<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: aicoding-guide</title>
    <description>The latest articles on DEV Community by aicoding-guide (@aicoding-guide).</description>
    <link>https://dev.to/aicoding-guide</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4119600%2F36237543-8d98-41c3-b6a3-c8202a09085f.png</url>
      <title>DEV Community: aicoding-guide</title>
      <link>https://dev.to/aicoding-guide</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/aicoding-guide"/>
    <language>en</language>
    <item>
      <title>What changed in Codex CLI v0.168 and v0.169, and when codex exec needs --trust</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Tue, 17 Nov 2026 19:19:07 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/what-changed-in-codex-cli-v0168-and-v0169-and-when-codex-exec-needs-trust-4d9p</link>
      <guid>https://dev.to/aicoding-guide/what-changed-in-codex-cli-v0168-and-v0169-and-when-codex-exec-needs-trust-4d9p</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-169/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Codex CLI v0.168.0 and v0.169.0 carry one change that matters for CI: &lt;strong&gt;&lt;code&gt;codex exec&lt;/code&gt; no longer loads configuration or MCP servers from a project you have not trusted.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;There is a second thing worth knowing before you edit any config. Of the settings keys named in these release notes, &lt;strong&gt;three are absent from the official configuration reference, and one has a documented value list that contradicts the notes.&lt;/strong&gt; It is worth sorting the verified from the unverified first.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What you will learn&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;What the &lt;code&gt;codex exec&lt;/code&gt; trust check disables, and when to pass &lt;code&gt;--trust&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Setting project trust in a file with &lt;code&gt;projects.&amp;lt;path&amp;gt;.trust_level&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Which release-note keys cannot be confirmed against the configuration reference&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  codex exec ignores an untrusted project's configuration
&lt;/h2&gt;

&lt;p&gt;The first v0.169.0 bug fix reads:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;code&gt;codex exec&lt;/code&gt; keeps project configuration and MCP servers disabled in untrusted directories, even with &lt;code&gt;--sandbox workspace-write&lt;/code&gt;. Use &lt;code&gt;--trust&lt;/code&gt; with &lt;code&gt;codex exec&lt;/code&gt; or &lt;code&gt;codex review&lt;/code&gt; to explicitly save project trust.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Widening the sandbox does not widen trust. The configuration reference states the premise: "Codex loads project-scoped config files only when you trust the project."&lt;/p&gt;

&lt;p&gt;What untrusted actually turns off is spelled out in Advanced Configuration:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;If the project is untrusted, Codex ignores project &lt;code&gt;.codex/&lt;/code&gt; layers, including &lt;code&gt;.codex/config.toml&lt;/code&gt;, project-local hooks,&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;and project-local rules. User-level and system-level layers are separate and keep loading. For hooks specifically: "Project-local hooks load only when the project &lt;code&gt;.codex/&lt;/code&gt; layer is trusted." and "User-level hooks remain independent of project trust."&lt;/p&gt;

&lt;p&gt;Here is the flag:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Save this project's trust for this run and for future chats&lt;/span&gt;
codex &lt;span class="nb"&gt;exec&lt;/span&gt; --trust &lt;span class="s2"&gt;"Run the test suite and summarize failures"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;blockquote&gt;
&lt;p&gt;Save project trust for this run and future chats, even with &lt;code&gt;--ephemeral&lt;/code&gt;. Conflicts with &lt;code&gt;--ignore-user-config&lt;/code&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;So trust is saved even when you pass &lt;code&gt;--ephemeral&lt;/code&gt; (which keeps session rollout files off disk), and it cannot be combined with &lt;code&gt;--ignore-user-config&lt;/code&gt;. On the review side, &lt;code&gt;codex review --trust&lt;/code&gt; is documented as "Trust this project for this review and future chats, including its configured MCP servers."&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;The MCP part could not be confirmed in the configuration reference&lt;/strong&gt;&lt;br&gt;
The release notes say MCP servers stay disabled too. But &lt;strong&gt;no statement that an untrusted project's MCP servers do not load could be found in the configuration reference or Advanced Configuration.&lt;/strong&gt; What those pages list as the consequences of being untrusted stops at the project &lt;code&gt;.codex/&lt;/code&gt; layers: project config, project-local hooks, project-local rules. The wording of &lt;code&gt;codex review --trust&lt;/code&gt;, "including its configured MCP servers", is the only text pointing that way.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;You can also declare trust ahead of time.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Key&lt;/th&gt;
&lt;th&gt;Values&lt;/th&gt;
&lt;th&gt;What the docs say&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;projects.&amp;lt;path&amp;gt;.trust_level&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;"trusted"&lt;/code&gt; / &lt;code&gt;"untrusted"&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;"Mark a project or worktree as trusted or untrusted". An untrusted project will "skip project-scoped &lt;code&gt;.codex/&lt;/code&gt; layers, including project-local config, hooks, and rules"&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Do not confuse this with &lt;code&gt;approval_policy = "untrusted"&lt;/code&gt;, which is gone. The reference says "Codex and ChatGPT Work no longer support &lt;code&gt;approval_policy = \"untrusted\"&lt;/code&gt;" while noting that "Project entries with &lt;code&gt;trust_level = \"untrusted\"&lt;/code&gt; in user-level &lt;code&gt;~/.codex/config.toml&lt;/code&gt; remain supported." For that distinction, see &lt;a href="https://aicoding-guide.com/en/posts/codex-project-trust-level/" rel="noopener noreferrer"&gt;Stop Codex asking whether to trust this folder with trust_level in config.toml&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Release-note keys that are missing from the reference
&lt;/h2&gt;

&lt;p&gt;The v0.168.0 notes name three configuration keys. Checked against the configuration reference as of 2026-11-18:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;What the release notes say&lt;/th&gt;
&lt;th&gt;What the reference says&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;features.tui_workspace&lt;/code&gt; (a side conversation beside the main chat)&lt;/td&gt;
&lt;td&gt;
&lt;strong&gt;No entry for the key.&lt;/strong&gt; The name appears once, inside the &lt;code&gt;/side&lt;/code&gt; slash-command description&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;features.image_generation_animation&lt;/code&gt; (animated GIF generation)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;No entry&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;tui.notification_method = "osc777"&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;The key exists, but its values are &lt;strong&gt;&lt;code&gt;auto&lt;/code&gt;, &lt;code&gt;osc9&lt;/code&gt; and &lt;code&gt;bel&lt;/code&gt;&lt;/strong&gt; — there is no &lt;code&gt;osc777&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;incremental_tools_legacy_gcp&lt;/code&gt; (v0.167.2)&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;No entry&lt;/strong&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The documented description of &lt;code&gt;tui.notification_method&lt;/code&gt; is "Notification method for terminal notifications (default: auto)." &lt;code&gt;auto&lt;/code&gt; is explained too: Codex prefers OSC 9 where the terminal looks capable and "falls back to BEL (&lt;code&gt;\x07&lt;/code&gt;) otherwise." Related keys are &lt;code&gt;tui.notification_condition&lt;/code&gt; (&lt;code&gt;unfocused&lt;/code&gt; / &lt;code&gt;always&lt;/code&gt;, default &lt;code&gt;unfocused&lt;/code&gt;) and &lt;code&gt;tui.notifications&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;So if you want to change the notification method, &lt;strong&gt;&lt;code&gt;osc9&lt;/code&gt; and &lt;code&gt;bel&lt;/code&gt; are what you can verify today.&lt;/strong&gt; &lt;code&gt;osc777&lt;/code&gt; exists only in the release notes. For the file as a whole, see &lt;a href="https://aicoding-guide.com/en/posts/codex-config-toml/" rel="noopener noreferrer"&gt;How to write Codex config.toml and its main settings keys&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The features.multi_agent deprecation is not in the docs
&lt;/h2&gt;

&lt;p&gt;One v0.168.0 chore reads:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Codex warns when &lt;code&gt;features.multi_agent&lt;/code&gt; is explicitly configured and points to &lt;code&gt;agents.enabled&lt;/code&gt; for disabling subagent tools.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Both keys exist in the reference. But &lt;strong&gt;nothing marks &lt;code&gt;features.multi_agent&lt;/code&gt; as deprecated.&lt;/strong&gt;&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Key&lt;/th&gt;
&lt;th&gt;Documented description&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;features.multi_agent&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Enable multi-agent collaboration tools (&lt;code&gt;spawn_agent&lt;/code&gt;, &lt;code&gt;send_input&lt;/code&gt;, &lt;code&gt;resume_agent&lt;/code&gt;, &lt;code&gt;wait_agent&lt;/code&gt;, and &lt;code&gt;close_agent&lt;/code&gt;)" plus "(stable; on by default)."&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;agents.enabled&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Enable or disable multi-agent tools (default: true)."&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The same table does flag deprecation elsewhere: &lt;code&gt;features.codex_hooks&lt;/code&gt; is called a deprecated alias of &lt;code&gt;features.hooks&lt;/code&gt;, &lt;code&gt;features.web_search&lt;/code&gt; a "Deprecated legacy toggle", and &lt;code&gt;agents.max_threads&lt;/code&gt; a "Legacy alias" of &lt;code&gt;agents.max_concurrent_threads_per_session&lt;/code&gt;. &lt;code&gt;features.multi_agent&lt;/code&gt; carries none of those markers and is labelled stable.&lt;/p&gt;

&lt;p&gt;Given that a warning now appears at startup, &lt;code&gt;agents.enabled&lt;/code&gt; is the safer thing to write going forward. But &lt;strong&gt;which one is the official mechanism could not be confirmed in the official documentation.&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The other changes
&lt;/h2&gt;

&lt;p&gt;TUI additions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The side conversation supports multiple tabs, and saved local side chats reopen after a restart (v0.169.0)&lt;/li&gt;
&lt;li&gt;Turning on &lt;strong&gt;Collapse turn details&lt;/strong&gt; in the full-screen transcript folds completed activity beneath each prompt, keeping answers visible and tool output searchable (v0.169.0)&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;Ctrl+S&lt;/code&gt; in the terminal report editor saves a diagnostic ZIP locally, "even when uploads are disabled" (v0.169.0)&lt;/li&gt;
&lt;li&gt;Model providers are browsable under Sessions → Model providers, where you pick the default for new chats (v0.168.0)&lt;/li&gt;
&lt;/ul&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;There is no documented leader key&lt;/strong&gt;&lt;br&gt;
Describing &lt;code&gt;features.tui_workspace&lt;/code&gt;, the v0.168.0 notes say to "press &lt;code&gt;leader z&lt;/code&gt; to zoom the focused pane". But &lt;strong&gt;no leader-key concept could be found in the official documentation.&lt;/strong&gt; The documented binding syntax for &lt;code&gt;tui.keymap.&amp;lt;context&amp;gt;.&amp;lt;action&amp;gt;&lt;/code&gt; uses normalized names such as &lt;code&gt;ctrl-a&lt;/code&gt;, &lt;code&gt;shift-enter&lt;/code&gt;, &lt;code&gt;page-down&lt;/code&gt; and &lt;code&gt;minus&lt;/code&gt;, with no prefix-sequence form.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Beyond the trust change, the fixes cover Windows sandbox handling for quoted &lt;code&gt;cmd.exe&lt;/code&gt; paths, a terminal that stayed busy waiting on MCP servers a shared app server never reported, and interrupted turns not closing cleanly. For how approvals and sandboxing pair up, see &lt;a href="https://aicoding-guide.com/en/posts/codex-approval-sandbox/" rel="noopener noreferrer"&gt;Codex approval_policy versus sandbox_mode&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;codex exec&lt;/code&gt; does not read a project's configuration in an untrusted directory; &lt;code&gt;--sandbox workspace-write&lt;/code&gt; does not change that, so pass &lt;code&gt;--trust&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;To declare it in a file, set &lt;code&gt;projects.&amp;lt;path&amp;gt;.trust_level&lt;/code&gt; to &lt;code&gt;"trusted"&lt;/code&gt;. This is not the retired &lt;code&gt;approval_policy = "untrusted"&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;features.tui_workspace&lt;/code&gt;, &lt;code&gt;features.image_generation_animation&lt;/code&gt; and &lt;code&gt;incremental_tools_legacy_gcp&lt;/code&gt; have no entry in the configuration reference&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;tui.notification_method&lt;/code&gt; documents &lt;code&gt;auto&lt;/code&gt;, &lt;code&gt;osc9&lt;/code&gt; and &lt;code&gt;bel&lt;/code&gt;; &lt;code&gt;osc777&lt;/code&gt; is not among them&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;features.multi_agent&lt;/code&gt; now warns at startup, but the reference does not mark it deprecated&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>codex</category>
      <category>codexcli</category>
      <category>releasenotes</category>
      <category>codexexec</category>
    </item>
    <item>
      <title>Gemini CLI v0.68 fixes rootless Podman sandboxing and path boundary checks</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Sun, 15 Nov 2026 19:16:07 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/gemini-cli-v068-fixes-rootless-podman-sandboxing-and-path-boundary-checks-2520</link>
      <guid>https://dev.to/aicoding-guide/gemini-cli-v068-fixes-rootless-podman-sandboxing-and-path-boundary-checks-2520</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-68/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Gemini CLI v0.68.0 ships &lt;strong&gt;with no summary prose at all&lt;/strong&gt; — the release page is just a list of pull requests. One feature; everything else is a fix.&lt;/p&gt;

&lt;p&gt;It looks unremarkable, but &lt;strong&gt;three of the fixes are about boundaries&lt;/strong&gt;, and each one is a case where behaviour the docs already describe was not holding in some environment.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What you will learn&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;When &lt;code&gt;--userns=keep-id&lt;/code&gt; is passed for rootless Podman, and how to override it&lt;/li&gt;
&lt;li&gt;The fixes to root checks on Windows and to symlink resolution in GlobTool&lt;/li&gt;
&lt;li&gt;That the new extension validation command and the hooks' &lt;code&gt;SubagentStop&lt;/code&gt; are both absent from the docs&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  --userns=keep-id reaches rootless Podman
&lt;/h2&gt;

&lt;p&gt;The fix is &lt;code&gt;fix(sandbox): supply --userns=keep-id for rootless podman&lt;/code&gt; (#30118). This behaviour is documented, under "Linux UID/GID handling":&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Gemini CLI automatically handles user permissions in container sandboxes on Linux. When the sandbox runs as your host UID/GID (the default on supported Linux distributions) and you use rootless Podman, Gemini CLI also passes &lt;code&gt;--userns=keep-id&lt;/code&gt;. This keeps files that the sandbox creates in mounted directories owned by your host user. To use a different user namespace mode, add your own &lt;code&gt;--userns&lt;/code&gt; option to &lt;code&gt;SANDBOX_FLAGS&lt;/code&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;The promise is that files the sandbox writes into mounted directories stay owned by your host user. If ownership was turning into root under rootless Podman, that promise was not being kept until this release.&lt;/p&gt;

&lt;p&gt;There are three ways to select the backend, in a fixed order of precedence.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Method&lt;/th&gt;
&lt;th&gt;How&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Command flag&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;-s&lt;/code&gt; or &lt;code&gt;--sandbox&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Environment variable&lt;/td&gt;
&lt;td&gt;`GEMINI_SANDBOX=true\&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Settings file&lt;/td&gt;
&lt;td&gt;{% raw %}&lt;code&gt;"sandbox": true&lt;/code&gt; inside the &lt;code&gt;tools&lt;/code&gt; object of &lt;code&gt;settings.json&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# Turn on sandboxing with rootless Podman&lt;/span&gt;
&lt;span class="kd"&gt;export&lt;/span&gt; &lt;span class="nv"&gt;GEMINI_SANDBOX&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;podman
gemini

&lt;span class="c"&gt;# Pick a different user namespace mode yourself&lt;/span&gt;
&lt;span class="kd"&gt;export&lt;/span&gt; &lt;span class="nv"&gt;SANDBOX_FLAGS&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"--userns=host"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;SANDBOX_SET_UID_GID&lt;/code&gt; switches the mapping itself: &lt;code&gt;true&lt;/code&gt; forces the host UID/GID, &lt;code&gt;false&lt;/code&gt; disables UID/GID mapping.&lt;/p&gt;

&lt;h2&gt;
  
  
  Two fixes to path boundary checks
&lt;/h2&gt;

&lt;p&gt;The other two concern which paths count as inside your working area.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;fix(core): make isWithinRoot case-insensitive on Windows&lt;/code&gt; (#29999) — root checks on Windows no longer depend on case, which matters if &lt;code&gt;C:\Project&lt;/code&gt; and &lt;code&gt;c:\project&lt;/code&gt; were being treated as different places&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(core): resolve symlinked paths canonically in GlobTool&lt;/code&gt; (#30042) — GlobTool now resolves symlinked paths canonically&lt;/li&gt;
&lt;/ul&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Neither behaviour is stated in the documentation&lt;/strong&gt;&lt;br&gt;
The workspace concept itself is documented. The sandbox page says "the sandbox only has access to the current project workspace" and points you at &lt;code&gt;SANDBOX_MOUNTS&lt;/code&gt; when something outside it must be reachable.&lt;/p&gt;

&lt;p&gt;But &lt;strong&gt;whether path comparison is case-sensitive on Windows, and how the file tools resolve symlinks, could not be confirmed in the official documentation.&lt;/strong&gt; The PR titles tell you what changed; no specification-level statement exists. If you need the outside of the workspace genuinely blocked, turn the sandbox on and pass what you need explicitly with &lt;code&gt;SANDBOX_MOUNTS&lt;/code&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;There is also a fix about git's exclude files: &lt;code&gt;fix(core): honor .git/info/exclude and core.excludesFile in shadow repo&lt;/code&gt; (#30034). Here too, &lt;strong&gt;no documented statement that &lt;code&gt;.git/info/exclude&lt;/code&gt; or &lt;code&gt;core.excludesFile&lt;/code&gt; are honored could be found.&lt;/strong&gt; These are the exclude settings that are documented:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Setting&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;context.fileFiltering.respectGitIgnore&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Respect &lt;code&gt;.gitignore&lt;/code&gt; (default &lt;code&gt;true&lt;/code&gt;)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;context.fileFiltering.respectGeminiIgnore&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Respect &lt;code&gt;.geminiignore&lt;/code&gt; (default &lt;code&gt;true&lt;/code&gt;)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;context.fileFiltering.customIgnoreFilePaths&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Additional ignore file paths, which take precedence over &lt;code&gt;.geminiignore&lt;/code&gt; and &lt;code&gt;.gitignore&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;If you want an extra ignore file to count on, naming it in &lt;code&gt;customIgnoreFilePaths&lt;/code&gt; is the documented route.&lt;/p&gt;

&lt;h2&gt;
  
  
  The extension validation command and the hooks' SubagentStop
&lt;/h2&gt;

&lt;p&gt;The release's only feature is &lt;code&gt;feat(cli): add extension validation command and evaluation tests&lt;/code&gt; (#29980). &lt;strong&gt;But as of 2026-11-16 no &lt;code&gt;validate&lt;/code&gt; subcommand appears in the official documentation.&lt;/strong&gt; The documented &lt;code&gt;gemini extensions&lt;/code&gt; subcommands are:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;install / uninstall / disable / enable / update / new / link / list / config
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The slash-command form, &lt;code&gt;/extensions&lt;/code&gt;, lists ten: &lt;code&gt;config&lt;/code&gt;, &lt;code&gt;disable&lt;/code&gt;, &lt;code&gt;enable&lt;/code&gt;, &lt;code&gt;explore&lt;/code&gt;, &lt;code&gt;install&lt;/code&gt;, &lt;code&gt;link&lt;/code&gt;, &lt;code&gt;list&lt;/code&gt;, &lt;code&gt;restart&lt;/code&gt;, &lt;code&gt;uninstall&lt;/code&gt;, &lt;code&gt;update&lt;/code&gt;. No &lt;code&gt;validate&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;There is a hooks fix too: &lt;code&gt;fix(hooks): support SubagentStop event mapping and merge hook definitions&lt;/code&gt; (#30111). Read that one carefully. &lt;strong&gt;&lt;code&gt;SubagentStop&lt;/code&gt; does not appear in the documented list of hook events.&lt;/strong&gt; The documented events are these eleven:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;SessionStart / SessionEnd / BeforeAgent / AfterAgent / BeforeModel / AfterModel
BeforeToolSelection / BeforeTool / AfterTool / PreCompress / Notification
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;SubagentStop&lt;/code&gt; is a Claude Code event name. Gemini CLI does carry a compatibility environment variable, &lt;code&gt;CLAUDE_PROJECT_DIR&lt;/code&gt;, so this could plausibly be about mapping a migrated Claude Code configuration — but &lt;strong&gt;no official statement to that effect could be found.&lt;/strong&gt; For the hooks' &lt;code&gt;timeout&lt;/code&gt; being in milliseconds (default 60000), see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-67/" rel="noopener noreferrer"&gt;Gemini CLI v0.67 adds a models command and fixes the unit in Claude hook migration&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The other fixes
&lt;/h2&gt;

&lt;p&gt;Three concern model selection.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;fix(core,cli): preserve selected model across transient fallback, cancel, and rewind&lt;/code&gt; (#30059) — your chosen model survives a transient fallback, a cancel, and a rewind&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(core): preserve explicit gemini-2.5-flash model in dynamic model resolution&lt;/code&gt; (#30057) — an explicitly chosen &lt;code&gt;gemini-2.5-flash&lt;/code&gt; is not replaced by dynamic resolution&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(core): list Auto in /model when dynamicModelConfiguration is enabled and preview access is disabled&lt;/code&gt; (#29998)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The one you are most likely to notice day to day is &lt;code&gt;fix(cli): surface 429 rate limit cool-down and countdown in loading indicator&lt;/code&gt; (#30012): when you hit a rate limit, the cool-down and a countdown now appear in the loading indicator. For the free-tier ceilings, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-free-tier-limits/" rel="noopener noreferrer"&gt;Gemini CLI free tier limits and how paid plans differ&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;On extensions, concurrent installs are now serialized with a per-extension lock (#30039), and stream failures during extension release downloads are handled (#30100). For MCP server configuration, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-mcp/" rel="noopener noreferrer"&gt;Add an MCP server to Gemini CLI&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;v0.68.0 has no summary prose in its release notes: one feature and 28 fixes&lt;/li&gt;
&lt;li&gt;Rootless Podman gets &lt;code&gt;--userns=keep-id&lt;/code&gt;, which is documented behaviour; to use another mode, add &lt;code&gt;--userns&lt;/code&gt; to &lt;code&gt;SANDBOX_FLAGS&lt;/code&gt; yourself&lt;/li&gt;
&lt;li&gt;Root checks on Windows and GlobTool symlink resolution were fixed, but neither is described as a specification in the docs&lt;/li&gt;
&lt;li&gt;The new extension validation command and the hooks' &lt;code&gt;SubagentStop&lt;/code&gt; are both missing from the official documentation&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>gemini</category>
      <category>releasenotes</category>
      <category>sandbox</category>
      <category>podman</category>
    </item>
    <item>
      <title>This week in Claude Code, Codex and Gemini CLI (week of November 15, 2026)</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Sat, 14 Nov 2026 19:11:52 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/this-week-in-claude-code-codex-and-gemini-cli-week-of-november-15-2026-4pmm</link>
      <guid>https://dev.to/aicoding-guide/this-week-in-claude-code-codex-and-gemini-cli-week-of-november-15-2026-4pmm</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/weekly-2026-11-15/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;The biggest change this week is &lt;strong&gt;four permission and sandbox fixes in Claude Code v2.1.321&lt;/strong&gt;. All four are cases where a prompt that should have appeared did not, so your behaviour changes without you touching any setting. There is no reason to defer this update.&lt;/p&gt;

&lt;p&gt;On the feature side, v2.1.323 adds &lt;strong&gt;Code Mode&lt;/strong&gt;, an experimental opt-in where Claude writes JavaScript to call its tools. Codex shipped four releases, 0.167.1 through 0.169.0.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
This week in brief&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Claude Code v2.1.321 fixes four permission and sandbox gaps; v2.1.325 fixes a security issue in zstd decoding&lt;/li&gt;
&lt;li&gt;v2.1.323 adds Code Mode as an experimental opt-in, plus &lt;code&gt;engines&lt;/code&gt; version-range gating in &lt;code&gt;plugin.json&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Codex v0.169.0 changes how &lt;code&gt;codex exec&lt;/code&gt; decides trust: &lt;code&gt;--sandbox workspace-write&lt;/code&gt; alone no longer enables project config or MCP&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Claude Code
&lt;/h2&gt;

&lt;p&gt;Four releases: v2.1.321, v2.1.322, v2.1.323 and v2.1.325. &lt;strong&gt;There is no v2.1.320 and no v2.1.324&lt;/strong&gt; — the CHANGELOG has no entry for either.&lt;/p&gt;

&lt;h3&gt;
  
  
  v2.1.321
&lt;/h3&gt;

&lt;p&gt;Four permission and sandbox fixes, quoted from the CHANGELOG:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;"Fixed a case where a read-only Bash command ran outside the sandbox without a permission prompt"&lt;/li&gt;
&lt;li&gt;"Fixed a case where Bash permission checks judged a file write from the wrong folder and didn't prompt"&lt;/li&gt;
&lt;li&gt;"Fixed deny rules and protected-file prompts not applying to the file that a &lt;code&gt;cp&lt;/code&gt; or &lt;code&gt;mv&lt;/code&gt; into a directory writes"&lt;/li&gt;
&lt;li&gt;"Fixed MCP OAuth requests sending secrets such as refresh tokens to another host when a server redirected them"&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The third is the easiest to miss in practice. When you name a directory as the destination, as in &lt;code&gt;cp src/x.env dist/&lt;/code&gt;, the file actually written (&lt;code&gt;dist/x.env&lt;/code&gt;) was not checked against deny rules or the protected-files prompt. If you rely on deny rules to guard paths, that route was open until this fix. For how to write those rules, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-settings-json-permissions/" rel="noopener noreferrer"&gt;Claude Code permissions in settings.json and settings.local.json&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;A model landed too: "Added Claude Fable 5.2 (&lt;code&gt;claude-fable-5-2&lt;/code&gt;), now the default Fable model — 1M context, $8/$40 per Mtok with $0.20/Mtok cache reads" (confirmed in the CHANGELOG on 2026-11-15).&lt;/p&gt;

&lt;p&gt;There is also a &lt;code&gt;/config&lt;/code&gt; display fix. Where new sessions start in auto mode, the default permission mode was shown as Manual; and pressing Space on the "Auto (default)" row now saves Don't Ask rather than Plan. For how the modes differ, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-default-mode/" rel="noopener noreferrer"&gt;Claude Code starts in auto mode by default&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  v2.1.322
&lt;/h3&gt;

&lt;p&gt;A single bullet: "Added fast mode on Vertex in the global, us and eu regions, falling back to standard speed if Vertex refuses it".&lt;/p&gt;

&lt;h3&gt;
  
  
  v2.1.323
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;Code Mode&lt;/strong&gt; arrives: "Added Code Mode, an experimental opt-in in which Claude writes JavaScript to call its tools", with &lt;code&gt;https://code.claude.com/docs/en/code-mode&lt;/code&gt; covering how to turn it on, or off for an organization. Being an opt-in, it does nothing by default.&lt;/p&gt;

&lt;p&gt;Two plugin changes matter. &lt;code&gt;plugin.json&lt;/code&gt; gains &lt;code&gt;engines["claude-code"]&lt;/code&gt;: a plugin outside its stated version range is not installed, updated or loaded, and install and update pick the newest version that fits. And &lt;code&gt;claude plugin validate&lt;/code&gt; now type-checks a plugin's TypeScript hooks module — each finding is a warning, and &lt;code&gt;--strict&lt;/code&gt; fails on it.&lt;/p&gt;

&lt;p&gt;There is a new warning about rule syntax: "Added a start-up warning for Bash allow rules with a &lt;code&gt;*&lt;/code&gt; in a variable's value, such as &lt;code&gt;Bash(PORT=* npm run dev)&lt;/code&gt;".&lt;/p&gt;

&lt;p&gt;Worth noting among the fixes: running &lt;code&gt;/compact&lt;/code&gt; again, or after a command such as &lt;code&gt;/context&lt;/code&gt;, no longer summarizes away a message Claude had not answered yet.&lt;/p&gt;

&lt;h3&gt;
  
  
  v2.1.325
&lt;/h3&gt;

&lt;p&gt;One security fix leads the entry: "Fixed a security issue in how zstd-compressed responses are decoded".&lt;/p&gt;

&lt;p&gt;Two fixes help long sessions. One is "Fixed sessions on many third-party gateways and Foundry stopping with a context length error on every turn instead of compacting". The other restores recent turns that could go missing from a resumed session after Claude Code stopped without a clean exit.&lt;/p&gt;

&lt;p&gt;Subagents get two fixes: one where a subagent with its own model was moved to the plan-mode model when the model setting is &lt;code&gt;opusplan&lt;/code&gt; or &lt;code&gt;haiku&lt;/code&gt;, and one where a subagent on a different model sized file reads and MCP results using the main session's model.&lt;/p&gt;

&lt;h2&gt;
  
  
  Codex
&lt;/h2&gt;

&lt;p&gt;Four releases: 0.167.1, 0.167.2, 0.168.0 and 0.169.0. &lt;strong&gt;0.167.1 shipped with no release notes at all&lt;/strong&gt; — the page carries only a compare link.&lt;/p&gt;

&lt;h3&gt;
  
  
  v0.167.2
&lt;/h3&gt;

&lt;p&gt;One item. On Google Cloud Vertex AI, Codex now sends tool updates incrementally by default. To opt out, set &lt;code&gt;incremental_tools_legacy_gcp = false&lt;/code&gt; under &lt;code&gt;[features]&lt;/code&gt; in &lt;code&gt;config.toml&lt;/code&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  v0.168.0
&lt;/h3&gt;

&lt;p&gt;Three new configuration keys.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Key&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;features.tui_workspace&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;View a side conversation beside the main chat. Drag the divider to resize the panes, or press &lt;code&gt;leader z&lt;/code&gt; to zoom the focused pane&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;features.image_generation_animation&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Generate or edit animated GIFs, with optional frame count and timing controls&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;tui.notification_method = "osc777"&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Terminal notifications. Approval notifications show longer command previews and distinguish terminal input from network access requests&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;One deprecation notice: "Codex warns when &lt;code&gt;features.multi_agent&lt;/code&gt; is explicitly configured and points to &lt;code&gt;agents.enabled&lt;/code&gt; for disabling subagent tools."&lt;/p&gt;

&lt;h3&gt;
  
  
  v0.169.0
&lt;/h3&gt;

&lt;p&gt;One behavioural change is the thing to know here.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;How codex exec decides trust has changed&lt;/strong&gt;&lt;br&gt;
"&lt;code&gt;codex exec&lt;/code&gt; keeps project configuration and MCP servers disabled in untrusted directories, even with &lt;code&gt;--sandbox workspace-write&lt;/code&gt;. Use &lt;code&gt;--trust&lt;/code&gt; with &lt;code&gt;codex exec&lt;/code&gt; or &lt;code&gt;codex review&lt;/code&gt; to explicitly save project trust."&lt;/p&gt;

&lt;p&gt;In an untrusted directory, &lt;code&gt;--sandbox workspace-write&lt;/code&gt; is no longer enough to bring project config and MCP servers in. If a CI job using &lt;code&gt;codex exec&lt;/code&gt; has stopped loading your MCP servers, pass &lt;code&gt;--trust&lt;/code&gt;. For how trust levels work, see &lt;a href="https://aicoding-guide.com/en/posts/codex-project-trust-level/" rel="noopener noreferrer"&gt;Stop Codex asking whether to trust this folder with trust_level in config.toml&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Among the new features, the opt-in terminal workspace now supports multiple side-chat tabs, switchable by keyboard or mouse, and saved local side chats reopen after a Codex restart. For getting diagnostics out of a locked-down environment: "Press &lt;code&gt;Ctrl+S&lt;/code&gt; in the terminal report editor to save a diagnostic ZIP locally, even when uploads are disabled."&lt;/p&gt;

&lt;h2&gt;
  
  
  Gemini CLI
&lt;/h2&gt;

&lt;p&gt;One release, v0.68.0. It ships &lt;strong&gt;with no summary prose at all&lt;/strong&gt; — just a list of pull requests. One feature, the rest fixes.&lt;/p&gt;

&lt;p&gt;The ones touching sandboxing and path handling:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;fix(sandbox): supply --userns=keep-id for rootless podman&lt;/code&gt; (#30118) — sandboxed execution under rootless podman&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(core): make isWithinRoot case-insensitive on Windows&lt;/code&gt; (#29999) — root checks no longer depend on case&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(core): resolve symlinked paths canonically in GlobTool&lt;/code&gt; (#30042) — symlinked paths resolved canonically&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(core): honor .git/info/exclude and core.excludesFile in shadow repo&lt;/code&gt; (#30034) — both exclude sources respected&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;fix(hooks): support SubagentStop event mapping and merge hook definitions&lt;/code&gt; (#30111) — hooks gain &lt;code&gt;SubagentStop&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The one feature is &lt;code&gt;feat(cli): add extension validation command and evaluation tests&lt;/code&gt; (#29980), which adds a command for validating extensions. Rate-limit feedback improved too: a 429 cool-down and countdown now appear in the loading indicator (#30012).&lt;/p&gt;

&lt;h2&gt;
  
  
  What to check now
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Update Claude Code to v2.1.321 or later.&lt;/strong&gt; Four permission gaps and a security fix. No configuration changes needed&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Re-examine the paths your deny rules guard.&lt;/strong&gt; If anything in your workflow uses &lt;code&gt;cp&lt;/code&gt; or &lt;code&gt;mv&lt;/code&gt; with a directory destination, that route was unchecked before v2.1.321&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Look for allow rules shaped like &lt;code&gt;Bash(PORT=* npm run dev)&lt;/code&gt;.&lt;/strong&gt; From v2.1.323 these produce a start-up warning&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If CI uses &lt;code&gt;codex exec&lt;/code&gt;, check whether you now need &lt;code&gt;--trust&lt;/code&gt;.&lt;/strong&gt; From v0.169.0, MCP servers stay disabled in untrusted directories&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If Codex on Vertex AI misbehaves, try &lt;code&gt;incremental_tools_legacy_gcp = false&lt;/code&gt;.&lt;/strong&gt; Incremental sending became the default in v0.167.2&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Nine releases this week: four for Claude Code, four for Codex, one for Gemini CLI&lt;/li&gt;
&lt;li&gt;Top priority is Claude Code v2.1.321's four permission and sandbox fixes, plus the zstd decoding fix in v2.1.325&lt;/li&gt;
&lt;li&gt;v2.1.323 adds Code Mode as an experimental opt-in, and plugins are now gated by &lt;code&gt;engines["claude-code"]&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Codex v0.169.0 changes &lt;code&gt;codex exec&lt;/code&gt; trust handling, so &lt;code&gt;--sandbox workspace-write&lt;/code&gt; alone no longer enables project config or MCP&lt;/li&gt;
&lt;li&gt;Gemini CLI v0.68.0 is mostly sandbox and path-resolution fixes, with one new extension validation command&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>weeklydigest</category>
      <category>updates</category>
      <category>claudecode</category>
      <category>codex</category>
    </item>
    <item>
      <title>Designing allow, deny and hooks is becoming the new code review</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Thu, 12 Nov 2026 19:17:08 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/designing-allow-deny-and-hooks-is-becoming-the-new-code-review-3epg</link>
      <guid>https://dev.to/aicoding-guide/designing-allow-deny-and-hooks-is-becoming-the-new-code-review-3epg</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/essay-permissions-are-code-review/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;On a team that uses coding agents daily, &lt;strong&gt;the highest-leverage thing to review shifts from the code itself to the design of allow, deny and hooks.&lt;/strong&gt; Reading diffs afterwards does not go away. But a layer appears in front of it, and that layer stops more incidents than the reading does.&lt;/p&gt;

&lt;p&gt;The argument rests on a line the tools' own documentation draws, not on a hunch.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
The claim&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The docs separate instruction from enforcement, and CLAUDE.md sits on the instruction side&lt;/li&gt;
&lt;li&gt;Permission rules and hooks are the executable form of the review checklist items a machine can decide&lt;/li&gt;
&lt;li&gt;But permissions never review correctness — this is a layer added in front of review, not a replacement&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  The docs draw the line explicitly
&lt;/h2&gt;

&lt;p&gt;The Claude Code memory page settles the question about organization-wide CLAUDE.md in one sentence:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Settings rules are enforced by the client regardless of what Claude decides to do. CLAUDE.md instructions shape Claude's behavior but are not a hard enforcement layer.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;The same page names where to go when you need certainty:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Claude treats them as context, not enforced configuration. To block an action regardless of what Claude decides, use a PreToolUse hook instead.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;A document holding your project's rules is something that gets &lt;strong&gt;read, and followed on a best-effort basis&lt;/strong&gt;. Its content arrives as a user message after the system prompt, not as part of it. No amount of work sharpening its prose changes which side of the line it sits on.&lt;/p&gt;

&lt;p&gt;That line maps directly onto review. Writing "please don't force push" in a review comment and writing it into &lt;code&gt;permissions.deny&lt;/code&gt; differ in strength. The first may need writing again next week; the second is finished when you write it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Permission design is the review checklist made executable
&lt;/h2&gt;

&lt;p&gt;A review checklist mixes two kinds of item: the ones a person must read to judge — design intent, naming, whether the abstraction holds — and the ones a machine can decide, such as "did this touch production config" or "did a credential reach the logs."&lt;/p&gt;

&lt;p&gt;The second kind can be written as a rule. And when you do, the docs guarantee different strengths.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Where it lives&lt;/th&gt;
&lt;th&gt;What the docs say&lt;/th&gt;
&lt;th&gt;Strength&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;permissions.deny&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Blocks before the classifier is consulted. Neither the classifier nor user intent can override it"&lt;/td&gt;
&lt;td&gt;Before the classifier; intent cannot reach it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;permissions.ask&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Always prompts for a command that matches a content-scoped rule like the recipe above. The classifier cannot auto-approve a matching action"&lt;/td&gt;
&lt;td&gt;Always reaches a human&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Critical-path removals&lt;/td&gt;
&lt;td&gt;"Recognized &lt;code&gt;rm&lt;/code&gt; and &lt;code&gt;rmdir&lt;/code&gt; critical-path removals, which no allow rule or &lt;code&gt;PreToolUse&lt;/code&gt; hook &lt;code&gt;\"allow\"&lt;/code&gt; approves"&lt;/td&gt;
&lt;td&gt;Neither a rule nor a hook can approve it&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;PreToolUse hook&lt;/td&gt;
&lt;td&gt;"To block an action regardless of what Claude decides"&lt;/td&gt;
&lt;td&gt;Runs without going through Claude's judgement&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;CLAUDE.md&lt;/td&gt;
&lt;td&gt;"not a hard enforcement layer"&lt;/td&gt;
&lt;td&gt;Instruction&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The top four rows are checklist items you can express as configuration instead of comments. For anything in &lt;code&gt;deny&lt;/code&gt;, the probability of missing it in review is not reduced — it leaves the realm of probability.&lt;/p&gt;

&lt;p&gt;For the mechanics, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-settings-json-permissions/" rel="noopener noreferrer"&gt;Claude Code permissions in settings.json and settings.local.json&lt;/a&gt; and &lt;a href="https://aicoding-guide.com/en/posts/claude-code-permissions-examples/" rel="noopener noreferrer"&gt;Claude Code permissions by example&lt;/a&gt;. For the hook side and where it stops being reliable, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-hooks-block-rm/" rel="noopener noreferrer"&gt;Block rm -rf with a Claude Code PreToolUse hook, and know its limits&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  There are three layers to design
&lt;/h2&gt;

&lt;p&gt;"Designing permissions" covers three distinct things.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Per-tool allow, ask and deny.&lt;/strong&gt; In Claude Code that is &lt;code&gt;permissions&lt;/code&gt;; in Codex it is named profiles, &lt;code&gt;[permissions.&amp;lt;name&amp;gt;]&lt;/code&gt;, with &lt;code&gt;filesystem&lt;/code&gt;, &lt;code&gt;workspace_roots&lt;/code&gt; and the &lt;code&gt;network&lt;/code&gt; keys beneath them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Lifecycle hooks.&lt;/strong&gt; PreToolUse and its siblings run without passing through the model's judgement. Only what you put here is in the "always" category.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The execution environment itself&lt;/strong&gt; — the sandbox and the reach of the working directory. Codex ships &lt;code&gt;:read-only&lt;/code&gt;, &lt;code&gt;:workspace&lt;/code&gt; and &lt;code&gt;:danger-full-access&lt;/code&gt; as built-in profiles, which &lt;code&gt;extends&lt;/code&gt; can narrow. For how approvals and sandboxing pair up, see &lt;a href="https://aicoding-guide.com/en/posts/codex-approval-sandbox/" rel="noopener noreferrer"&gt;Codex approval_policy versus sandbox_mode&lt;/a&gt;.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Three layers exist because one leaks. Claude Code's auto mode suspends broad rules the moment you enter it:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;By default, narrow Bash and PowerShell allow rules such as &lt;code&gt;Bash(npm test)&lt;/code&gt; stay in effect in auto mode. Claude Code resolves them before the classifier runs, unless the command carries per-command allowed domains. Claude Code suspends only the broad rules that grant arbitrary code execution, such as &lt;code&gt;Bash(*)&lt;/code&gt; or wildcarded interpreters, together with every rule that names &lt;code&gt;Monitor&lt;/code&gt;, because Monitor commands run through the shell.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;So a blanket &lt;code&gt;Bash(*)&lt;/code&gt; buys nothing in auto mode. Keeping the grain fine is not style advice; it is a precondition for the rule working at all. For how the modes differ, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-default-mode/" rel="noopener noreferrer"&gt;Claude Code starts in auto mode by default&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Objections and limits
&lt;/h2&gt;

&lt;p&gt;Four conditions make this claim fail.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Permissions never review correctness.&lt;/strong&gt; This is the hard limit. &lt;code&gt;deny&lt;/code&gt; stops writes to places that are off-limits; it says nothing about whether the logic written in an allowed place is right. Algorithm choice, boundary conditions, naming and design coherence stay with human readers. Permission design is not a substitute for review — it carves out the part review no longer has to carry.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;It does not pay off where nothing is irreversible.&lt;/strong&gt; On a local library with no deploys and no outbound calls, the design cost is pure prepayment. The argument holds where production, an external service, or credentials are within reach.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Rules cannot express intent.&lt;/strong&gt; "Don't touch this directory, except during a migration" resists being written as a rule. Claude Code's classifier does let user intent override blocks, but within tight bounds:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;code&gt;hard_deny&lt;/code&gt; rules block unconditionally. User intent and &lt;code&gt;allow&lt;/code&gt; exceptions don't apply.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Intent only reaches &lt;code&gt;soft_deny&lt;/code&gt;, and even there the required specificity is stated: "General requests don't count as explicit intent. Asking Claude to 'clean up the repo' doesn't authorize force-pushing, but asking Claude to 'force-push this branch' does." Judgement that a rule cannot capture goes back to an in-the-moment prompt.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The configuration itself becomes something to review.&lt;/strong&gt; A permissions file grows and turns into another thing that is hard to read. Review moves; it does not disappear.&lt;/p&gt;

&lt;h2&gt;
  
  
  Three things to decide first
&lt;/h2&gt;

&lt;p&gt;Given those limits, the list that actually earns its place is short.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Put irreversible operations in &lt;code&gt;deny&lt;/code&gt;.&lt;/strong&gt; Force push, production deploys, and the paths an &lt;code&gt;rm -rf&lt;/code&gt; could reach. That is enough for the layer intent cannot override.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Put anything that leaves the machine in &lt;code&gt;ask&lt;/code&gt;.&lt;/strong&gt; Pushes, writes to external APIs, reads of credentials. Not blocked — guaranteed to reach a person.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep the rest narrow.&lt;/strong&gt; When you want to write &lt;code&gt;Bash(*)&lt;/code&gt;, the question is not what to allow but which permission mode you are in.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Together that is the exercise of pulling out of your review checklist the items you keep writing every week. Whatever you pull out is reading time returned to design and correctness.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;The docs separate instruction from enforcement, and CLAUDE.md is not on the enforcement side&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;permissions.deny&lt;/code&gt;, &lt;code&gt;ask&lt;/code&gt;, critical paths and PreToolUse hooks are the executable form of review items&lt;/li&gt;
&lt;li&gt;Three layers matter: rules, hooks and the execution environment. Overly broad allow rules are suspended in auto mode&lt;/li&gt;
&lt;li&gt;Permissions never review correctness; this is a layer in front of code review, not a replacement&lt;/li&gt;
&lt;li&gt;Start with deny for irreversible operations and ask for anything that leaves the machine&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>aicoding</category>
      <category>permissiondesign</category>
      <category>codereview</category>
      <category>permissions</category>
    </item>
    <item>
      <title>Every Claude Code slash command, ordered by how often you actually use it</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Tue, 10 Nov 2026 19:23:34 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/every-claude-code-slash-command-ordered-by-how-often-you-actually-use-it-jlc</link>
      <guid>https://dev.to/aicoding-guide/every-claude-code-slash-command-ordered-by-how-often-you-actually-use-it-jlc</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/claude-code-slash-commands/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;The official Claude Code command reference lists more than 110 slash commands. You do not need to know them all. About six come up every day; the rest are things you look up when something goes wrong. This article regroups the official list by how often you actually reach for each one.&lt;/p&gt;

&lt;p&gt;One thing to get out of the way first: &lt;strong&gt;the built-in command list has moved.&lt;/strong&gt;&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What you will learn&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Where the official list lives, and how to read &lt;code&gt;&amp;lt;arg&amp;gt;&lt;/code&gt; versus &lt;code&gt;[arg]&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;The six commands you use in every session, and their arguments&lt;/li&gt;
&lt;li&gt;What the Skill and Workflow markers mean, and why availability differs per person&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Where the list lives and how to read it
&lt;/h2&gt;

&lt;p&gt;The official built-in command reference is at &lt;a href="https://code.claude.com/docs/en/commands" rel="noopener noreferrer"&gt;https://code.claude.com/docs/en/commands&lt;/a&gt;. The &lt;code&gt;/docs/en/slash-commands&lt;/code&gt; URL now serves a page titled "Extend Claude with skills", which has no built-in commands table. That page points you onward itself: "For built-in commands like &lt;code&gt;/help&lt;/code&gt; and &lt;code&gt;/compact&lt;/code&gt;, and bundled skills like &lt;code&gt;/debug&lt;/code&gt; and &lt;code&gt;/code-review&lt;/code&gt;, see the commands reference".&lt;/p&gt;

&lt;p&gt;The table has just &lt;strong&gt;two columns, Command and Purpose&lt;/strong&gt;. Argument notation is fixed.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Notation&lt;/th&gt;
&lt;th&gt;Meaning&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;&amp;lt;arg&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Required. The docs say "&lt;code&gt;&amp;lt;arg&amp;gt;&lt;/code&gt; indicates a required argument"&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;[arg]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Optional. "&lt;code&gt;[arg]&lt;/code&gt; indicates an optional one"&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;From inside the CLI, &lt;code&gt;/help&lt;/code&gt; gives you "Show help and available commands". When you want to know what you actually have rather than what exists, that is the reliable source.&lt;/p&gt;

&lt;h2&gt;
  
  
  The six you use every session
&lt;/h2&gt;

&lt;p&gt;Learn these and day-to-day work is covered.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Command&lt;/th&gt;
&lt;th&gt;What the docs say (condensed)&lt;/th&gt;
&lt;th&gt;When to reach for it&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/clear [name]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Start a new conversation with empty context. A name labels the previous conversation in the &lt;code&gt;/resume&lt;/code&gt; picker. Aliases &lt;code&gt;/reset&lt;/code&gt;, &lt;code&gt;/new&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Switching to unrelated work&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/compact [instructions]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;"Free up context by summarizing the conversation so far". You can pass focus instructions for the summary&lt;/td&gt;
&lt;td&gt;Freeing context without losing the thread&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/context [all]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Visualize context usage as a colored grid. Past the context window, the output warns how far over you are and which command frees space&lt;/td&gt;
&lt;td&gt;Finding out what made the session heavy&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/model [model]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Switch model and save it as the default for new sessions. With no argument a picker opens; press &lt;code&gt;s&lt;/code&gt; on a row to switch &lt;strong&gt;for this session only&lt;/strong&gt;
&lt;/td&gt;
&lt;td&gt;Before and after a hard task&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/resume [session]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Resume a conversation by ID or name, or open the session picker. Alias &lt;code&gt;/continue&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Picking up yesterday's work&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/rewind&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Rewind the conversation and/or code to a previous point, or summarize from a selected message. Aliases &lt;code&gt;/checkpoint&lt;/code&gt;, &lt;code&gt;/undo&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;After Claude heads the wrong way&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The split between &lt;code&gt;/clear&lt;/code&gt; and &lt;code&gt;/compact&lt;/code&gt; is stated right in the &lt;code&gt;/clear&lt;/code&gt; entry: "To free up context while continuing the same conversation, use &lt;code&gt;/compact&lt;/code&gt; instead." One discards the conversation; the other summarizes and keeps it. For the longer version see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-context-compact/" rel="noopener noreferrer"&gt;Managing context in Claude Code with /compact versus /clear&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;To move the threshold where automatic compaction fires, use &lt;code&gt;/autocompact [auto|&amp;lt;tokens&amp;gt;]&lt;/code&gt;. Pass a size such as &lt;code&gt;500k&lt;/code&gt;, or &lt;code&gt;auto&lt;/code&gt; to return to the window tuned for your model; the value is saved to user settings and applied to the current session. Requires Claude Code v2.1.221 or later.&lt;/p&gt;

&lt;h2&gt;
  
  
  Checking settings, permissions and usage
&lt;/h2&gt;

&lt;p&gt;The next most frequent group answers "what is the state of things right now".&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Command&lt;/th&gt;
&lt;th&gt;What the docs say (condensed)&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/permissions&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Interactive dialog for allow, ask and deny rules. View rules by scope, add or remove them, manage working directories, and review recent auto mode denials. The &lt;strong&gt;Auto mode&lt;/strong&gt; tab also shows and edits classifier rules. Alias &lt;code&gt;/allowed-tools&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/config [key=value ...]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Open the Settings interface, or set a value directly without opening it, such as &lt;code&gt;/config theme=dark&lt;/code&gt;. &lt;code&gt;/config --help&lt;/code&gt; lists the keys it accepts. Alias &lt;code&gt;/settings&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/status&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Open Settings on the Status tab: version, model, account and connectivity. Works while Claude is responding&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/usage&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Usage. &lt;code&gt;/cost&lt;/code&gt; is "Alias for &lt;code&gt;/usage&lt;/code&gt;", and &lt;code&gt;/stats&lt;/code&gt; opens it on the Stats tab&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/memory&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Edit &lt;code&gt;CLAUDE.md&lt;/code&gt; files, enable or disable auto memory, and view auto memory entries&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/hooks&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;View hook configurations&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/mcp&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Manage MCP server connections and OAuth. Pass &lt;code&gt;reconnect&lt;/code&gt;, &lt;code&gt;enable&lt;/code&gt; or &lt;code&gt;disable&lt;/code&gt; with a server name or &lt;code&gt;all&lt;/code&gt; to change state without opening the list&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/add-dir &amp;lt;path&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Add a working directory for file access during this session. Partial paths suggest matches; &lt;code&gt;Tab&lt;/code&gt; accepts one&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/cd &amp;lt;path&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Move the session to a new working directory, keeping the conversation&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;code&gt;/add-dir&lt;/code&gt; carries a caveat worth knowing: "Most &lt;code&gt;.claude/&lt;/code&gt; configuration isn't discovered from the added directory." For reading the permissions dialog, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-permissions-command/" rel="noopener noreferrer"&gt;Claude Code /permissions shows every rule and the settings file it came from&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Handing the work itself over, with Skills and Workflows
&lt;/h2&gt;

&lt;p&gt;Some rows in the list carry a marker. The official definitions:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Glossary&lt;/strong&gt;&lt;br&gt;
&lt;strong&gt;Skill&lt;/strong&gt;: "a bundled skill. It works like skills you write yourself: a prompt handed to Claude."&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Workflow&lt;/strong&gt;: "a bundled dynamic workflow that fans work out across many subagents and runs in the background."&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Both only run on demand — the docs make the point with &lt;code&gt;/verify&lt;/code&gt; and &lt;code&gt;/deep-research&lt;/code&gt;: "runs only when you invoke it."&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Command&lt;/th&gt;
&lt;th&gt;Kind&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/code-review&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Skill&lt;/td&gt;
&lt;td&gt;Review the current diff, or a PR number, branch or path you pass, for correctness bugs. &lt;code&gt;--fix&lt;/code&gt; applies findings, &lt;code&gt;--comment&lt;/code&gt; posts them to the PR or merge request. Alias &lt;code&gt;/review&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/simplify [target]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Skill&lt;/td&gt;
&lt;td&gt;Review changed code for cleanup opportunities and apply the fixes, with four review agents in parallel. "The review doesn't look for correctness bugs" — that is &lt;code&gt;/code-review&lt;/code&gt;'s job&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/debug [description]&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Skill&lt;/td&gt;
&lt;td&gt;Turn on debug logging for this session and troubleshoot from the log. Logging is off by default, so running it mid-session captures from that point forward&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/doctor&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Skill&lt;/td&gt;
&lt;td&gt;A setup checkup over installation, settings, extension and &lt;code&gt;CLAUDE.md&lt;/code&gt; problems, proposing fixes you confirm. Alias &lt;code&gt;/checkup&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/batch &amp;lt;instruction&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Skill&lt;/td&gt;
&lt;td&gt;Parallelize a large change: decompose into 5 to 30 independent units, present a plan, then run one background subagent per unit in an isolated worktree&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/deep-research &amp;lt;question&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Workflow&lt;/td&gt;
&lt;td&gt;Fan out web searches, fetch and cross-check sources, and synthesize a cited report&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/security-review&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;—&lt;/td&gt;
&lt;td&gt;Analyze the changes on your current branch for security vulnerabilities. Needs an &lt;code&gt;origin&lt;/code&gt; remote&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;To build commands on the same mechanism yourself, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-custom-slash-commands/" rel="noopener noreferrer"&gt;Create your own slash commands in Claude Code with a single Markdown file&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Commands that never appear, and ones that went away
&lt;/h2&gt;

&lt;p&gt;A command can be in the list and still be missing from your session. That is by design: "Not every command appears for every user. Availability depends on your platform, plan, and environment." The docs give two examples.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;/desktop&lt;/code&gt; shows only on macOS and x64 Windows, signed in with a Claude subscription&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;/upgrade&lt;/code&gt; does not show on Enterprise plans&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Some commands are genuinely gone.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Command&lt;/th&gt;
&lt;th&gt;Status&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/pr-comments&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Removed in v2.1.91. Ask Claude directly to view pull request comments&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/ultraplan&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Removed. Use plan mode instead&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;/agents&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Now "Print a reminder to ask Claude to create or manage subagents". On v2.1.197 and earlier it opened an interactive management interface&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Some commands are only hidden from the menu&lt;/strong&gt;&lt;br&gt;
&lt;code&gt;/heapdump&lt;/code&gt;, &lt;code&gt;/setup-bedrock&lt;/code&gt; and &lt;code&gt;/setup-vertex&lt;/code&gt; are documented as "Hidden from the command menu" — they do not autocomplete, but typing them in full works. Setting &lt;code&gt;CLAUDE_CODE_USE_BEDROCK=1&lt;/code&gt; or &lt;code&gt;CLAUDE_CODE_USE_VERTEX=1&lt;/code&gt; makes the matching wizard appear in the menu.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;The official built-in list is at &lt;code&gt;/docs/en/commands&lt;/code&gt;; &lt;code&gt;/docs/en/slash-commands&lt;/code&gt; is now the skills page&lt;/li&gt;
&lt;li&gt;The table has two columns, and &lt;code&gt;&amp;lt;arg&amp;gt;&lt;/code&gt; is required while &lt;code&gt;[arg]&lt;/code&gt; is optional&lt;/li&gt;
&lt;li&gt;The daily six are &lt;code&gt;/clear&lt;/code&gt;, &lt;code&gt;/compact&lt;/code&gt;, &lt;code&gt;/context&lt;/code&gt;, &lt;code&gt;/model&lt;/code&gt;, &lt;code&gt;/resume&lt;/code&gt; and &lt;code&gt;/rewind&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Skill means a prompt handed to Claude; Workflow means work fanned out across subagents. Both run only when invoked&lt;/li&gt;
&lt;li&gt;Platform, plan and environment decide what you see, so check your own list with &lt;code&gt;/help&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>claudecode</category>
      <category>slashcommands</category>
      <category>commandreference</category>
      <category>cli</category>
    </item>
    <item>
      <title>What changed in Codex CLI v0.167, codex agents and the auto-compaction fix</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Sun, 08 Nov 2026 19:27:07 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/what-changed-in-codex-cli-v0167-codex-agents-and-the-auto-compaction-fix-542b</link>
      <guid>https://dev.to/aicoding-guide/what-changed-in-codex-cli-v0167-codex-agents-and-the-auto-compaction-fix-542b</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-167/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Codex CLI v0.167.0 ships six new features, six bug fixes, two documentation changes and two chores. Two of them matter for everyday work: the &lt;strong&gt;agent command center&lt;/strong&gt; gets unread markers, and &lt;strong&gt;automatic context compaction&lt;/strong&gt; now counts tokens it used to miss.&lt;/p&gt;

&lt;p&gt;One caveat first. This release contains a flag spelling that the official documentation contradicts outright, so it is worth knowing before a command fails on you.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What you will learn&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;What &lt;code&gt;codex agents&lt;/code&gt; opens, and whether you may pass &lt;code&gt;--no-daemon&lt;/code&gt; to it&lt;/li&gt;
&lt;li&gt;What automatic context compaction now counts, and where to set its threshold&lt;/li&gt;
&lt;li&gt;Which v0.167.0 settings are still missing from the configuration reference&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Opening the agent command center with codex agents
&lt;/h2&gt;

&lt;p&gt;The first release note reads: "Open the agent command center with &lt;code&gt;codex agents --no-daemon&lt;/code&gt;. The command center and session picker now mark unread results". The screen that lists chats on a shared server now flags results you have not read.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;codex agents&lt;/code&gt; itself is documented, with maturity Stable. The command overview describes it as "Open the agent command center on a shared local or remote server", and its own section as "Open the command center to browse chats and switch between agents on a shared server".&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Flag&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;--remote &amp;lt;URL&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Connect to a remote server&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;--cd &amp;lt;DIR&amp;gt;&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Set the directory used for new remote tasks&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;--remote-auth-token-env&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A global flag, documented as requiring &lt;code&gt;--remote&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;You can reach the same screen from inside the TUI with &lt;code&gt;/agents&lt;/code&gt;, listed in the slash-command table as "Open the agent command center." with the when-to-use note "Browse chats and switch between agents on the connected server." A separate &lt;code&gt;/subagents&lt;/code&gt; command switches the active subagent thread.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;The docs say --no-daemon cannot be combined with codex agents&lt;/strong&gt;&lt;br&gt;
As of 2026-11-09 the &lt;code&gt;codex agents&lt;/code&gt; section states "&lt;code&gt;codex agents&lt;/code&gt; doesn't support &lt;code&gt;--no-daemon&lt;/code&gt;". The global &lt;code&gt;--no-daemon&lt;/code&gt; flag ("Run the interactive CLI without the shared background server.") carries the matching note "Can't be combined with &lt;code&gt;--remote&lt;/code&gt; or &lt;code&gt;codex agents&lt;/code&gt;". The v0.167.0 release notes spell the command with that flag anyway. &lt;strong&gt;Which one reflects the current implementation could not be confirmed in the official documentation.&lt;/strong&gt; Start with a plain &lt;code&gt;codex agents&lt;/code&gt; and only reach for the flag if that fails.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Automatic compaction now counts failed responses
&lt;/h2&gt;

&lt;p&gt;The widest-reaching fix is: "Automatic context compaction now accounts for retained output from failed responses and responses with missing token usage data". Retained output from failed responses, and responses whose token usage never arrived, now feed into the compaction decision.&lt;/p&gt;

&lt;p&gt;Until now those tokens went uncounted, so Codex judged compaction against a figure lower than real consumption. That is exactly the kind of drift that lets a long session run out of context without compaction ever firing.&lt;/p&gt;

&lt;p&gt;The threshold lives in &lt;code&gt;config.toml&lt;/code&gt;. These four keys are the ones documented today.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight toml"&gt;&lt;code&gt;&lt;span class="c"&gt;# Token threshold that triggers automatic history compaction&lt;/span&gt;
&lt;span class="n"&gt;model_auto_compact_token_limit&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;300000&lt;/span&gt;

&lt;span class="c"&gt;# How the threshold is counted: "total" counts the full active context,&lt;/span&gt;
&lt;span class="c"&gt;# "body_after_prefix" counts only growth after the carried compaction prefix&lt;/span&gt;
&lt;span class="n"&gt;model_auto_compact_token_limit_scope&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s"&gt;"total"&lt;/span&gt;

&lt;span class="c"&gt;# Turn-end compaction, as a percentage of the usable context window&lt;/span&gt;
&lt;span class="n"&gt;model_post_turn_compact_threshold_percent&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;85&lt;/span&gt;

&lt;span class="c"&gt;# Replace the prompt used for compaction&lt;/span&gt;
&lt;span class="n"&gt;compact_prompt&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s"&gt;"Summarize the work so far, keeping file paths and decisions."&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Key&lt;/th&gt;
&lt;th&gt;Value&lt;/th&gt;
&lt;th&gt;Description&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_auto_compact_token_limit&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;number&lt;/td&gt;
&lt;td&gt;Token threshold for automatic compaction. Unset uses model defaults&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_auto_compact_token_limit_scope&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;total&lt;/code&gt; / &lt;code&gt;body_after_prefix&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Defaults to &lt;code&gt;total&lt;/code&gt;, which counts the full active context&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_post_turn_compact_threshold_percent&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;integer 0–100&lt;/td&gt;
&lt;td&gt;Omit or set &lt;code&gt;0&lt;/code&gt; to disable turn-end compaction. The usual auto-compaction limits still apply&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;compact_prompt&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;string&lt;/td&gt;
&lt;td&gt;Inline override for the compaction prompt&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;To compact by hand, use &lt;code&gt;/compact&lt;/code&gt;, documented as "Summarize the visible chat to free tokens." with the advice "Use after long runs so Codex retains key points without blowing the context window." For the rest of the file, see &lt;a href="https://aicoding-guide.com/en/posts/codex-config-toml/" rel="noopener noreferrer"&gt;How to write Codex config.toml and its main settings keys&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Vertex AI and system proxy settings are not in the reference yet
&lt;/h2&gt;

&lt;p&gt;Two release notes touch configuration, and neither has reached the configuration reference.&lt;/p&gt;

&lt;p&gt;The first reads: "Find API key and cloud provider options under 'Sign in another way.' Enable &lt;code&gt;gcp_setup_wizard&lt;/code&gt; to set up Google Cloud Vertex AI with Application Default Credentials or a credential file". &lt;strong&gt;A key named &lt;code&gt;gcp_setup_wizard&lt;/code&gt; could not be found in the configuration reference as of 2026-11-09.&lt;/strong&gt; These are the Vertex AI keys that are documented.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Key&lt;/th&gt;
&lt;th&gt;Description&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_providers.gcp.project_id&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Google Cloud project ID for the Vertex AI endpoint. Takes precedence over &lt;code&gt;GOOGLE_CLOUD_PROJECT&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_providers.gcp.location&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Location for the Vertex AI endpoint. Takes precedence over &lt;code&gt;GOOGLE_CLOUD_LOCATION&lt;/code&gt;, defaults to &lt;code&gt;global&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_providers.gcp.credential_file&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Absolute path to an Application Default Credentials file. Overrides &lt;code&gt;GOOGLE_APPLICATION_CREDENTIALS&lt;/code&gt; only for the provider's authentication command&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;model_providers.gcp.base_url&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Override the built-in Vertex AI endpoint. Otherwise Codex builds it from the project ID and location&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The second reads: "Opt into macOS or Windows system proxy settings from &lt;code&gt;/experimental&lt;/code&gt;". The &lt;code&gt;/experimental&lt;/code&gt; command is documented ("Toggle experimental features." / "Enable options such as Network proxy or Prevent sleep while running."), and so is the matching key &lt;code&gt;features.network_proxy&lt;/code&gt; ("Start the network proxy for sandboxed commands (experimental; off by default)."). But &lt;strong&gt;reading the operating system's proxy configuration, and the handling of variables such as &lt;code&gt;HTTP_PROXY&lt;/code&gt;, could not be confirmed on those two pages.&lt;/strong&gt; Environment-variable behaviour lives on a separate page, so check that before you turn the setting on.&lt;/p&gt;

&lt;h2&gt;
  
  
  The remaining fixes
&lt;/h2&gt;

&lt;p&gt;The rest of the release is about keeping state that used to be dropped.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Saved chats stay visible after you switch model providers. If a chat cannot resume, you can still read its transcript and prepare an editable prompt to continue elsewhere&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;Ctrl+L&lt;/code&gt; clears the terminal display while earlier messages stay reachable through &lt;code&gt;Ctrl+T&lt;/code&gt;. The docs confirm only the first half, "&lt;code&gt;Ctrl+L&lt;/code&gt; only clears the terminal view and keeps the current chat." — &lt;strong&gt;no documentation for &lt;code&gt;Ctrl+T&lt;/code&gt; could be found&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;Deleting text that exposes a leading &lt;code&gt;!&lt;/code&gt; keeps an ordinary draft in chat mode&lt;/li&gt;
&lt;li&gt;Network approval prompts close when their requests end, and concurrent requests to the same endpoint prompt one at a time&lt;/li&gt;
&lt;li&gt;Long realtime V3 conversations save completed speech into chat history as the transcript grows&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;On the documentation side, the app-server README now covers managed network constraints under &lt;code&gt;[permissions.network]&lt;/code&gt; and &lt;code&gt;prompt-user&lt;/code&gt; command rules that require fresh human approval. Note that the configuration reference has no section by that literal name: it documents &lt;strong&gt;named profiles as &lt;code&gt;[permissions.&amp;lt;name&amp;gt;]&lt;/code&gt; with &lt;code&gt;network.*&lt;/code&gt; keys beneath them&lt;/strong&gt;. For how approvals and sandboxing interact, see &lt;a href="https://aicoding-guide.com/en/posts/codex-approval-sandbox/" rel="noopener noreferrer"&gt;Codex approval_policy and sandbox_mode combinations&lt;/a&gt;; for the previous release, see &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-166/" rel="noopener noreferrer"&gt;What changed in Codex CLI v0.166, codex queue --resume and codex doctor --network-only&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;The headline changes in v0.167.0 are unread markers in the agent command center and drafts that survive a backend switch&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;codex agents&lt;/code&gt; is a Stable command, but the docs deny that it accepts &lt;code&gt;--no-daemon&lt;/code&gt;, contradicting the release notes; which one is current could not be confirmed&lt;/li&gt;
&lt;li&gt;Automatic context compaction now counts retained output from failed responses and responses with missing usage data&lt;/li&gt;
&lt;li&gt;Tune the threshold with &lt;code&gt;model_auto_compact_token_limit&lt;/code&gt; and &lt;code&gt;model_auto_compact_token_limit_scope&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;gcp_setup_wizard&lt;/code&gt; and the system-proxy option do not appear in the configuration reference pages checked here&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>codex</category>
      <category>codexcli</category>
      <category>releasenotes</category>
      <category>codexagents</category>
    </item>
    <item>
      <title>This week in Claude Code, Codex and Gemini CLI (week of November 8, 2026)</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Sat, 07 Nov 2026 19:10:05 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/this-week-in-claude-code-codex-and-gemini-cli-week-of-november-8-2026-485k</link>
      <guid>https://dev.to/aicoding-guide/this-week-in-claude-code-codex-and-gemini-cli-week-of-november-8-2026-485k</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/weekly-2026-11-08/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;The change with the widest reach this week is that &lt;strong&gt;auto mode no longer falls back to the built-in classifier&lt;/strong&gt;. Where telemetry or feature flags are off, including behind an LLM gateway, calls the server cannot check are &lt;strong&gt;asked about or denied&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;For unattended runs, &lt;code&gt;permissions.denyInsteadOfPrompting&lt;/code&gt; is new: instead of stalling on a prompt, the call is denied and the turn continues.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
This week in brief&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Auto mode asks about or denies calls the server can't check, with no classifier fallback&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;permissions.denyInsteadOfPrompting&lt;/code&gt; replaces a stalled prompt with a denial&lt;/li&gt;
&lt;li&gt;Web and shell tool calls now ask before sending your account email address&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Claude Code
&lt;/h2&gt;

&lt;p&gt;Three releases: 2.1.317, 2.1.318 and 2.1.319.&lt;/p&gt;

&lt;h3&gt;
  
  
  Auto mode lost its fallback (2.1.319)
&lt;/h3&gt;

&lt;p&gt;The CHANGELOG reads: "Changed auto mode with telemetry or feature flags off, including behind LLM gateways: calls the server can't check are asked about or denied, with no fallback to the built-in classifier; set &lt;code&gt;CLAUDE_CODE_AUTO_MODE_SERVER=0&lt;/code&gt; to keep using it (Bedrock, Vertex and Foundry are unchanged)."&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Environment&lt;/th&gt;
&lt;th&gt;Effect&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Telemetry or feature flags off&lt;/td&gt;
&lt;td&gt;Calls the server can't check are asked about or denied&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Behind an LLM gateway&lt;/td&gt;
&lt;td&gt;The same&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Bedrock / Vertex / Foundry&lt;/td&gt;
&lt;td&gt;Unchanged&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;code&gt;CLAUDE_CODE_AUTO_MODE_SERVER=0&lt;/code&gt; restores the previous behaviour of using the built-in classifier. In the other direction, &lt;code&gt;CLAUDE_CODE_AUTO_MODE_SERVER=1&lt;/code&gt; turns the server-side check on behind a gateway even with &lt;code&gt;CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS&lt;/code&gt; set — and where managed settings set the latter, they must set the &lt;code&gt;1&lt;/code&gt; too.&lt;/p&gt;

&lt;p&gt;Relatedly, &lt;code&gt;/status&lt;/code&gt; no longer reports "Auto mode server: Enabled" when the server-side review is not actually reaching the session. For how the modes differ, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-default-mode/" rel="noopener noreferrer"&gt;Claude Code starts in auto mode by default&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  New settings and variables (2.1.319)
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Setting&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;permissions.denyInsteadOfPrompting&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Denies a call that would wait on a permission prompt and continues the turn, in terminal sessions, subagents and teammates&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;CLAUDE_CODE_DISABLE_FEATURE_FLAGS&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Stops feature-flag fetching without turning telemetry off; every flag uses its built-in default&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Waits for an MCP task inside its tool call instead of backgrounding it immediately&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;MCP tasks are now supported in the terminal: when a server runs a tool call as a task, Claude Code moves it to the background right away and reports the result when it finishes.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Disabling feature flags also stops idle compaction&lt;/strong&gt;&lt;br&gt;
&lt;code&gt;CLAUDE_CODE_DISABLE_FEATURE_FLAGS&lt;/code&gt; stops flag fetching while leaving telemetry on. One of idle compaction's conditions is that Claude Code can fetch feature flags from Anthropic, so setting this variable stops it running. For the full list of conditions, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-autocompact-window/" rel="noopener noreferrer"&gt;Making Claude Code compact earlier, or never&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h3&gt;
  
  
  A check before your email address goes out (2.1.317)
&lt;/h3&gt;

&lt;p&gt;New in 2.1.317: "Added a check that asks before a web or shell tool call sends your account email address to the network."&lt;/p&gt;

&lt;p&gt;Three security fixes also landed:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A &lt;code&gt;sed&lt;/code&gt; command could skip the permission check in accept-edits and auto mode (2.1.319)&lt;/li&gt;
&lt;li&gt;Bash deny and ask rules missed commands written with quoted words (2.1.319)&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;/rewind&lt;/code&gt; could restore or delete a file through a directory symlink (2.1.319)&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  Sign-in expiry is now visible (2.1.318)
&lt;/h3&gt;

&lt;p&gt;This one matters for unattended runs. The expiry date of a stored claude.ai sign-in is now available as &lt;code&gt;loginExpiresAt&lt;/code&gt; in &lt;code&gt;claude auth status&lt;/code&gt; and &lt;code&gt;credential_expires_at&lt;/code&gt; in &lt;code&gt;claude -p&lt;/code&gt; JSON results, with a one-line stderr warning starting 10 days before.&lt;/p&gt;

&lt;p&gt;The release itself is mostly fixes for configuration files and stored logins being corrupted on network home directories, so it is worth taking if you run on a shared home. Claude Code also no longer installs or updates the IDE extension when started from an IDE terminal; use &lt;code&gt;/ide&lt;/code&gt; from an outside terminal instead.&lt;/p&gt;

&lt;h2&gt;
  
  
  Codex
&lt;/h2&gt;

&lt;p&gt;v0.167.0 ships six features.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Addition&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;codex agents --no-daemon&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Opens the agent command center; it and the session picker now mark unread results&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Backend switching&lt;/td&gt;
&lt;td&gt;Switch between connected backends from the terminal session picker, keeping drafts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Image preview&lt;/td&gt;
&lt;td&gt;Preview local image attachments while composing, on supported terminals&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;gcp_setup_wizard&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Sets up Google Cloud Vertex AI with Application Default Credentials or a credential file&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;System proxy&lt;/td&gt;
&lt;td&gt;Opt into macOS or Windows system proxy settings from &lt;code&gt;/experimental&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Patch details&lt;/td&gt;
&lt;td&gt;Expand patch details in the transcript while an approval prompt is open&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;code&gt;gcp_setup_wizard&lt;/code&gt; continues from v0.165.0, which made Vertex AI a built-in provider via &lt;code&gt;model_provider = "gcp"&lt;/code&gt; — see &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-165/" rel="noopener noreferrer"&gt;Codex CLI v0.165 makes Vertex AI a built-in provider&lt;/a&gt;. The API key and cloud provider options moved under the "Sign in another way" menu.&lt;/p&gt;

&lt;p&gt;Three fixes stand out. Saved chats stay visible after switching model providers, and a chat that cannot resume can still be read and turned into an editable prompt for a new chat. Network approval prompts close when their requests end, and concurrent requests to the same endpoint prompt one at a time. Automatic context compaction now accounts for retained output from failed responses.&lt;/p&gt;

&lt;h2&gt;
  
  
  Gemini CLI
&lt;/h2&gt;

&lt;p&gt;v0.67.0 (November 4) has two features among 27 entries: the &lt;code&gt;models&lt;/code&gt; command, and the fix for a Claude hook timeout that migrated 1000x too small. See &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-67/" rel="noopener noreferrer"&gt;Gemini CLI v0.67 adds a models command and fixes the unit in Claude hook migration&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What to check now
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;If you run auto mode behind an LLM gateway, note the 2.1.319 change and set &lt;code&gt;CLAUDE_CODE_AUTO_MODE_SERVER&lt;/code&gt; if needed&lt;/li&gt;
&lt;li&gt;For unattended runs that must not stall, consider &lt;code&gt;permissions.denyInsteadOfPrompting&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Check your sign-in expiry with &lt;code&gt;claude auth status&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Move to 2.1.318 if you run on a network home directory&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Auto mode asks about or denies calls the server can't check, with no classifier fallback&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;CLAUDE_CODE_AUTO_MODE_SERVER=0&lt;/code&gt; restores the old behaviour; Bedrock, Vertex and Foundry are unchanged&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;permissions.denyInsteadOfPrompting&lt;/code&gt; turns a stalled prompt into a denial&lt;/li&gt;
&lt;li&gt;Web and shell tool calls ask before sending your account email address&lt;/li&gt;
&lt;li&gt;Codex 0.167.0 adds &lt;code&gt;gcp_setup_wizard&lt;/code&gt; and opt-in system proxy settings&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>weeklydigest</category>
      <category>updates</category>
      <category>claudecode</category>
      <category>codex</category>
    </item>
    <item>
      <title>Gemini CLI v0.67 adds a models command and fixes the unit in Claude hook migration</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Thu, 05 Nov 2026 19:17:46 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/gemini-cli-v067-adds-a-models-command-and-fixes-the-unit-in-claude-hook-migration-5b98</link>
      <guid>https://dev.to/aicoding-guide/gemini-cli-v067-adds-a-models-command-and-fixes-the-unit-in-claude-hook-migration-5b98</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-67/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Gemini CLI v0.67.0 carries 27 changes: two features and 22 fixes.&lt;/p&gt;

&lt;p&gt;One of those fixes had real consequences for anyone running both Claude Code and Gemini CLI. Migrating a Claude hook configuration made its timeout &lt;strong&gt;1000 times smaller&lt;/strong&gt;.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What this article covers&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Listing models with the &lt;code&gt;models&lt;/code&gt; command and &lt;code&gt;--list-models&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Claude's hook &lt;code&gt;timeout&lt;/code&gt; is seconds; Gemini's is milliseconds&lt;/li&gt;
&lt;li&gt;The IDE companion now works inside container sandboxes&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  A models command for listing models
&lt;/h2&gt;

&lt;p&gt;PR #29880 adds a &lt;code&gt;models&lt;/code&gt; command and a &lt;code&gt;--list-models&lt;/code&gt; flag for model discovery.&lt;/p&gt;

&lt;p&gt;Until now there was no direct way to see which models you could pass to &lt;code&gt;-m&lt;/code&gt;. This fills that gap.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Not in the documentation yet&lt;/strong&gt;&lt;br&gt;
As of 2026-11-06, neither the &lt;code&gt;models&lt;/code&gt; command nor &lt;code&gt;--list-models&lt;/code&gt; appears in the official documentation. v0.67.0 shipped on November 4, and the docs have not caught up. Check &lt;code&gt;--help&lt;/code&gt; locally for the output format and options.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Migrating a Claude hook made its timeout 1000x smaller
&lt;/h2&gt;

&lt;p&gt;This is PR #29899: "fix(cli): convert Claude hook timeout from seconds to milliseconds during migration."&lt;/p&gt;

&lt;p&gt;The cause is a unit mismatch, visible when you put both references side by side.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Tool&lt;/th&gt;
&lt;th&gt;Unit of &lt;code&gt;timeout&lt;/code&gt;
&lt;/th&gt;
&lt;th&gt;Default&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Claude Code&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Seconds&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;600 for &lt;code&gt;command&lt;/code&gt; / &lt;code&gt;http&lt;/code&gt; / &lt;code&gt;mcp_tool&lt;/code&gt;, 30 for &lt;code&gt;prompt&lt;/code&gt;, 60 for &lt;code&gt;agent&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gemini CLI&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Milliseconds&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;60000&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Claude Code's reference describes &lt;code&gt;timeout&lt;/code&gt; as "Seconds before canceling". Gemini CLI's says "Execution timeout in milliseconds (default: 60000)".&lt;/p&gt;

&lt;p&gt;So a Claude hook set to &lt;code&gt;timeout: 600&lt;/code&gt; (ten minutes), moved across without conversion, became &lt;strong&gt;600 milliseconds&lt;/strong&gt; on the Gemini side — short enough that almost any command times out immediately. If you migrated a configuration before v0.67.0, check those values.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="nl"&gt;"hooks"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"PreToolUse"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
      &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
        &lt;/span&gt;&lt;span class="nl"&gt;"hooks"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
          &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"name"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"security-check"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"type"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"$GEMINI_PROJECT_DIR/.gemini/hooks/security.sh"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
            &lt;/span&gt;&lt;span class="nl"&gt;"timeout"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="mi"&gt;5000&lt;/span&gt;&lt;span class="w"&gt;
          &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
        &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
      &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;The environment variable names differ too&lt;/strong&gt;&lt;br&gt;
The other thing easily missed in a migration is the environment. Gemini CLI passes hooks &lt;code&gt;GEMINI_PROJECT_DIR&lt;/code&gt;, &lt;code&gt;GEMINI_PLANS_DIR&lt;/code&gt;, &lt;code&gt;GEMINI_SESSION_ID&lt;/code&gt; and &lt;code&gt;GEMINI_CWD&lt;/code&gt;. A command string referencing Claude Code's &lt;code&gt;CLAUDE_PROJECT_DIR&lt;/code&gt; will not resolve when carried over as-is. The documentation also states hooks run with "a sanitized environment", so avoid relying on shell environment variables generally.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;For the Claude Code side of the contract, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-hooks-stdin-json/" rel="noopener noreferrer"&gt;Claude Code hook input and output&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Relatedly, PR #29867 preserves &lt;code&gt;${VAR}&lt;/code&gt; placeholders during deprecated settings migration. v0.66.0 fixed placeholder resolution after &lt;code&gt;.env&lt;/code&gt; files load, so this area has had attention in two consecutive releases.&lt;/p&gt;

&lt;h2&gt;
  
  
  IDE integration and containers
&lt;/h2&gt;

&lt;p&gt;The second feature is PR #29858, which bridges the IDE companion into container sandboxes over exec stdio — so IDE integration works where the sandbox runs in a container.&lt;/p&gt;

&lt;p&gt;Two IDE fixes also landed:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;IDE settings files and &lt;code&gt;GEMINI.md&lt;/code&gt; are excluded from the open-file context (PR #29848)&lt;/li&gt;
&lt;li&gt;Relative file paths resolve against the active IDE workspace root (PR #29838)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The first avoids &lt;code&gt;GEMINI.md&lt;/code&gt; entering the context twice: it is already loaded as project context, so passing it again as an open file duplicates it. For how it is loaded, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-setup-gemini-md/" rel="noopener noreferrer"&gt;Installing Gemini CLI and writing GEMINI.md&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cancellation and display fixes
&lt;/h2&gt;

&lt;p&gt;Three fixes matter for unattended runs:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Termination signals are handled in non-interactive mode, and tools no longer execute after cancellation (PR #29849)&lt;/li&gt;
&lt;li&gt;Continuation stream processing and auto-nudge stop after a request is cancelled (PR #29868)&lt;/li&gt;
&lt;li&gt;Project spending cap errors are classified as terminal quota errors (PR #29961)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The third stops pointless retries once a spending cap is reached.&lt;/p&gt;

&lt;p&gt;On display: Thai and Lao SARA AM width is corrected, and terminals that render ambiguous-width characters as wide are now handled. The latter may help where CJK text renders misaligned.&lt;/p&gt;

&lt;p&gt;Also fixed: duplicate agent loading when the workspace is the home directory, read-only files and hard links respected in atomic &lt;code&gt;writeTextFile&lt;/code&gt;, ripgrep resolution on Android (Termux), CRLF normalisation in &lt;code&gt;getDiffContextSnippet&lt;/code&gt;, and MCP server and tool metadata included in ACP permission requests. For the previous release, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-66/" rel="noopener noreferrer"&gt;Gemini CLI v0.66 is fixes only, with the keybindings.json keys array now working&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;v0.67.0 has two features and 22 fixes among 27 entries&lt;/li&gt;
&lt;li&gt;A &lt;code&gt;models&lt;/code&gt; command and &lt;code&gt;--list-models&lt;/code&gt; list available models, though neither is documented yet&lt;/li&gt;
&lt;li&gt;Migrating a Claude hook made its timeout 1000x smaller: Claude uses seconds, Gemini milliseconds&lt;/li&gt;
&lt;li&gt;Environment variables need rewriting too, from &lt;code&gt;CLAUDE_PROJECT_DIR&lt;/code&gt; to &lt;code&gt;GEMINI_PROJECT_DIR&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;The IDE companion now works inside container sandboxes&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>gemini</category>
      <category>updates</category>
      <category>hooks</category>
      <category>models</category>
    </item>
    <item>
      <title>Gemini CLI v0.66 is fixes only, with the keybindings.json keys array now working</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Tue, 03 Nov 2026 19:17:47 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/gemini-cli-v066-is-fixes-only-with-the-keybindingsjson-keys-array-now-working-213k</link>
      <guid>https://dev.to/aicoding-guide/gemini-cli-v066-is-fixes-only-with-the-keybindingsjson-keys-array-now-working-213k</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-66/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Gemini CLI v0.66.0 adds no features. All 26 entries under "What's Changed" are fixes, refactors or release chores.&lt;/p&gt;

&lt;p&gt;One of them is the kind worth knowing about, though: &lt;strong&gt;a setting the documentation already describes was not actually working&lt;/strong&gt;. The &lt;code&gt;keys&lt;/code&gt; array in &lt;code&gt;keybindings.json&lt;/code&gt; works as of this release.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What this article covers&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;How to write the &lt;code&gt;keys&lt;/code&gt; array, and what to check when a binding does nothing&lt;/li&gt;
&lt;li&gt;Settings placeholders now resolved after &lt;code&gt;.env&lt;/code&gt; files load&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;~/.gitconfig&lt;/code&gt; preserved in shell and git environments&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  The keys array in keybindings.json now works
&lt;/h2&gt;

&lt;p&gt;The fix is PR #29773. Key bindings live in your home gemini directory, typically &lt;code&gt;~/.gemini/keybindings.json&lt;/code&gt;. It is a JSON array in a schema close to VS Code's, with a &lt;code&gt;command&lt;/code&gt; and either a &lt;code&gt;key&lt;/code&gt; or a &lt;code&gt;keys&lt;/code&gt; array per entry.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"edit.clear"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"key"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"cmd+l"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"-app.toggleYolo"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"key"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"ctrl+y"&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"-input.newline"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"keys"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"alt+enter"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"ctrl+enter"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"shift+enter"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"ctrl+j"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;},&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"command"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"input.newline"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt;
    &lt;/span&gt;&lt;span class="nl"&gt;"keys"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"enter"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s2"&gt;"shift+enter"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
  &lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Form&lt;/th&gt;
&lt;th&gt;What it means&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;-&lt;/code&gt; prefixed on &lt;code&gt;command&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Unbinds the key, defaults included&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;A &lt;code&gt;keys&lt;/code&gt; array&lt;/td&gt;
&lt;td&gt;Binds or unbinds several keys in one entry; the first key is the primary shortcut shown in the UI&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Modifiers&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;ctrl&lt;/code&gt;, &lt;code&gt;shift&lt;/code&gt;, &lt;code&gt;alt&lt;/code&gt; (&lt;code&gt;opt&lt;/code&gt; / &lt;code&gt;option&lt;/code&gt;), &lt;code&gt;cmd&lt;/code&gt; (&lt;code&gt;meta&lt;/code&gt;)&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;When a binding never fires, suspect the terminal&lt;/strong&gt;&lt;br&gt;
The documentation notes that terminals often translate complex combinations into special characters, losing the modifier information on the way. Its example is &lt;code&gt;shift+5&lt;/code&gt; arriving as &lt;code&gt;%&lt;/code&gt;, where "you must bind to the literal character &lt;code&gt;%&lt;/code&gt; as bindings to &lt;code&gt;shift+5&lt;/code&gt; will never fire." To see exactly what is sent, enable &lt;code&gt;Debug Keystroke Logging&lt;/code&gt; and press &lt;code&gt;f12&lt;/code&gt; to open the debug log console.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Matching is explicit: a binding for &lt;code&gt;ctrl+f&lt;/code&gt; fires on &lt;code&gt;ctrl+f&lt;/code&gt; only, not &lt;code&gt;ctrl+shift+f&lt;/code&gt; or &lt;code&gt;alt+ctrl+f&lt;/code&gt;. And because the same key can be bound to different commands in different contexts, &lt;strong&gt;adding a binding does not unbind the key from anything else&lt;/strong&gt;. To remove the old one, add a separate &lt;code&gt;-&lt;/code&gt; prefixed entry.&lt;/p&gt;

&lt;h2&gt;
  
  
  Fixes around .env and the git environment
&lt;/h2&gt;

&lt;p&gt;Three fixes concern the order configuration is loaded in.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;PR&lt;/th&gt;
&lt;th&gt;What changed&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;#29750&lt;/td&gt;
&lt;td&gt;Settings placeholders are resolved &lt;strong&gt;after&lt;/strong&gt; &lt;code&gt;.env&lt;/code&gt; files load&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;#29832&lt;/td&gt;
&lt;td&gt;The user-level &lt;code&gt;~/.gitconfig&lt;/code&gt; is preserved in shell and git environments&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;#29760&lt;/td&gt;
&lt;td&gt;Interactive credential and SSH prompts are disabled in &lt;code&gt;getSafeGitEnv&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The first matters if your settings file references a value defined in &lt;code&gt;.env&lt;/code&gt;. Resolving before the load leaves the reference empty.&lt;/p&gt;

&lt;p&gt;A related documented setting is &lt;code&gt;advanced.ignoreLocalEnv&lt;/code&gt; (default &lt;code&gt;false&lt;/code&gt;), which controls whether generic &lt;code&gt;.env&lt;/code&gt; files in the project directory are ignored — the lever to reach for when a project's &lt;code&gt;.env&lt;/code&gt; should not be read at all.&lt;/p&gt;

&lt;p&gt;The third matters for unattended runs. When git asks for a credential or an SSH passphrase interactively, nobody is there to answer and the run hangs. Disabling those prompts makes it fail instead.&lt;/p&gt;

&lt;h2&gt;
  
  
  MCP and authentication fixes
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;MCP OAuth supports private network endpoints&lt;/strong&gt; (PR #29844), on an opt-in basis — relevant when the MCP server lives inside your network&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;False positives in &lt;code&gt;isAuthenticationError&lt;/code&gt;&lt;/strong&gt; are prevented (PR #29835), where non-authentication failures were classified as authentication errors&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Update verification tightened&lt;/strong&gt; (PR #29757): the active binary is checked after an update, so a &lt;code&gt;PATH&lt;/code&gt;-shadowed older binary is no longer reported as a successful update&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For MCP configuration generally, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-mcp/" rel="noopener noreferrer"&gt;Setting up MCP servers in Gemini CLI&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The rest, and a Gemini 4 banner
&lt;/h2&gt;

&lt;p&gt;The visible addition is a &lt;strong&gt;banner saying Gemini CLI is not optimized for Gemini 4&lt;/strong&gt; (PR #29774).&lt;/p&gt;

&lt;p&gt;On rendering: synchronized output with isolated spinner rendering, debug console height calculation and flickering, right-hand border bleed in scrollable tool containers, preserved terminal scrollback during viewport refreshes, and rounding before unit selection in &lt;code&gt;formatDuration&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Also: &lt;code&gt;openDiff&lt;/code&gt; now formats and quotes Windows diff command arguments, &lt;code&gt;SKILL.md&lt;/code&gt; frontmatter parsing handles single-line descriptions, slash-command conflict dedupe resets when conflicts reappear, and the bootstrap parent forwards termination signals to its child process.&lt;/p&gt;

&lt;p&gt;For the previous release, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-65/" rel="noopener noreferrer"&gt;Gemini CLI v0.65 enforces read-only settings in untrusted folders&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;v0.66.0 adds no features; all 26 "What's Changed" entries are fixes, refactors or release chores&lt;/li&gt;
&lt;li&gt;The &lt;code&gt;keys&lt;/code&gt; array in &lt;code&gt;keybindings.json&lt;/code&gt; works now; unbind by prefixing &lt;code&gt;command&lt;/code&gt; with &lt;code&gt;-&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;When a binding never fires, check what the terminal sends with &lt;code&gt;Debug Keystroke Logging&lt;/code&gt; and &lt;code&gt;f12&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Settings placeholders are resolved after &lt;code&gt;.env&lt;/code&gt; files load&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;~/.gitconfig&lt;/code&gt; is preserved, and git's interactive prompts are disabled&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>gemini</category>
      <category>updates</category>
      <category>keybindingsjson</category>
      <category>bugfixes</category>
    </item>
    <item>
      <title>Codex CLI v0.166 adds codex queue --resume and codex doctor --network-only</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Sun, 01 Nov 2026 19:15:17 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/codex-cli-v0166-adds-codex-queue-resume-and-codex-doctor-network-only-5ae0</link>
      <guid>https://dev.to/aicoding-guide/codex-cli-v0166-adds-codex-queue-resume-and-codex-doctor-network-only-5ae0</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-166/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Codex CLI v0.166.0 ships six features and seven fixes. Two matter day to day: &lt;strong&gt;&lt;code&gt;codex queue --resume&lt;/code&gt;&lt;/strong&gt; and &lt;strong&gt;&lt;code&gt;codex doctor --network-only&lt;/code&gt;&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Both are flags on existing commands, so knowing what each command already does is the quickest way to see what changed.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What this article covers&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;How &lt;code&gt;codex queue&lt;/code&gt; works and the situation &lt;code&gt;--resume&lt;/code&gt; fixes&lt;/li&gt;
&lt;li&gt;What &lt;code&gt;codex doctor&lt;/code&gt; checks, and what &lt;code&gt;--network-only&lt;/code&gt; narrows it to&lt;/li&gt;
&lt;li&gt;Approval settings and token accounting now surviving resume and fork&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  codex queue --resume resumes the thread
&lt;/h2&gt;

&lt;p&gt;The release notes read: "Added &lt;code&gt;codex queue --resume&lt;/code&gt; to resume unloaded threads after queueing a message."&lt;/p&gt;

&lt;p&gt;&lt;code&gt;codex queue&lt;/code&gt; itself is documented: it queues a text message for an existing session.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;codex queue &lt;span class="nt"&gt;--thread&lt;/span&gt; &amp;lt;SESSION_ID&amp;gt; &lt;span class="nt"&gt;--message&lt;/span&gt; &lt;span class="s2"&gt;"Run the remaining checks"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Option&lt;/th&gt;
&lt;th&gt;Detail&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;--thread&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A UUID or an exact session name; use a UUID when sessions share a name&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;--message&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;The text to queue&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Others&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;--remote&lt;/code&gt; and &lt;code&gt;--remote-auth-token-env&lt;/code&gt; are accepted; image attachments are not&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Previously, queueing a message to a thread that was not loaded left it waiting. With &lt;code&gt;--resume&lt;/code&gt;, queueing and resuming happen in one step — useful when a script or CI job adds work to an existing session.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;--resume is not in the commands documentation&lt;/strong&gt;&lt;br&gt;
As of 2026-11-02, the flags documented for &lt;code&gt;codex queue&lt;/code&gt; are &lt;code&gt;--thread&lt;/code&gt;, &lt;code&gt;--message&lt;/code&gt;, &lt;code&gt;--remote&lt;/code&gt; and &lt;code&gt;--remote-auth-token-env&lt;/code&gt;. There is no entry for &lt;code&gt;--resume&lt;/code&gt;; it appears only in the release notes.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  codex doctor --network-only narrows the report
&lt;/h2&gt;

&lt;p&gt;The other addition: "Added &lt;code&gt;codex doctor --network-only&lt;/code&gt; for focused configuration, connectivity, and WebSocket diagnostics."&lt;/p&gt;

&lt;p&gt;&lt;code&gt;codex doctor&lt;/code&gt; is the diagnostic command to run before filing a support issue or while investigating a broken installation. The documentation describes its scope as "installation, configuration, authentication, runtime, Git, terminal, app-server, and thread inventory health".&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# The full report&lt;/span&gt;
codex doctor

&lt;span class="c"&gt;# Narrowed to configuration, connectivity and WebSocket&lt;/span&gt;
codex doctor &lt;span class="nt"&gt;--network-only&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The full report includes the thread inventory, so it takes a while. When the symptom is clearly network-side — connections failing behind a proxy, WebSockets dropping — &lt;code&gt;--network-only&lt;/code&gt; is the faster path.&lt;/p&gt;

&lt;p&gt;This flag could not be confirmed in the documentation either: &lt;code&gt;codex doctor&lt;/code&gt;'s option table is rendered dynamically on that page, so individual flags are not readable from the Markdown.&lt;/p&gt;

&lt;p&gt;For the network proxy settings themselves, see &lt;a href="https://aicoding-guide.com/en/posts/codex-approval-sandbox/" rel="noopener noreferrer"&gt;Combining approval_policy and sandbox_mode in Codex&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Transcript handling and the other features
&lt;/h2&gt;

&lt;p&gt;The remaining four are display and interaction changes.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Addition&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Transcript selection&lt;/td&gt;
&lt;td&gt;Keyboard-driven, with word, line and page motions, integrated search, and configurable shortcuts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Markdown rendering&lt;/td&gt;
&lt;td&gt;Made consistent in async questions, choices, answers and transcript replies&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPT-6.1 Terra&lt;/td&gt;
&lt;td&gt;Added to the bundled, Vertex AI and Amazon Bedrock model catalogs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Safety-block explanations&lt;/td&gt;
&lt;td&gt;Clearer reasons in the TUI, with selectable safer follow-up prompts&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;GPT-6.1 Terra reached the bundled catalog in v0.165.0; v0.166.0 adds it to the Vertex AI and Amazon Bedrock catalogs. That follows v0.165.0 making Vertex AI a built-in provider via &lt;code&gt;model_provider = "gcp"&lt;/code&gt; — see &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-165/" rel="noopener noreferrer"&gt;Codex CLI v0.165 makes Vertex AI a built-in provider&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Fixes around resume and fork
&lt;/h2&gt;

&lt;p&gt;Two of the seven fixes concern resuming and forking sessions.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Approval and permission settings persist&lt;/strong&gt;: "Preserved acknowledged approval and permission settings when resuming or forking threads."&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reasoning-token accounting persists&lt;/strong&gt;: "Preserved reasoning-token accounting across resume and fork to avoid premature compaction."&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The second has the clearer practical effect. When the accounting reset, a resumed session could be judged to have met the compaction threshold while it actually had room left, so context was lost earlier than it needed to be.&lt;/p&gt;

&lt;p&gt;Also fixed: spawned descendant threads are restored when their parent is unarchived, app-server thread startup and shutdown races are resolved with completed runtimes released promptly, and TUI reconnection recovery now handles offline &lt;code&gt;/quit&lt;/code&gt; and &lt;code&gt;/exit&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;codex queue --resume&lt;/code&gt; queues a message and resumes an unloaded thread in one step&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;codex queue&lt;/code&gt; takes a UUID or exact session name in &lt;code&gt;--thread&lt;/code&gt; and does not support image attachments&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;codex doctor --network-only&lt;/code&gt; narrows diagnostics to configuration, connectivity and WebSocket&lt;/li&gt;
&lt;li&gt;Approval settings and reasoning-token accounting now survive resume and fork&lt;/li&gt;
&lt;li&gt;Neither &lt;code&gt;--resume&lt;/code&gt; nor &lt;code&gt;--network-only&lt;/code&gt; could be confirmed in the official commands documentation&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>codex</category>
      <category>codexcli</category>
      <category>updates</category>
      <category>codexdoctor</category>
    </item>
    <item>
      <title>This week in Claude Code, Codex and Gemini CLI (week of November 1, 2026)</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Sat, 31 Oct 2026 19:08:39 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/this-week-in-claude-code-codex-and-gemini-cli-week-of-november-1-2026-15pn</link>
      <guid>https://dev.to/aicoding-guide/this-week-in-claude-code-codex-and-gemini-cli-week-of-november-1-2026-15pn</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/weekly-2026-11-01/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;This week has a &lt;strong&gt;deadline in it&lt;/strong&gt;. Fast mode for Opus 4.8 is deprecated as of 2026-10-30 and &lt;strong&gt;removed on 2026-11-30&lt;/strong&gt;. The CHANGELOG is specific that after removal, fast mode requests on Opus 4.8 return an error and do not fall back to standard Opus 4.8.&lt;/p&gt;

&lt;p&gt;On the feature side, idle compaction spread to VS Code and the desktop app — two weeks after it arrived, terminal-only.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
This week in brief&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Fast mode on Opus 4.8 is removed on November 30; move to Opus 5.5 to keep the speedup&lt;/li&gt;
&lt;li&gt;Idle compaction was added to VS Code and the desktop app's Code tab&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;permissions.blockWritesOutsideWorkingDirectories&lt;/code&gt; was added&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Claude Code
&lt;/h2&gt;

&lt;p&gt;Four releases: 2.1.311, 2.1.312, 2.1.314 and 2.1.316.&lt;/p&gt;

&lt;h3&gt;
  
  
  A deprecation with a date (2.1.316)
&lt;/h3&gt;

&lt;p&gt;The CHANGELOG reads: "Deprecated fast mode for Opus 4.8 as of October 30, 2026; it will be removed on November 30, 2026. After removal, fast mode requests on Opus 4.8 return an error and do not fall back to standard Opus 4.8. Migrate to Opus 5.5 to keep the speedup."&lt;/p&gt;

&lt;p&gt;If a script or settings file assumes fast mode, move it to Opus 5.5 during November. After removal it does not quietly get slower — it errors.&lt;/p&gt;

&lt;h3&gt;
  
  
  Confining writes to the working directories (2.1.314)
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Setting&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;permissions.blockWritesOutsideWorkingDirectories&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Write, Edit and NotebookEdit refuse paths outside the working directories in every permission mode&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;It pairs with &lt;code&gt;permissions.blockReadsOutsideWorkingDirectories&lt;/code&gt; on the read side. With both set, the reach of the file tools no longer depends on the permission mode.&lt;/p&gt;

&lt;h3&gt;
  
  
  Idle compaction spread (2.1.314)
&lt;/h3&gt;

&lt;p&gt;"Added idle compaction to VS Code and Claude Desktop's Code tab: a 200K+ token conversation left idle is summarized, and the summary is used on your next message if the prompt cache has expired."&lt;/p&gt;

&lt;p&gt;What landed in the terminal in 2.1.299 reached VS Code and the desktop Code tab two weeks later. For the conditions and how to turn it off, see &lt;a href="https://aicoding-guide.com/en/posts/claude-code-autocompact-window/" rel="noopener noreferrer"&gt;Making Claude Code compact earlier, or never&lt;/a&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  New environment variables (2.1.316)
&lt;/h3&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Variable&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;CLAUDE_CODE_DISABLE_ULTRATHINK&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Turns the &lt;code&gt;ultrathink&lt;/code&gt; keyword off&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;CLAUDE_CODE_SUBAGENT_EFFORT&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Sets the effort subagents run at when nothing else names one, instead of inheriting the session's&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h3&gt;
  
  
  Telemetry and permissions (2.1.312, 2.1.316)
&lt;/h3&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;claude_code.safety_event stopped being sent after a week&lt;/strong&gt;&lt;br&gt;
The &lt;code&gt;claude_code.safety_event&lt;/code&gt; OpenTelemetry event added in v2.1.307 was withdrawn in v2.1.312: "Changed Claude Code to stop sending the &lt;code&gt;claude_code.safety_event&lt;/code&gt; OpenTelemetry event." We covered it in last week's digest; it lasted a week. A dashboard built on that event needs rebuilding.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;On security, &lt;code&gt;--restricted&lt;/code&gt; sessions could upload files from outside the working directories when a Read allow rule covered them; that is fixed. Separately, for Team and Enterprise logins, server-managed settings fetched from Anthropic's API now apply without an approval prompt, and a stored copy that doesn't verify is refused.&lt;/p&gt;

&lt;h3&gt;
  
  
  Easy ones to miss
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;Sessions failing turn after turn with "Invalid effort level" where an admin caps a model's effort are fixed: Claude Code lowers the effort, retries, and says which level it uses (2.1.314)&lt;/li&gt;
&lt;li&gt;A &lt;code&gt;PreCompact&lt;/code&gt; hook with very large output could make compaction fail or repeat; &lt;strong&gt;each hook's output is now cut at 10,000 characters&lt;/strong&gt; (2.1.314)&lt;/li&gt;
&lt;li&gt;Claude could overwrite a very large &lt;code&gt;CLAUDE.md&lt;/code&gt;, rules file or @-included file after seeing only its start; Write now asks Claude to read it all or edit in place (2.1.312)&lt;/li&gt;
&lt;li&gt;Standard-context sessions that stopped with "Usage credits required for 1M context" past 200K tokens now compact and continue (2.1.311)&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;SubagentStop&lt;/code&gt; hooks were firing about twice a minute per running subagent instead of once at the end (2.1.311)&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Codex
&lt;/h2&gt;

&lt;p&gt;v0.165.0 brought the Vertex AI provider and memories by default; see &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-165/" rel="noopener noreferrer"&gt;Codex CLI v0.165&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;v0.166.0 adds six features.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Addition&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;codex queue --resume&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Resumes unloaded threads after queueing a message&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;codex doctor --network-only&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Runs focused configuration, connectivity and WebSocket diagnostics&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Transcript selection&lt;/td&gt;
&lt;td&gt;Keyboard-driven, with word, line and page motions, search and configurable shortcuts&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPT-6.1 Terra&lt;/td&gt;
&lt;td&gt;Added to the bundled, Vertex AI and Amazon Bedrock model catalogs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Safety-block explanations&lt;/td&gt;
&lt;td&gt;Clearer reasons in the TUI, with selectable safer follow-up prompts&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Two fixes stand out, both about resuming: acknowledged approval and permission settings are now preserved when resuming or forking threads, and reasoning-token accounting is preserved across resume and fork to avoid premature compaction.&lt;/p&gt;

&lt;h2&gt;
  
  
  Gemini CLI
&lt;/h2&gt;

&lt;p&gt;v0.66.0 is 26 entries of fixes and chores with no features. The configuration-related ones:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Settings placeholders are now resolved after &lt;code&gt;.env&lt;/code&gt; files are loaded&lt;/li&gt;
&lt;li&gt;The user-level &lt;code&gt;~/.gitconfig&lt;/code&gt; is preserved in shell and git environments&lt;/li&gt;
&lt;li&gt;Interactive credential and SSH prompts are disabled in &lt;code&gt;getSafeGitEnv&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;MCP OAuth flows support opt-in private network endpoints&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;keybindings.json&lt;/code&gt; supports a &lt;code&gt;keys&lt;/code&gt; array&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A banner was also added saying Gemini CLI is not optimized for Gemini 4. For the previous release, see &lt;a href="https://aicoding-guide.com/en/posts/gemini-cli-update-0-65/" rel="noopener noreferrer"&gt;Gemini CLI v0.65 enforces read-only settings in untrusted folders&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What to check now
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;If you use fast mode on Opus 4.8, move to Opus 5.5 before November 30&lt;/li&gt;
&lt;li&gt;Rebuild any telemetry that relied on &lt;code&gt;claude_code.safety_event&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Note that idle compaction now runs in VS Code and the desktop app too&lt;/li&gt;
&lt;li&gt;Set &lt;code&gt;permissions.blockWritesOutsideWorkingDirectories&lt;/code&gt; if you want writes confined&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Fast mode on Opus 4.8 is removed on 2026-11-30 and will error, with no fallback&lt;/li&gt;
&lt;li&gt;Idle compaction was added to VS Code and the desktop app's Code tab&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;permissions.blockWritesOutsideWorkingDirectories&lt;/code&gt; confines writes in every permission mode&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;claude_code.safety_event&lt;/code&gt; stopped being sent a week after it was added&lt;/li&gt;
&lt;li&gt;Codex 0.166.0 adds &lt;code&gt;codex queue --resume&lt;/code&gt; and &lt;code&gt;codex doctor --network-only&lt;/code&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>weeklydigest</category>
      <category>updates</category>
      <category>claudecode</category>
      <category>codex</category>
    </item>
    <item>
      <title>Codex CLI v0.165 makes Vertex AI a built-in provider and turns memories on by default</title>
      <dc:creator>aicoding-guide</dc:creator>
      <pubDate>Thu, 29 Oct 2026 19:19:07 +0000</pubDate>
      <link>https://dev.to/aicoding-guide/codex-cli-v0165-makes-vertex-ai-a-built-in-provider-and-turns-memories-on-by-default-403k</link>
      <guid>https://dev.to/aicoding-guide/codex-cli-v0165-makes-vertex-ai-a-built-in-provider-and-turns-memories-on-by-default-403k</guid>
      <description>&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-165/" rel="noopener noreferrer"&gt;https://aicoding-guide.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Codex CLI v0.165.0 brings six features and six fixes. Two of them touch configuration: a &lt;strong&gt;built-in Vertex AI provider&lt;/strong&gt; and &lt;strong&gt;memories on by default&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;For both, the release notes are ahead of the official configuration reference. Check the behaviour on your own machine before relying on either.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Key point&lt;/strong&gt;&lt;br&gt;
What this article covers&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The Vertex AI provider configured with &lt;code&gt;model_provider = "gcp"&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Why the release notes and the configuration reference disagree about memories&lt;/li&gt;
&lt;li&gt;Changing reasoning effort partway through a conversation&lt;/li&gt;
&lt;/ul&gt;
&lt;/blockquote&gt;

&lt;h2&gt;
  
  
  Vertex AI is now a built-in provider
&lt;/h2&gt;

&lt;p&gt;The release notes read: "Added a built-in Google Cloud Vertex AI provider, configured with &lt;code&gt;model_provider = "gcp"&lt;/code&gt;." Using Vertex AI previously meant writing your own &lt;code&gt;model_providers&lt;/code&gt; entry; now it is treated as built in.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;gcp is not in the configuration reference yet&lt;/strong&gt;&lt;br&gt;
As of 2026-10-30, the reference lists only three built-in provider IDs: "Built-in provider IDs (&lt;code&gt;openai&lt;/code&gt;, &lt;code&gt;ollama&lt;/code&gt;, and &lt;code&gt;lmstudio&lt;/code&gt;) are reserved and cannot be overridden." There is no mention of &lt;code&gt;gcp&lt;/code&gt; or Vertex AI. The ID appears only in the release notes.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Where you write the setting matters. The configuration reference states that &lt;code&gt;model_provider&lt;/code&gt; and &lt;code&gt;model_providers&lt;/code&gt; are &lt;strong&gt;ignored in a project-local &lt;code&gt;.codex/config.toml&lt;/code&gt;&lt;/strong&gt;.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight toml"&gt;&lt;code&gt;&lt;span class="c"&gt;# Put this in ~/.codex/config.toml; a project-local copy is ignored&lt;/span&gt;
&lt;span class="n"&gt;model_provider&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="s"&gt;"gcp"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The ignored keys include &lt;code&gt;openai_base_url&lt;/code&gt;, &lt;code&gt;chatgpt_base_url&lt;/code&gt;, &lt;code&gt;model_provider&lt;/code&gt;, &lt;code&gt;model_providers&lt;/code&gt;, &lt;code&gt;notify&lt;/code&gt;, &lt;code&gt;profile&lt;/code&gt;, &lt;code&gt;profiles&lt;/code&gt; and &lt;code&gt;otel&lt;/code&gt;. Provider, notification and telemetry keys belong in user-level config. For defining a custom provider, the fields are &lt;code&gt;model_providers.&amp;lt;id&amp;gt;.name&lt;/code&gt;, &lt;code&gt;.base_url&lt;/code&gt;, &lt;code&gt;.env_key&lt;/code&gt; and others, covered in &lt;a href="https://aicoding-guide.com/en/posts/codex-config-toml/" rel="noopener noreferrer"&gt;Configuring Codex with config.toml&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;GPT-6.1 Terra was also added to the model catalogs.&lt;/p&gt;

&lt;h2&gt;
  
  
  Memories are on by default for fresh profiles
&lt;/h2&gt;

&lt;p&gt;The release notes read: "Enabled memory v2 by default for fresh profiles and added support for importing external-agent memories into it."&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;The two sources disagree on the default&lt;/strong&gt;&lt;br&gt;
The configuration reference still describes &lt;code&gt;features.memories&lt;/code&gt; as "Enable Memories (off by default)" (checked 2026-10-30). The term "memory v2" does not appear in the memories documentation at all. Since the documents do not settle which applies, check the actual state with &lt;code&gt;/memories&lt;/code&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;The mechanism itself is documented.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Aspect&lt;/th&gt;
&lt;th&gt;Detail&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Storage&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;~/.codex/memories/&lt;/code&gt;, holding summaries, durable entries, recent inputs and evidence&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;When they are written&lt;/td&gt;
&lt;td&gt;In the background once a chat has been idle long enough, not at the end of every chat&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;What is skipped&lt;/td&gt;
&lt;td&gt;Active and short-lived sessions; secrets are redacted from generated memory fields&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Throttling&lt;/td&gt;
&lt;td&gt;A background pass is skipped when your rate-limit remaining percentage is below the configured threshold&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Per-chat control&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;/memories&lt;/code&gt; chooses whether the chat can use existing memories and whether it feeds future ones&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The documentation is also explicit about the limits of relying on them: "Keep required team guidance in &lt;code&gt;AGENTS.md&lt;/code&gt; or checked-in documentation. Treat memories as a helpful recall layer, not as the only source for rules that must always apply."&lt;/p&gt;

&lt;p&gt;On the files themselves, it says "Treat these files as generated state" — fine to inspect when troubleshooting or before sharing your Codex home directory, but not a control surface to edit by hand.&lt;/p&gt;

&lt;h2&gt;
  
  
  Reasoning effort can change mid-conversation
&lt;/h2&gt;

&lt;p&gt;Another entry reads: "Enabled reasoning effort changes during a conversation by default for supported models." On models that support it, you can change the depth after a conversation has started.&lt;/p&gt;

&lt;p&gt;Switching through &lt;code&gt;/model&lt;/code&gt; already existed; what changed is that mid-conversation changes are enabled by default. For the values and the ways to set them, see &lt;a href="https://aicoding-guide.com/en/posts/codex-reasoning-effort/" rel="noopener noreferrer"&gt;How to change reasoning effort in Codex CLI&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The session picker also gained a repository filter, and the agent command center now groups by directory, which makes sessions easier to find across linked worktrees.&lt;/p&gt;

&lt;h2&gt;
  
  
  Security and the other fixes
&lt;/h2&gt;

&lt;p&gt;There are six fixes. The security one reads: "Closed a marketplace URL validation gap that could allow Git to connect to a different host than the one checked against allowed patterns."&lt;/p&gt;

&lt;p&gt;The remaining fixes that change behaviour:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Fresh TUI sessions now honour the server's default approval reviewer when no explicit override is given&lt;/li&gt;
&lt;li&gt;TUI preferences stay editable and immediately applicable when execution settings are invalid&lt;/li&gt;
&lt;li&gt;Queued agent messages no longer restart work after an interrupt or a switch to Plan mode&lt;/li&gt;
&lt;li&gt;When a provider reports that a model was not found, Codex stops retrying and preserves the error message&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For the previous release, see &lt;a href="https://aicoding-guide.com/en/posts/codex-update-0-164/" rel="noopener noreferrer"&gt;Codex CLI v0.164 adds /presets and inline /permissions arguments&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Summary
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;A built-in Vertex AI provider is configured with &lt;code&gt;model_provider = "gcp"&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;gcp&lt;/code&gt; could not be confirmed in the configuration reference, which still lists only &lt;code&gt;openai&lt;/code&gt;, &lt;code&gt;ollama&lt;/code&gt; and &lt;code&gt;lmstudio&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;model_provider&lt;/code&gt; is ignored in a project-local &lt;code&gt;.codex/config.toml&lt;/code&gt;; put it in user-level config&lt;/li&gt;
&lt;li&gt;Memories are said to be on by default for fresh profiles, while the reference still says off by default&lt;/li&gt;
&lt;li&gt;A marketplace URL validation gap that could redirect Git to another host was closed&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>codex</category>
      <category>codexcli</category>
      <category>updates</category>
      <category>vertexai</category>
    </item>
  </channel>
</rss>
