CODE

Ollama patched a critical cache corruption bug that corrupts lazy snapshots during multi-request prefills, while llama.cpp hardened Windows junction traversal and vLLM removed 14 hardcoded MoE shards across quantization methods.