---
格式版本: 2
标题: "ChatGPT & Codex changelog | ChatGPT Learn"
原文链接: "https://learn.chatgpt.com/docs/changelog"
发布日期: "2026-09-04"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "llm:local:strict_original_body"
发布时间证据: "2026-09-04"
发布时间校准原因: "标题附近最新更新日期，为changelog页面当前版本的发布/更新时间。"
发布时间校准置信度: "1"
发布时间候选数量: 78
发布时间严格候选数量: 9
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-09-06T13:08:56+08:00"
发布时间仲裁状态: "confirmed"
发布时间仲裁尝试次数: 1
发布时间仲裁耗时毫秒: 56056
发现时间: "2026-09-06T13:04:47+08:00"
入库时间: "2026-09-06T05:09:57.134Z"
来源平台: "固定入口"
搜索渠道: "fixed_url"
搜索词: "https://developers.openai.com/blog"
匹配关键词:
  []
相关厂家:
  - "OpenAI"
  - "Microsoft"
相关专家:
  []
内容类型: "网页"
抓取工具: "CDP Render"
清洗工具: "CDP Text + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 35
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "正文主线是OpenAI官方发布的Codex CLI与ChatGPT iOS版本更新，新增模型选择、TUI、插件、MCP及缺陷修复等软件功能；来源为官方更新日志且可追溯至GitHub版本与PR。固定知识库未显示重复事件，但本文没有新增AI机架、互连、供电、液冷、RAS或规模部署事实，也无相关客户、订单或量产信号，不属于机架级AI基础设施，命中弱相关强否决。"
AI质检模型: "gpt-5.6-sol"
AI质检时间: "2026-09-08T16:37:23+08:00"
AI主题相关性: 0
AI来源权威性: 15
AI新颖性: 10
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 10
AI评分提示词版本: "v17-精简生产版"
AI评分提示词SHA256: "48fb9777f386026761b4873eaff30807694fb11e9b352d7c69bf2dfde750cc7d"
AI评分知识库版本: "knowledge_base_v1-20260819+runtime.89"
AI评分知识库SHA256: "d189746fc3aa9b189a3266c4540ac047c6c2d69b60f1544fdbc17f4bf8d040c2"
AI评分知识库检索词: "[\"OpenAI\",\"Microsoft\",\"https://developers.openai.com/blog\",\"NPU\",\"Meta\",\"CLI\",\"github.com/openai/codex/pull/42874\",\"github.com/openai/codex/pull/42878\",\"ID\",\"UI\",\"HTTP\",\"CI\"]"
AI评分知识库命中: "[{\"id\":\"runtime-bf4039a0342d37545e9459a2\",\"title\":\"Most Neoclouds Suck At Security\",\"sourceType\":\"ai_excellent_article\",\"time\":\"2026-08-30\",\"matchedTerms\":[\"OpenAI\",\"Microsoft\",\"Meta\",\"CLI\",\"ID\",\"UI\",\"HTTP\",\"CI\"],\"rank\":-12.940233043486701},{\"id\":\"july-correct-0001\",\"title\":\"全球首颗2nm GPU来了！苏姿丰甩出“最强AI机架”，CPU性能干翻英伟达 - 智东西\",\"sourceType\":\"labeled_article\",\"time\":\"2026-07\",\"matchedTerms\":[\"OpenAI\",\"NPU\",\"Meta\",\"ID\",\"HTTP\",\"CI\"],\"rank\":-12.018417421901468},{\"id\":\"july-correct-0033\",\"title\":\"Microsoft, Alphabet, Meta Pivot from Buy to Build in AI\",\"sourceType\":\"labeled_article\",\"time\":\"2026-07\",\"matchedTerms\":[\"OpenAI\",\"Microsoft\",\"Meta\",\"ID\",\"UI\",\"HTTP\",\"CI\"],\"rank\":-11.47055804714985},{\"id\":\"runtime-569639751a0dbe7ef3ffccc5\",\"title\":\"[2608.17503] Predict Before Replay: Joint FEC and Flight Control for Reliable Scale-Up Links\",\"sourceType\":\"ai_excellent_article\",\"time\":\"2026-08-18\",\"matchedTerms\":[\"ID\",\"UI\",\"HTTP\",\"CI\"],\"rank\":-11.114669757602412},{\"id\":\"july-correct-0018\",\"title\":\"AMD, Cerebras partner on joint Helios rack-scale AI inference platform\",\"sourceType\":\"labeled_article\",\"time\":\"2026-07\",\"matchedTerms\":[\"OpenAI\",\"Microsoft\",\"Meta\",\"ID\",\"HTTP\",\"CI\"],\"rank\":-10.306133075629765}]"
采集批次: "2026年9月6日13点04分37秒"
采集批次ID: "20260906-130437-d43ae737"
去重键: "https://learn.chatgpt.com/docs/changelog"
---

- ### Codex CLI 0.153.4
	```bash
	$ npm install -g @openai/codex@0.153.4
	```
	View details
	## Bug Fixes
	- Fixed Astra’s visibility in the bundled model picker and made it the bundled default when no model is explicitly configured. ([#42874](https://github.com/openai/codex/pull/42874))
	- Updated Astra’s guidance to use asynchronous questions only when the tool is available in the session. ([#42878](https://github.com/openai/codex/pull/42878))
	## Changelog
	Full Changelog: [rust-v0.153.3...rust-v0.153.4](https://github.com/openai/codex/compare/rust-v0.153.3...rust-v0.153.4)
	- [#42874](https://github.com/openai/codex/pull/42874) \[0.153 hotfix\] Show Astra in bundled model picker [@rhan-oai](https://github.com/rhan-oai)
	- [#42878](https://github.com/openai/codex/pull/42878) \[0.153 hotfix\] Qualify Astra async-question guidance by tool availability [@rhan-oai](https://github.com/rhan-oai)
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.153.4)
- ### Codex CLI 0.153.3
	```bash
	$ npm install -g @openai/codex@0.153.3
	```
	View details
	## New Features
	- Added GPT-6-Astra to the Amazon Bedrock model picker for Mantle and Runtime global/US routes. ([#42805](https://github.com/openai/codex/pull/42805))
	## Bug Fixes
	- Corrected GPT-6-Astra’s guidance for asynchronous clarification questions to use the supported tool and recognize that it accepts text only. ([#42809](https://github.com/openai/codex/pull/42809))
	## Changelog
	Full Changelog: [rust-v0.153.2...rust-v0.153.3](https://github.com/openai/codex/compare/rust-v0.153.2...rust-v0.153.3)
	- [#42805](https://github.com/openai/codex/pull/42805) \[0.153 hotfix\] Add GPT-6-Astra to Amazon Bedrock catalogs [@celia-oai](https://github.com/celia-oai)
	- [#42809](https://github.com/openai/codex/pull/42809) \[0.153 hotfix\] Update Astra async-question guidance [@rhan-oai](https://github.com/rhan-oai)
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.153.3)
- ### Codex CLI 0.153.2
	```bash
	$ npm install -g @openai/codex@0.153.2
	```
	View details
	## Bug Fixes
	- Corrected the GPT-6-Astra Fast tier description to say “2x speed, increased usage” instead of “1.5x.” This changes only the displayed text, not how requests run. ([#42632](https://github.com/openai/codex/pull/42632))
	## Changelog
	Full Changelog: [rust-v0.153.1...rust-v0.153.2](https://github.com/openai/codex/compare/rust-v0.153.1...rust-v0.153.2)
	- [#42632](https://github.com/openai/codex/pull/42632) Fix GPT-6-Astra Fast tier description for 0.153.2 [@anp-oai](https://github.com/anp-oai)
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.153.2)
- ### Codex CLI 0.153.1
	```bash
	$ npm install -g @openai/codex@0.153.1
	```
	View details
	## New Features
	- Added support for configuring GPT-6-Astra through the API without changing the default model or showing it in the model picker. ([#42605](https://github.com/openai/codex/pull/42605))
	## Changelog
	Full Changelog: [rust-v0.153.0...rust-v0.153.1](https://github.com/openai/codex/compare/rust-v0.153.0...rust-v0.153.1)
	- [#42605](https://github.com/openai/codex/pull/42605) Backport GPT-6-Astra model catalog to 0.153 [@anp-oai](https://github.com/anp-oai)
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.153.1)
- ### Codex CLI 0.153.0
	```bash
	$ npm install -g @openai/codex@0.153.0
	```
	View details
	## New Features
	- Vim mode now supports undo with `u` and redo with `Ctrl+R`, preserving complete drafts including pasted content and attachments. ([#41941](https://github.com/openai/codex/pull/41941), [#42140](https://github.com/openai/codex/pull/42140))
	- The plugin CLI can list, install, and remove plugins from remote marketplaces. ([#42150](https://github.com/openai/codex/pull/42150))
	- Set `tui.auto_recap = false` to disable automatic recaps while keeping manual `/recap` available. ([#42101](https://github.com/openai/codex/pull/42101))
	- TUI history shows complete patches, input sent to background terminals, and individual completed commands. ([#41893](https://github.com/openai/codex/pull/41893), [#42107](https://github.com/openai/codex/pull/42107))
	- Plus and Team users receive an earlier warning when less than half of their allowance remains in an approximately five-hour usage window. ([#42142](https://github.com/openai/codex/pull/42142))
	## Bug Fixes
	- TUI sessions reconnect after an external app-server connection drops, preserving drafts and transcripts while keeping uncertain or queued submissions paused for review. ([#41911](https://github.com/openai/codex/pull/41911), [#41916](https://github.com/openai/codex/pull/41916), [#41918](https://github.com/openai/codex/pull/41918))
	- Full Access skips Guardian reviews for confirmation-only actions. User approval mode skips background Guardian scoring and prewarming, while sensitive-action checks and requests for user input retain their existing handling. ([#42147](https://github.com/openai/codex/pull/42147), [#42256](https://github.com/openai/codex/pull/42256))
	- Guardian review history survives compaction, restarts, and user-created forks while respecting rollback boundaries and isolating subagent history. ([#41879](https://github.com/openai/codex/pull/41879), [#42065](https://github.com/openai/codex/pull/42065))
	- Remembered MCP tool approvals are scoped to the selected app account, and relative MCP executable paths start more reliably on macOS. ([#42133](https://github.com/openai/codex/pull/42133), [#42117](https://github.com/openai/codex/pull/42117))
	- Rollout compression includes shared histories, `codex exec resume` handles compressed rollouts when selecting by working directory, and thread forks work with symlinked session roots. ([#42039](https://github.com/openai/codex/pull/42039), [#42135](https://github.com/openai/codex/pull/42135))
	## Configuration and API Updates
	- App-server thread metadata includes nullable `model` and `reasoningEffort` fields. Structured asynchronous questions are supported through `request_user_input_async` when enabled by the model catalog. ([#42151](https://github.com/openai/codex/pull/42151), [#42178](https://github.com/openai/codex/pull/42178))
	- `tui.disable_paste_burst` replaces the top-level setting, which remains supported as a fallback. ([#41976](https://github.com/openai/codex/pull/41976))
	- Adds the disabled-by-default `features.context_management.experimental_mode` configuration. When enabled for eligible ChatGPT Plus, Pro, or Pro Lite sessions using the Codex backend, it activates token-budget context, history notes, and the `new_context` tool. API-key sessions, custom providers, and temporary structured threads remain excluded. ([#42385](https://github.com/openai/codex/pull/42385))
	## Changelog
	[Full Changelog](https://github.com/openai/codex/compare/rust-v0.152.1...rust-v0.153.0)
	- [#41870](https://github.com/openai/codex/pull/41870) Use shared transcript collection for Guardian reviews
	- [#41879](https://github.com/openai/codex/pull/41879) Preserve Guardian review evidence across compaction
	- [#41884](https://github.com/openai/codex/pull/41884) Add pinned native voice source preparation
	- [#41890](https://github.com/openai/codex/pull/41890) Add native voice dependency build recipe
	- [#41892](https://github.com/openai/codex/pull/41892) Retain the MCP client for event streams
	- [#41893](https://github.com/openai/codex/pull/41893) Show successful TUI commands individually
	- [#41894](https://github.com/openai/codex/pull/41894) Fix Windows native voice dependency builds
	- [#41897](https://github.com/openai/codex/pull/41897) Add the voice helper lifecycle foundation
	- [#41899](https://github.com/openai/codex/pull/41899) Keep MCP event subscriptions alive after task unloading
	- [#41901](https://github.com/openai/codex/pull/41901) Load bounded context after empty wake turns
	- [#41902](https://github.com/openai/codex/pull/41902) Add installed voice host lifecycle support
	- [#41906](https://github.com/openai/codex/pull/41906) Add a manager for MCP event streams
	- [#41908](https://github.com/openai/codex/pull/41908) Avoid scanning archived rollouts when archiving threads
	- [#41909](https://github.com/openai/codex/pull/41909) Make permission transforms aware of executor path context
	- [#41911](https://github.com/openai/codex/pull/41911) Preserve TUI drafts after app-server disconnects
	- [#41912](https://github.com/openai/codex/pull/41912) Persist response token usage in rollout history
	- [#41913](https://github.com/openai/codex/pull/41913) Preserve TUI status timing when the status row is hidden
	- [#41915](https://github.com/openai/codex/pull/41915) Move the config schema generator into a dedicated crate
	- [#41916](https://github.com/openai/codex/pull/41916) Reconnect TUI app-server sessions automatically
	- [#41917](https://github.com/openai/codex/pull/41917) Open the agents overview from an empty composer
	- [#41918](https://github.com/openai/codex/pull/41918) Restore agent navigation after TUI reconnects
	- [#41923](https://github.com/openai/codex/pull/41923) Allow per-call sideband endpoints for existing realtime calls
	- [#41924](https://github.com/openai/codex/pull/41924) Record realtime conversation history in Core
	- [#41925](https://github.com/openai/codex/pull/41925) Test repository-wide Rust formatter discovery
	- [#41928](https://github.com/openai/codex/pull/41928) Use executor path context for permission preapproval
	- [#41929](https://github.com/openai/codex/pull/41929) Open the agents overview directly in the reconnect test
	- [#41933](https://github.com/openai/codex/pull/41933) Report configured sandbox policy consistently
	- [#41934](https://github.com/openai/codex/pull/41934) Omit undersized WAV output from Code Mode
	- [#41936](https://github.com/openai/codex/pull/41936) Attach failed Guardian reviews to diagnostic reports
	- [#41937](https://github.com/openai/codex/pull/41937) Limit background terminal input previews
	- [#41938](https://github.com/openai/codex/pull/41938) Clarify resume guidance in exit summaries
	- [#41940](https://github.com/openai/codex/pull/41940) Preserve transcript layout caches during backtrack selection
	- [#41941](https://github.com/openai/codex/pull/41941) Add Vim undo to the TUI composer
	- [#41944](https://github.com/openai/codex/pull/41944) Emit turn cost telemetry for ChatGPT sessions
	- [#41946](https://github.com/openai/codex/pull/41946) Expand extension permission regression coverage
	- [#41949](https://github.com/openai/codex/pull/41949) Add plugin reconciliation app-server API
	- [#41950](https://github.com/openai/codex/pull/41950) Improve tracing for nested tool calls and exec processes
	- [#41953](https://github.com/openai/codex/pull/41953) Enforce marketplace source policy for curated plugins
	- [#41974](https://github.com/openai/codex/pull/41974) Track TUI starts by app server mode
	- [#41976](https://github.com/openai/codex/pull/41976) Move `disable_paste_burst` under `[tui]`
	- [#41980](https://github.com/openai/codex/pull/41980) Preserve raw response usage metadata
	- [#42003](https://github.com/openai/codex/pull/42003) Report turn trigger and source in turn analytics
	- [#42031](https://github.com/openai/codex/pull/42031) Share Guardian user-message retention logic
	- [#42033](https://github.com/openai/codex/pull/42033) Improve Guardian report diagnostics
	- [#42039](https://github.com/openai/codex/pull/42039) Include shared histories in rollout compression
	- [#42043](https://github.com/openai/codex/pull/42043) Tag Codex home size metrics with compression state
	- [#42047](https://github.com/openai/codex/pull/42047) Add per-account approval settings for apps
	- [#42054](https://github.com/openai/codex/pull/42054) Honor explicit account selectors for Apps tool calls
	- [#42056](https://github.com/openai/codex/pull/42056) Honor app link settings for MCP tool approvals
	- [#42065](https://github.com/openai/codex/pull/42065) Preserve Guardian history across thread reconstruction
	- [#42066](https://github.com/openai/codex/pull/42066) Remove selected core test cases
	- [#42068](https://github.com/openai/codex/pull/42068) Detect standalone installs from the macOS CLI bundle
	- [#42069](https://github.com/openai/codex/pull/42069) Remove redundant test coverage
	- [#42071](https://github.com/openai/codex/pull/42071) Detect Vite+-managed Codex installs
	- [#42076](https://github.com/openai/codex/pull/42076) Unify Guardian context section collection
	- [#42082](https://github.com/openai/codex/pull/42082) Attribute nested REPL reviews to their tool calls
	- [#42085](https://github.com/openai/codex/pull/42085) Centralize Guardian context composition
	- [#42086](https://github.com/openai/codex/pull/42086) Attribute Guardian reviews to OpenAI app tools
	- [#42094](https://github.com/openai/codex/pull/42094) Record Windows MXC availability
	- [#42096](https://github.com/openai/codex/pull/42096) Make diagnostic report uploads resilient to slow networks
	- [#42100](https://github.com/openai/codex/pull/42100) Prefer remote Sites over the bundled plugin
	- [#42101](https://github.com/openai/codex/pull/42101) Add a TUI setting to disable automatic recaps
	- [#42102](https://github.com/openai/codex/pull/42102) Extract OTEL trace WebSocket into a reusable crate
	- [#42104](https://github.com/openai/codex/pull/42104) Show recent sessions in the agent command center
	- [#42107](https://github.com/openai/codex/pull/42107) Show full patches and terminal input in TUI history
	- [#42109](https://github.com/openai/codex/pull/42109) Format Python scripts across the repository
	- [#42110](https://github.com/openai/codex/pull/42110) Treat bundled cleanup hooks as built-ins
	- [#42113](https://github.com/openai/codex/pull/42113) Apply remote platform semantics to exec safety checks
	- [#42114](https://github.com/openai/codex/pull/42114) Centralize remote plugin mutations in `PluginsManager`
	- [#42117](https://github.com/openai/codex/pull/42117) Fix relative MCP server spawning on macOS
	- [#42118](https://github.com/openai/codex/pull/42118) Refine hook activity rendering in the TUI
	- [#42121](https://github.com/openai/codex/pull/42121) Allow updating the approval reviewer for active turns
	- [#42123](https://github.com/openai/codex/pull/42123) Preserve descriptive labels on local file links
	- [#42126](https://github.com/openai/codex/pull/42126) Ignore non-JSON files in plugin catalog test helpers
	- [#42128](https://github.com/openai/codex/pull/42128) Prepare MCP connections for coordinated OAuth refresh
	- [#42132](https://github.com/openai/codex/pull/42132) Bound Git root discovery for metadata enrichment
	- [#42133](https://github.com/openai/codex/pull/42133) Scope session MCP approvals to app account links
	- [#42134](https://github.com/openai/codex/pull/42134) Include app link metadata in MCP approval elicitations
	- [#42135](https://github.com/openai/codex/pull/42135) Support thread forks from symlinked session roots
	- [#42137](https://github.com/openai/codex/pull/42137) Prewarm shell snapshots for eligible turns
	- [#42140](https://github.com/openai/codex/pull/42140) Add redo support to Vim composer history
	- [#42142](https://github.com/openai/codex/pull/42142) Add early rate-limit warnings for Plus and Team plans
	- [#42144](https://github.com/openai/codex/pull/42144) Add Guardian V2 analytics events
	- [#42146](https://github.com/openai/codex/pull/42146) Resolve permission requests in the executor context
	- [#42147](https://github.com/openai/codex/pull/42147) Skip Guardian reviews in Full Access
	- [#42149](https://github.com/openai/codex/pull/42149) Upgrade Git marketplaces from merged configuration
	- [#42150](https://github.com/openai/codex/pull/42150) Support remote marketplaces in the plugin CLI
	- [#42151](https://github.com/openai/codex/pull/42151) Expose model settings in app-server thread metadata
	- [#42161](https://github.com/openai/codex/pull/42161) Split tool JSON Schema code into focused modules
	- [#42164](https://github.com/openai/codex/pull/42164) Record result sources in app tool analytics
	- [#42173](https://github.com/openai/codex/pull/42173) Support header injections in network requirements
	- [#42174](https://github.com/openai/codex/pull/42174) Add a cacheable Bazel app-server schema bundle
	- [#42178](https://github.com/openai/codex/pull/42178) Add structured asynchronous user input requests
	- [#42256](https://github.com/openai/codex/pull/42256) Skip Guardian scoring in User approval mode
	- [#42385](https://github.com/openai/codex/pull/42385) Add experimental context management activation
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.153.0)
- ### ChatGPT for iOS 1.2026.237
	### New features
	- Attachments now work across all connected hosts, including Windows and Linux, and support videos from the Photo Library.
	- Press and hold the attachment button to attach recent photos.
	- A new Priority view brings running tasks, unread updates, and tasks awaiting your response to the top of the task list.
	- Queued prompts now sync with the connected host, remain editable, and send even when the app is in the background.
	- Long-running tasks now show their live working time.
	- Task menus now include an option to copy the thread ID.
	### Improvements and bug fixes
	- Task list loading and organization are faster and more reliable, with simpler date sections and fewer stalls or disappearing projects.
	- Reconnects are more reliable, resolving stuck Send states, missing approvals, and stale task updates.
	- Long responses now stream with fewer visual interruptions.
	- Side chat messages remain available until you close them, even after the chat can no longer reconnect.
- ### Codex CLI 0.152.0
	```bash
	$ npm install -g @openai/codex@0.152.0
	```
	View details
	## New Features
	- Vim mode supports `/` and `?` searches within drafts, highlighted matches, and repeat navigation with `n` and `N`. ([#41586](https://github.com/openai/codex/pull/41586))
	- Rate-limit banners offer actions for checking usage, managing credits, resetting limits, and managing plans. ([#41742](https://github.com/openai/codex/pull/41742))
	- The terminal UI and `codex exec` show credential-refresh progress, including Amazon Bedrock reauthentication. ([#41239](https://github.com/openai/codex/pull/41239))
	- MCP server names can contain `:`, `@`, `/`, and `.`, supporting package-style names throughout CLI commands and authentication. ([#41700](https://github.com/openai/codex/pull/41700))
	- Individual MCP tools support an `output_token_limit` setting, with consistent truncation across session resumes. ([#41421](https://github.com/openai/codex/pull/41421))
	- App-server clients can configure `thread/shellCommand` timeouts, including deadlines longer than one hour. ([#41384](https://github.com/openai/codex/pull/41384))
	## Bug Fixes
	- Vim-enabled composers now start fresh drafts in Insert mode, including after submitting messages or dispatching slash commands. ([#41921](https://github.com/openai/codex/pull/41921))
	- Automatic approval reviews can retain longer messages and a larger conversation transcript. ([#41931](https://github.com/openai/codex/pull/41931))
	- Automatic approval reviews preserve user instructions, answers, and valid authorizations across history compaction. ([#41660](https://github.com/openai/codex/pull/41660), [#41846](https://github.com/openai/codex/pull/41846), [#41852](https://github.com/openai/codex/pull/41852))
	- Resumed threads restore their saved working directory when none is supplied, and client metadata updates preserve filesystem permissions. ([#41567](https://github.com/openai/codex/pull/41567), [#41464](https://github.com/openai/codex/pull/41464))
	- MCP tools remain available through cache refreshes and remote plugin changes; authentication retries use refreshed helper-provided headers. ([#41336](https://github.com/openai/codex/pull/41336), [#41344](https://github.com/openai/codex/pull/41344), [#41396](https://github.com/openai/codex/pull/41396), [#41400](https://github.com/openai/codex/pull/41400))
	- Opening the model picker refreshes available models without losing the highlighted choice. ([#41467](https://github.com/openai/codex/pull/41467))
	- Fixed Windows sandbox execution with Microsoft Store PowerShell, subprocess hangs on terminal queries, and cursor-related display corruption in older JediTerm terminals. ([#41227](https://github.com/openai/codex/pull/41227), [#41436](https://github.com/openai/codex/pull/41436), [#41673](https://github.com/openai/codex/pull/41673))
	- Cloud task requests reject untrusted backend URLs and disable redirects to protect saved credentials. ([#41403](https://github.com/openai/codex/pull/41403))
	## Chores
	- The planning tool is disabled by default; enable it with `tools.update_plan.enabled = true`. ([#41744](https://github.com/openai/codex/pull/41744))
	- Plugin recommendations begin loading during startup, reducing delays before the first turn. ([#41375](https://github.com/openai/codex/pull/41375))
	## Changelog
	Full Changelog: [rust-v0.151.0...rust-v0.152.0](https://github.com/openai/codex/compare/rust-v0.151.0...rust-v0.152.0)
	- [#41210](https://github.com/openai/codex/pull/41210) Enable clock tools from model metadata [@copyberry](https://github.com/copyberry)
	- [#41215](https://github.com/openai/codex/pull/41215) Roll over Guardian context before follow-up reviews [@copyberry](https://github.com/copyberry)
	- [#41218](https://github.com/openai/codex/pull/41218) Share linked tool mention parsing in the TUI [@copyberry](https://github.com/copyberry)
	- [#41219](https://github.com/openai/codex/pull/41219) Retry confirmed remote registration conflicts [@copyberry](https://github.com/copyberry)
	- [#41221](https://github.com/openai/codex/pull/41221) Honor turn token budgets in Guardian review rollover [@copyberry](https://github.com/copyberry)
	- [#41223](https://github.com/openai/codex/pull/41223) Add recency sorting to `project/list` [@copyberry](https://github.com/copyberry)
	- [#41226](https://github.com/openai/codex/pull/41226) Move Guardian review session tests to a separate file [@copyberry](https://github.com/copyberry)
	- [#41227](https://github.com/openai/codex/pull/41227) Use compatible PowerShell for elevated Windows sandbox commands [@copyberry](https://github.com/copyberry)
	- [#41230](https://github.com/openai/codex/pull/41230) Apply app routing policy to unauthenticated plugin reads [@copyberry](https://github.com/copyberry)
	- [#41231](https://github.com/openai/codex/pull/41231) Instrument the loaded plugin cache [@copyberry](https://github.com/copyberry)
	- [#41232](https://github.com/openai/codex/pull/41232) Expose the PowerShell version in environment context [@copyberry](https://github.com/copyberry)
	- [#41235](https://github.com/openai/codex/pull/41235) Sanitize history notes backend errors [@copyberry](https://github.com/copyberry)
	- [#41239](https://github.com/openai/codex/pull/41239) Surface model provider authentication recovery progress [@copyberry](https://github.com/copyberry)
	- [#41243](https://github.com/openai/codex/pull/41243) Add configurable gating for the sleep tool [@copyberry](https://github.com/copyberry)
	- [#41250](https://github.com/openai/codex/pull/41250) Include thread source in realtime connection metadata [@copyberry](https://github.com/copyberry)
	- [#41260](https://github.com/openai/codex/pull/41260) Let the history backend enforce tool output budgets [@copyberry](https://github.com/copyberry)
	- [#41285](https://github.com/openai/codex/pull/41285) Drive keymap conflict checks from the action registry [@copyberry](https://github.com/copyberry)
	- [#41292](https://github.com/openai/codex/pull/41292) Forward history note images to the model [@copyberry](https://github.com/copyberry)
	- [#41308](https://github.com/openai/codex/pull/41308) Make subagents follow the root service tier [@copyberry](https://github.com/copyberry)
	- [#41309](https://github.com/openai/codex/pull/41309) Honor required reviews when reusing Guardian scores [@copyberry](https://github.com/copyberry)
	- [#41313](https://github.com/openai/codex/pull/41313) Decouple HTTP retry backoff from overload integration testing [@copyberry](https://github.com/copyberry)
	- [#41322](https://github.com/openai/codex/pull/41322) Isolate required-model Guardian approval coverage [@copyberry](https://github.com/copyberry)
	- [#41328](https://github.com/openai/codex/pull/41328) Review terminal input against retained permissions [@copyberry](https://github.com/copyberry)
	- [#41331](https://github.com/openai/codex/pull/41331) Classify clock tools as built-in control tools [@copyberry](https://github.com/copyberry)
	- [#41336](https://github.com/openai/codex/pull/41336) Preserve cached MCP tools during binding capture [@copyberry](https://github.com/copyberry)
	- [#41344](https://github.com/openai/codex/pull/41344) Use refreshed MCP tool caches during binding capture [@copyberry](https://github.com/copyberry)
	- [#41349](https://github.com/openai/codex/pull/41349) Assign stable IDs to generated Responses input items [@copyberry](https://github.com/copyberry)
	- [#41354](https://github.com/openai/codex/pull/41354) Reject NUL bytes in reviewed terminal input [@copyberry](https://github.com/copyberry)
	- [#41357](https://github.com/openai/codex/pull/41357) Support compression for shared rollout lineages [@copyberry](https://github.com/copyberry)
	- [#41360](https://github.com/openai/codex/pull/41360) Measure Codex home usage at app-server startup [@copyberry](https://github.com/copyberry)
	- [#41364](https://github.com/openai/codex/pull/41364) Test resuming compressed shared rollouts [@copyberry](https://github.com/copyberry)
	- [#41365](https://github.com/openai/codex/pull/41365) Restrict async user messages to questions [@copyberry](https://github.com/copyberry)
	- [#41368](https://github.com/openai/codex/pull/41368) Match Windows shell guidance to the executor platform [@copyberry](https://github.com/copyberry)
	- [#41375](https://github.com/openai/codex/pull/41375) Preload plugin recommendations during session startup [@copyberry](https://github.com/copyberry)
	- [#41380](https://github.com/openai/codex/pull/41380) Clarify proactive multi-agent delegation guidance [@copyberry](https://github.com/copyberry)
	- [#41384](https://github.com/openai/codex/pull/41384) Support configurable timeouts for thread shell commands [@copyberry](https://github.com/copyberry)
	- [#41385](https://github.com/openai/codex/pull/41385) Give Guardian classifications distinct turn identities [@copyberry](https://github.com/copyberry)
	- [#41392](https://github.com/openai/codex/pull/41392) Add shared Guardian context primitives [@copyberry](https://github.com/copyberry)
	- [#41393](https://github.com/openai/codex/pull/41393) Preserve one-shot exec when unified exec is disabled [@copyberry](https://github.com/copyberry)
	- [#41396](https://github.com/openai/codex/pull/41396) Refresh runtimes for remote plugin state changes [@copyberry](https://github.com/copyberry)
	- [#41400](https://github.com/openai/codex/pull/41400) Refresh MCP HTTP helper headers after authorization failures [@copyberry](https://github.com/copyberry)
	- [#41403](https://github.com/openai/codex/pull/41403) Restrict cloud task credentials to trusted origins [@copyberry](https://github.com/copyberry)
	- [#41413](https://github.com/openai/codex/pull/41413) Optimize history item lookups [@copyberry](https://github.com/copyberry)
	- [#41416](https://github.com/openai/codex/pull/41416) Add app-server notification media filtering [@copyberry](https://github.com/copyberry)
	- [#41421](https://github.com/openai/codex/pull/41421) Support per-tool MCP output limits [@copyberry](https://github.com/copyberry)
	- [#41422](https://github.com/openai/codex/pull/41422) Add shared Guardian transcript collection [@copyberry](https://github.com/copyberry)
	- [#41424](https://github.com/openai/codex/pull/41424) Preserve context baselines across nested agent forks [@copyberry](https://github.com/copyberry)
	- [#41427](https://github.com/openai/codex/pull/41427) Filter media from function call output notifications [@copyberry](https://github.com/copyberry)
	- [#41429](https://github.com/openai/codex/pull/41429) Retain the last selected step context for each turn [@copyberry](https://github.com/copyberry)
	- [#41432](https://github.com/openai/codex/pull/41432) Run executor hooks for interrupted turns [@copyberry](https://github.com/copyberry)
	- [#41435](https://github.com/openai/codex/pull/41435) Allow bundled browser cleanup hooks on subagent stop [@copyberry](https://github.com/copyberry)
	- [#41436](https://github.com/openai/codex/pull/41436) Respond to terminal queries from TTY subprocesses [@copyberry](https://github.com/copyberry)
	- [#41447](https://github.com/openai/codex/pull/41447) Support `openai/elicitation` form requests [@copyberry](https://github.com/copyberry)
	- [#41448](https://github.com/openai/codex/pull/41448) Clarify question handling in Default collaboration mode [@copyberry](https://github.com/copyberry)
	- [#41449](https://github.com/openai/codex/pull/41449) Rename the read-only Seatbelt platform defaults policy [@copyberry](https://github.com/copyberry)
	- [#41452](https://github.com/openai/codex/pull/41452) Report code mode host request durations [@copyberry](https://github.com/copyberry)
	- [#41454](https://github.com/openai/codex/pull/41454) Block goals after repeated execution host failures [@copyberry](https://github.com/copyberry)
	- [#41456](https://github.com/openai/codex/pull/41456) Support app targets in executor plugin hooks [@copyberry](https://github.com/copyberry)
	- [#41457](https://github.com/openai/codex/pull/41457) Source proactive multi-agent instructions from the model catalog [@copyberry](https://github.com/copyberry)
	- [#41461](https://github.com/openai/codex/pull/41461) Source async user message descriptions from the model catalog [@copyberry](https://github.com/copyberry)
	- [#41464](https://github.com/openai/codex/pull/41464) Preserve permissions when updating session metadata [@copyberry](https://github.com/copyberry)
	- [#41467](https://github.com/openai/codex/pull/41467) Refresh the TUI model picker from the app server [@copyberry](https://github.com/copyberry)
	- [#41476](https://github.com/openai/codex/pull/41476) Use rules\_rs platforms for release binaries [@copyberry](https://github.com/copyberry)
	- [#41477](https://github.com/openai/codex/pull/41477) Organize bundled Rust resources under asset directories [@copyberry](https://github.com/copyberry)
	- [#41562](https://github.com/openai/codex/pull/41562) Preserve turn lineage across goal continuations [@copyberry](https://github.com/copyberry)
	- [#41567](https://github.com/openai/codex/pull/41567) Restore thread cwd from owned settings snapshots [@copyberry](https://github.com/copyberry)
	- [#41569](https://github.com/openai/codex/pull/41569) Harden diagnostic report uploads [@copyberry](https://github.com/copyberry)
	- [#41570](https://github.com/openai/codex/pull/41570) Fix proactive multi-agent instruction grammar [@copyberry](https://github.com/copyberry)
	- [#41586](https://github.com/openai/codex/pull/41586) Add Vim search motions to the composer [@copyberry](https://github.com/copyberry)
	- [#41613](https://github.com/openai/codex/pull/41613) Move Vim history tests into the history search module [@copyberry](https://github.com/copyberry)
	- [#41630](https://github.com/openai/codex/pull/41630) Update tests for default-enabled update\_plan [@copyberry](https://github.com/copyberry)
	- [#41660](https://github.com/openai/codex/pull/41660) Preserve Guardian authorization across history compaction [@copyberry](https://github.com/copyberry)
	- [#41666](https://github.com/openai/codex/pull/41666) Approve the first Node REPL execution without a Guardian wait [@copyberry](https://github.com/copyberry)
	- [#41673](https://github.com/openai/codex/pull/41673) Repair cursor-style rendering on older JediTerm terminals [@copyberry](https://github.com/copyberry)
	- [#41683](https://github.com/openai/codex/pull/41683) Set working directories for environment MCP tests [@copyberry](https://github.com/copyberry)
	- [#41700](https://github.com/openai/codex/pull/41700) Support package-style MCP server names [@copyberry](https://github.com/copyberry)
	- [#41742](https://github.com/openai/codex/pull/41742) Show actionable rate-limit banners in the TUI [@copyberry](https://github.com/copyberry)
	- [#41743](https://github.com/openai/codex/pull/41743) Mark history ingestion requests in turn metadata [@copyberry](https://github.com/copyberry)
	- [#41744](https://github.com/openai/codex/pull/41744) Make the update\_plan tool opt-in [@copyberry](https://github.com/copyberry)
	- [#41803](https://github.com/openai/codex/pull/41803) Allow models to enable token budgeting by default [@copyberry](https://github.com/copyberry)
	- [#41840](https://github.com/openai/codex/pull/41840) Use the async stack budget for approval reviews [@copyberry](https://github.com/copyberry)
	- [#41846](https://github.com/openai/codex/pull/41846) Preserve Guardian review evidence across compaction [@copyberry](https://github.com/copyberry)
	- [#41852](https://github.com/openai/codex/pull/41852) Preserve Guardian user answers across compaction [@copyberry](https://github.com/copyberry)
	- [#41853](https://github.com/openai/codex/pull/41853) Box the session startup future at its API boundary [@copyberry](https://github.com/copyberry)
	- [#41857](https://github.com/openai/codex/pull/41857) Preserve Guardian user answers from current history [@copyberry](https://github.com/copyberry)
	- [#41858](https://github.com/openai/codex/pull/41858) Preserve user text when Guardian history drops oversized images [@copyberry](https://github.com/copyberry)
	- [#41861](https://github.com/openai/codex/pull/41861) Keep history extension tools out of Guardian reviews [@copyberry](https://github.com/copyberry)
	- [#41921](https://github.com/openai/codex/pull/41921) Start fresh Vim drafts in Insert mode
	- [#41931](https://github.com/openai/codex/pull/41931) Increase Guardian message transcript limits
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.152.0)

- ### Codex CLI 0.151.0
	```bash
	$ npm install -g @openai/codex@0.151.0
	```
	View details
	## New Features
	- Added a configurable grace period for discovering tools from optional MCP servers. ([#41199](https://github.com/openai/codex/pull/41199))
	- Extensions can now inspect or replace MCP tool results before they reach the model. ([#41202](https://github.com/openai/codex/pull/41202))
	- Plugin catalogs now combine per-repository configuration and report invalid project marketplaces without hiding valid plugins. ([#41208](https://github.com/openai/codex/pull/41208))
	## Bug Fixes
	- Preserved restored permission profiles across TUI turns and prevented `/cd` from weakening sandbox restrictions. ([#41192](https://github.com/openai/codex/pull/41192))
	- Kept tool availability and reasoning effort correct when switching models or falling back to another model. ([#41195](https://github.com/openai/codex/pull/41195), [#41206](https://github.com/openai/codex/pull/41206))
	- Improved remote sandbox enforcement using the executor’s actual home directory, operating system, and path conventions. ([#41196](https://github.com/openai/codex/pull/41196), [#41204](https://github.com/openai/codex/pull/41204), [#41207](https://github.com/openai/codex/pull/41207), [#41209](https://github.com/openai/codex/pull/41209))
	- Preserved structured MCP tool and resource errors in app-server responses. ([#41196](https://github.com/openai/codex/pull/41196))
	- Counted nested subagent token usage toward root goal budgets. ([#41183](https://github.com/openai/codex/pull/41183))
	- Prevented stale Guardian classifications from authorizing actions after permission state changes. ([#41196](https://github.com/openai/codex/pull/41196))
	## Chores
	- Added telemetry for escalated stdin reviews and remote executor MCP discovery. ([#41189](https://github.com/openai/codex/pull/41189), [#41205](https://github.com/openai/codex/pull/41205))
	- Stabilized Guardian WebSocket and core fixture tests under slow or highly concurrent CI. ([#41191](https://github.com/openai/codex/pull/41191), [#41194](https://github.com/openai/codex/pull/41194))
	## Changelog
	Full Changelog: [rust-v0.150.0...rust-v0.151.0](https://github.com/openai/codex/compare/rust-v0.150.0...rust-v0.151.0)
	- [#41183](https://github.com/openai/codex/pull/41183) Account subagent token usage toward root goals [@copyberry](https://github.com/copyberry)
	- [#41189](https://github.com/openai/codex/pull/41189) Instrument stdin review size checks [@copyberry](https://github.com/copyberry)
	- [#41191](https://github.com/openai/codex/pull/41191) Stabilize Guardian WebSocket tests [@copyberry](https://github.com/copyberry)
	- [#41192](https://github.com/openai/codex/pull/41192) Preserve restored permission profiles in TUI sessions [@copyberry](https://github.com/copyberry)
	- [#41193](https://github.com/openai/codex/pull/41193) Report affected capabilities from remote plugin syncs [@copyberry](https://github.com/copyberry)
	- [#41194](https://github.com/openai/codex/pull/41194) Harden core test fixture startup assertions [@copyberry](https://github.com/copyberry)
	- [#41195](https://github.com/openai/codex/pull/41195) Finalize model-specific tool plans in `ToolRouter` [@copyberry](https://github.com/copyberry)
	- [#41196](https://github.com/openai/codex/pull/41196) Improve sandboxing, MCP errors, and cached approvals [@copyberry](https://github.com/copyberry)
	- [#41199](https://github.com/openai/codex/pull/41199) Make the optional MCP startup grace configurable [@copyberry](https://github.com/copyberry)
	- [#41202](https://github.com/openai/codex/pull/41202) Let extensions process MCP tool results [@copyberry](https://github.com/copyberry)
	- [#41204](https://github.com/openai/codex/pull/41204) Propagate executor home directories into sandbox contexts [@copyberry](https://github.com/copyberry)
	- [#41205](https://github.com/openai/codex/pull/41205) Track executor MCP discovery telemetry [@copyberry](https://github.com/copyberry)
	- [#41206](https://github.com/openai/codex/pull/41206) Make Ultra reasoning fallback model-aware [@copyberry](https://github.com/copyberry)
	- [#41207](https://github.com/openai/codex/pull/41207) Propagate executor OS into turn environments [@copyberry](https://github.com/copyberry)
	- [#41208](https://github.com/openai/codex/pull/41208) Honor per-repository plugin configuration in catalog requests [@copyberry](https://github.com/copyberry)
	- [#41209](https://github.com/openai/codex/pull/41209) Align deny-read matching with executor path semantics [@copyberry](https://github.com/copyberry)
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.151.0)
- ### Codex CLI 0.150.1
	```bash
	$ npm install -g @openai/codex@0.150.1
	```
	View details
	## Bug Fixes
	- Remote compaction now counts retained images toward its token budget by default, trimming older images as needed. ([#41003](https://github.com/openai/codex/pull/41003))
	## Changelog
	Full Changelog: [rust-v0.150.0...rust-v0.150.1](https://github.com/openai/codex/compare/rust-v0.150.0...rust-v0.150.1)
	- [#41003](https://github.com/openai/codex/pull/41003) Backport retained-image compaction budgeting to 0.150 [@rhan-oai](https://github.com/rhan-oai)
	[Full release on Github](https://github.com/openai/codex/releases/tag/rust-v0.150.1)
- ### ChatGPT for iOS 1.2026.230
	### New features
	- Added task search across titles and conversation content on connected hosts.
	- Added a compact composer gauge for viewing and adjusting reasoning effort.
	- Added a full-screen editor for longer prompts.
	- Added configurable Home Screen shortcuts for ChatGPT, Work, and Codex Remote.
	- Added optional comments to selected response annotations.
	### Improvements and bug fixes
	- Improved loading for long threads, with older history fetched as needed.
	- Refined task list layout, ordering, and pinning to better match desktop.
	- Inline visualizations now follow iOS appearance and accent settings.
- ### Codex CLI 0.150.0
	```bash
	$ npm install -g @openai/codex@0.150.0
	```
	View details
	## New Features
	- Reference other Codex tasks with `@` mentions, and ask agents to read, create, or message tasks from the terminal. ([#40308](https://github.com/openai/codex/pull/40308), [#40315](https://github.com/openai/codex/pull/40315))
	- `/copy` now offers a picker for full responses, individual code blocks, and blockquotes. ([#39997](https://github.com/openai/codex/pull/39997))
	- Unnamed terminal tasks receive descriptive titles automatically, and `/rename` suggests an editable title based on the conversation. ([#40492](https://github.com/openai/codex/pull/40492), [#40495](https://github.com/openai/codex/pull/40495))
	- Markdown links appear as clickable labels in supported terminals, with visible URLs retained elsewhere. ([#40471](https://github.com/openai/codex/pull/40471))
	- Bind shortcuts to cycle permission modes, and use `.` in Vim mode to repeat your last edit. ([#39873](https://github.com/openai/codex/pull/39873), [#40521](https://github.com/openai/codex/pull/40521))
	- New `Interrupt` hooks can run commands or MCP handlers when an active top-level turn is interrupted. ([#40511](https://github.com/openai/codex/pull/40511))
	## Bug Fixes
	- Untrusted projects no longer supply project-level `AGENTS.md` instructions, and managed deny-read rules remain enforced after permission changes. ([#39837](https://github.com/openai/codex/pull/39837), [#40004](https://github.com/openai/codex/pull/40004))
	- Improved credential redaction in app-server diagnostics, including provider, authentication refresh, and attestation fields. ([#39993](https://github.com/openai/codex/pull/39993))
	- Fixed remote MCP bearer-token lookup and required-server startup while preserving compatibility with older executors. ([#39926](https://github.com/openai/codex/pull/39926), [#39952](https://github.com/openai/codex/pull/39952), [#39979](https://github.com/openai/codex/pull/39979))
	- Fixed elevated Windows sandbox setup and launch aliases under Unicode user paths. ([#39971](https://github.com/openai/codex/pull/39971), [#40570](https://github.com/openai/codex/pull/40570))
	- Prevented Unix shutdown hangs caused by detached processes retaining a terminal or full terminal output buffers. ([#40460](https://github.com/openai/codex/pull/40460))
	- Fixed conversation compaction and multi-agent compatibility for Amazon Bedrock models. ([#39804](https://github.com/openai/codex/pull/39804), [#39825](https://github.com/openai/codex/pull/39825))
	## Changelog
	Full Changelog: [rust-v0.149.0...rust-v0.150.0](https://github.com/openai/codex/compare/rust-v0.149.0...rust-v0.150.0)
	- [#39667](https://github.com/openai/codex/pull/39667) Trace MCP runtime refresh coordination [@copyberry](https://github.com/copyberry)
	- [#39691](https://github.com/openai/codex/pull/39691) Make Guardian v2 parent compaction reuse configurable [@copyberry](https://github.com/copyberry)
	- [#39700](https://github.com/openai/codex/pull/39700) Fail closed on unsafe config and sed parsing [@copyberry](https://github.com/copyberry)
	- [#39702](https://github.com/openai/codex/pull/39702) Wait for turn completion events in multi-agent resume tests [@copyberry](https://github.com/copyberry)
	- [#39703](https://github.com/openai/codex/pull/39703) Move the global scope check into the code-mode runtime [@copyberry](https://github.com/copyberry)
	- [#39706](https://github.com/openai/codex/pull/39706) Harden Seatbelt writable root path binding [@copyberry](https://github.com/copyberry)
	- [#39707](https://github.com/openai/codex/pull/39707) Remove redundant code mode image helper test [@copyberry](https://github.com/copyberry)
	- [#39709](https://github.com/openai/codex/pull/39709) Log TUI app event variants without their payloads [@copyberry](https://github.com/copyberry)
	- [#39712](https://github.com/openai/codex/pull/39712) Reduce unified exec output buffer allocations [@copyberry](https://github.com/copyberry)
	- [#39717](https://github.com/openai/codex/pull/39717) Pass CI workflow inputs through environment variables [@copyberry](https://github.com/copyberry)
	- [#39719](https://github.com/openai/codex/pull/39719) Stop persisting checkout credentials in V8 workflows [@copyberry](https://github.com/copyberry)
	- [#39720](https://github.com/openai/codex/pull/39720) Expose managed policy for browser settings imports [@copyberry](https://github.com/copyberry)
	- [#39722](https://github.com/openai/codex/pull/39722) Track multi-agent v2 spawn calls in analytics [@copyberry](https://github.com/copyberry)
	- [#39726](https://github.com/openai/codex/pull/39726) Box the WebSocket dial future [@copyberry](https://github.com/copyberry)
	- [#39731](https://github.com/openai/codex/pull/39731) Avoid rollout reads for configured TUI sessions [@copyberry](https://github.com/copyberry)
	- [#39736](https://github.com/openai/codex/pull/39736) Remove private executor directory creation [@copyberry](https://github.com/copyberry)
	- [#39738](https://github.com/openai/codex/pull/39738) Honor Guardian runtime settings from model defaults [@copyberry](https://github.com/copyberry)
	- [#39741](https://github.com/openai/codex/pull/39741) Use model-specific auto-review outcome instructions [@copyberry](https://github.com/copyberry)
	- [#39744](https://github.com/openai/codex/pull/39744) Skip postprocessing for short composer input [@copyberry](https://github.com/copyberry)
	- [#39746](https://github.com/openai/codex/pull/39746) Refresh resumed thread capability roots from executors [@copyberry](https://github.com/copyberry)
	- [#39749](https://github.com/openai/codex/pull/39749) Require filesystem backends to implement directory walks [@copyberry](https://github.com/copyberry)
	- [#39752](https://github.com/openai/codex/pull/39752) Expose uncompiled permission profile selection [@copyberry](https://github.com/copyberry)
	- [#39755](https://github.com/openai/codex/pull/39755) Add managed developer instructions to requirements [@copyberry](https://github.com/copyberry)
	- [#39756](https://github.com/openai/codex/pull/39756) Cache shell snapshots in the exec server [@copyberry](https://github.com/copyberry)
	- [#39757](https://github.com/openai/codex/pull/39757) Standardize shell execution on unified exec [@copyberry](https://github.com/copyberry)
	- [#39761](https://github.com/openai/codex/pull/39761) Add app-server MCP event streaming [@copyberry](https://github.com/copyberry)
	- [#39765](https://github.com/openai/codex/pull/39765) Include suggestion IDs in plugin install metadata [@copyberry](https://github.com/copyberry)
	- [#39770](https://github.com/openai/codex/pull/39770) Refresh bundled model definitions [@copyberry](https://github.com/copyberry)
	- [#39772](https://github.com/openai/codex/pull/39772) Standardize shell execution on unified exec [@copyberry](https://github.com/copyberry)
	- [#39776](https://github.com/openai/codex/pull/39776) Verify Codex app signatures before launch or install [@copyberry](https://github.com/copyberry)
	- [#39777](https://github.com/openai/codex/pull/39777) Retry transient registry failures during initial exec connection [@copyberry](https://github.com/copyberry)
	- [#39779](https://github.com/openai/codex/pull/39779) Make tool-result telemetry limits configurable [@copyberry](https://github.com/copyberry)
	- [#39782](https://github.com/openai/codex/pull/39782) Support standalone named function call outputs [@copyberry](https://github.com/copyberry)
	- [#39784](https://github.com/openai/codex/pull/39784) Classify rollout migration failures [@copyberry](https://github.com/copyberry)
	- [#39785](https://github.com/openai/codex/pull/39785) Support turn cost telemetry for custom model providers [@copyberry](https://github.com/copyberry)
	- [#39786](https://github.com/openai/codex/pull/39786) Support host-accepted exec-server WebSockets [@copyberry](https://github.com/copyberry)
	- [#39790](https://github.com/openai/codex/pull/39790) Deduplicate zsh fork test setup [@copyberry](https://github.com/copyberry)
	- [#39791](https://github.com/openai/codex/pull/39791) Handle standalone tool outputs as external context [@copyberry](https://github.com/copyberry)
	- [#39792](https://github.com/openai/codex/pull/39792) Reject settings updates for parent-owned subagents [@copyberry](https://github.com/copyberry)
	- [#39794](https://github.com/openai/codex/pull/39794) Install build tools in full Rust CI [@copyberry](https://github.com/copyberry)
	- [#39795](https://github.com/openai/codex/pull/39795) Add hostname to the configurable TUI status line [@copyberry](https://github.com/copyberry)
	- [#39797](https://github.com/openai/codex/pull/39797) Enrich thread archive analytics with thread context [@copyberry](https://github.com/copyberry)
	- [#39798](https://github.com/openai/codex/pull/39798) Update rmcp to 3.1.3 [@copyberry](https://github.com/copyberry)
	- [#39802](https://github.com/openai/codex/pull/39802) Optimize case-insensitive thread history matching [@copyberry](https://github.com/copyberry)
	- [#39804](https://github.com/openai/codex/pull/39804) Use multi-agent V1 for Amazon Bedrock models [@copyberry](https://github.com/copyberry)
	- [#39807](https://github.com/openai/codex/pull/39807) Finalize reserved PDF uploads with creation context [@copyberry](https://github.com/copyberry)
	- [#39809](https://github.com/openai/codex/pull/39809) Preserve WINDIR in core Windows shell environments [@copyberry](https://github.com/copyberry)
	- [#39811](https://github.com/openai/codex/pull/39811) Restrict macOS preference reads to full-disk policies [@copyberry](https://github.com/copyberry)
	- [#39812](https://github.com/openai/codex/pull/39812) Avoid materializing writable-root carveouts for presence checks [@copyberry](https://github.com/copyberry)
	- [#39813](https://github.com/openai/codex/pull/39813) Defer legacy filesystem policy projection [@copyberry](https://github.com/copyberry)
	- [#39822](https://github.com/openai/codex/pull/39822) Preserve uncapped Guardian classifier instructions [@copyberry](https://github.com/copyberry)
	- [#39825](https://github.com/openai/codex/pull/39825) Use Responses compaction for Amazon Bedrock [@copyberry](https://github.com/copyberry)
	- [#39827](https://github.com/openai/codex/pull/39827) Add history and notes tools for token-budget sessions [@copyberry](https://github.com/copyberry)
	- [#39830](https://github.com/openai/codex/pull/39830) Rename the history notes extension config option [@copyberry](https://github.com/copyberry)
	- [#39837](https://github.com/openai/codex/pull/39837) Ignore project instructions for untrusted projects [@copyberry](https://github.com/copyberry)
	- [#39847](https://github.com/openai/codex/pull/39847) Include context window IDs in response metadata [@copyberry](https://github.com/copyberry)
	- [#39852](https://github.com/openai/codex/pull/39852) Reset registry retries when refreshing Noise bundles [@copyberry](https://github.com/copyberry)
	- [#39864](https://github.com/openai/codex/pull/39864) Limit pending input preview wrapping work [@copyberry](https://github.com/copyberry)
	- [#39870](https://github.com/openai/codex/pull/39870) Allow session configuration with `codex agents` [@copyberry](https://github.com/copyberry)
	- [#39873](https://github.com/openai/codex/pull/39873) Add keybindings for cycling TUI permission modes [@copyberry](https://github.com/copyberry)
	- [#39876](https://github.com/openai/codex/pull/39876) Support attaching to existing realtime calls [@copyberry](https://github.com/copyberry)
	- [#39917](https://github.com/openai/codex/pull/39917) Honor request PATH in exec-server shell snapshots [@copyberry](https://github.com/copyberry)
	- [#39926](https://github.com/openai/codex/pull/39926) Resolve HTTP MCP bearer tokens in executor environments [@copyberry](https://github.com/copyberry)
	- [#39930](https://github.com/openai/codex/pull/39930) Track remote MCP header environment variables [@copyberry](https://github.com/copyberry)
	- [#39935](https://github.com/openai/codex/pull/39935) Enforce issuer binding for MCP OAuth endpoints [@copyberry](https://github.com/copyberry)
	- [#39937](https://github.com/openai/codex/pull/39937) Bound unified exec output delta frames [@copyberry](https://github.com/copyberry)
	- [#39940](https://github.com/openai/codex/pull/39940) Allow more time for local code-mode host startup [@copyberry](https://github.com/copyberry)
	- [#39941](https://github.com/openai/codex/pull/39941) Discover HTTP MCP servers from selected executors [@copyberry](https://github.com/copyberry)
	- [#39952](https://github.com/openai/codex/pull/39952) Honor required MCP servers from selected executors [@copyberry](https://github.com/copyberry)
	- [#39953](https://github.com/openai/codex/pull/39953) Support voice-aware configuration and version-skew builds [@copyberry](https://github.com/copyberry)
	- [#39957](https://github.com/openai/codex/pull/39957) Add in-memory shell snapshots to unified exec [@copyberry](https://github.com/copyberry)
	- [#39958](https://github.com/openai/codex/pull/39958) Stop advertising shell snapshots from local exec servers [@copyberry](https://github.com/copyberry)
	- [#39961](https://github.com/openai/codex/pull/39961) Test browser MCP bearer tokens over executor WebSockets [@copyberry](https://github.com/copyberry)
	- [#39962](https://github.com/openai/codex/pull/39962) Keep Guardian reviews isolated from executor MCP servers [@copyberry](https://github.com/copyberry)
	- [#39967](https://github.com/openai/codex/pull/39967) Upgrade pnpm to 10.34.5 [@copyberry](https://github.com/copyberry)
	- [#39969](https://github.com/openai/codex/pull/39969) Consolidate code mode output helper tests [@copyberry](https://github.com/copyberry)
	- [#39971](https://github.com/openai/codex/pull/39971) Fix elevated Windows sandbox setup activation [@copyberry](https://github.com/copyberry)
	- [#39975](https://github.com/openai/codex/pull/39975) Preserve root user authorization in subagent Guardian reviews [@copyberry](https://github.com/copyberry)
	- [#39976](https://github.com/openai/codex/pull/39976) Allow semaphore limit queries in the macOS sandbox [@copyberry](https://github.com/copyberry)
	- [#39979](https://github.com/openai/codex/pull/39979) Preserve MCP compatibility with older executors [@copyberry](https://github.com/copyberry)
	- [#39980](https://github.com/openai/codex/pull/39980) Enforce environment network policies for remote execution [@copyberry](https://github.com/copyberry)
	- [#39981](https://github.com/openai/codex/pull/39981) Bypass risk scoring for models that require automatic review [@copyberry](https://github.com/copyberry)
	- [#39985](https://github.com/openai/codex/pull/39985) Truncate Guardian instructions after rendering the policy [@copyberry](https://github.com/copyberry)
	- [#39991](https://github.com/openai/codex/pull/39991) Preserve TUI event ordering during active-thread draining [@copyberry](https://github.com/copyberry)
	- [#39992](https://github.com/openai/codex/pull/39992) Keep keymap action descriptions stable while navigating [@copyberry](https://github.com/copyberry)
	- [#39993](https://github.com/openai/codex/pull/39993) Keep credentials out of app-server logs [@copyberry](https://github.com/copyberry)
	- [#39994](https://github.com/openai/codex/pull/39994) Add Guardian internal session support [@copyberry](https://github.com/copyberry)
	- [#39995](https://github.com/openai/codex/pull/39995) Expand browser and computer use requirements [@copyberry](https://github.com/copyberry)
	- [#39997](https://github.com/openai/codex/pull/39997) Add a response target picker to `/copy` [@copyberry](https://github.com/copyberry)
	- [#39999](https://github.com/openai/codex/pull/39999) Hide Fast mode status for unsupported models [@copyberry](https://github.com/copyberry)
	- [#40000](https://github.com/openai/codex/pull/40000) Expose browser and computer-use requirements through app-server [@copyberry](https://github.com/copyberry)
	- [#40004](https://github.com/openai/codex/pull/40004) Preserve managed deny-read rules across permission updates [@copyberry](https://github.com/copyberry)