<?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: Roi Dayan</title>
    <description>The latest articles on DEV Community by Roi Dayan (@roidayan).</description>
    <link>https://dev.to/roidayan</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%2F4212764%2Fd09ec175-d20f-4bae-9199-fd5bb4d29a28.jpg</url>
      <title>DEV Community: Roi Dayan</title>
      <link>https://dev.to/roidayan</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/roidayan"/>
    <language>en</language>
    <item>
      <title>T3 Code, Herdr or Moontors? Three ways to run many coding agents</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Thu, 19 Nov 2026 13:14:12 +0000</pubDate>
      <link>https://dev.to/roidayan/t3-code-herdr-or-moontors-three-ways-to-run-many-coding-agents-5585</link>
      <guid>https://dev.to/roidayan/t3-code-herdr-or-moontors-three-ways-to-run-many-coding-agents-5585</guid>
      <description>&lt;p&gt;If you run more than one coding agent at a time, you've probably looked at a desktop app for it, and with Conductor &lt;a href="https://www.conductor.build/announcement" rel="noopener noreferrer"&gt;joining Meta&lt;/a&gt;, a lot of people are looking again. Three of the free ones are built three different ways:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;T3 Code&lt;/strong&gt; is one interface for the agents you've installed, with desktop, web and phone apps.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Herdr&lt;/strong&gt; is a terminal app: any program in a pane, with a status from the agents that report one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Moontors&lt;/strong&gt; drives five agents through their own protocols and shows them in its own views.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I build Moontors, so read this as the maker's view. Every fact about the other two comes from their own docs, checked on November 17, 2026, and linked.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short version
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;T3 Code&lt;/th&gt;
&lt;th&gt;Herdr&lt;/th&gt;
&lt;th&gt;Moontors&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Agents&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/faq.md" rel="noopener noreferrer"&gt;Claude Code, Codex, Cursor, Grok Build, OpenCode, Antigravity, Kiro, Pi, Muse Code, fx&lt;/a&gt;, any ACP agent&lt;/td&gt;
&lt;td&gt;&lt;a href="https://herdr.dev/" rel="noopener noreferrer"&gt;Any program in a terminal&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Claude Code, Codex, Gemini CLI, Grok Build, OpenCode&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;How you see them&lt;/td&gt;
&lt;td&gt;Its own interface&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/herdrdev/herdr-app" rel="noopener noreferrer"&gt;Each agent's terminal&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Moontors' own views&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Platforms&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/install.md" rel="noopener noreferrer"&gt;macOS, Windows, Linux&lt;/a&gt;, web, iOS, Android&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/herdrdev/herdr-app" rel="noopener noreferrer"&gt;macOS (Apple silicon), Linux, Windows&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;macOS, Windows, Linux; Android, iPhone coming soon&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Other machines&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/remote-access.md" rel="noopener noreferrer"&gt;T3 Connect, Tailscale, SSH&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/herdrdev/herdr-app" rel="noopener noreferrer"&gt;SSH&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Your own computer&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Licence&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/faq.md" rel="noopener noreferrer"&gt;MIT&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/herdrdev/herdr-app" rel="noopener noreferrer"&gt;Apache 2.0&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Closed source&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Price&lt;/td&gt;
&lt;td&gt;Free&lt;/td&gt;
&lt;td&gt;Free&lt;/td&gt;
&lt;td&gt;Free, no account needed&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  T3 Code
&lt;/h2&gt;

&lt;p&gt;T3 Code &lt;a href="https://github.com/pingdotgg/t3code" rel="noopener noreferrer"&gt;runs the coding agents on your machine and gives them one fast interface&lt;/a&gt;: the agents you've installed and signed in to, from a desktop app, a web app or its phone apps.&lt;/p&gt;

&lt;p&gt;What it does well:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;More agents&lt;/strong&gt; than Moontors: the ten &lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/faq.md" rel="noopener noreferrer"&gt;its FAQ&lt;/a&gt; names, and any ACP agent.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Agents that hand off&lt;/strong&gt;: &lt;a href="https://github.com/pingdotgg/t3code" rel="noopener noreferrer"&gt;Claude can hand a task to Codex&lt;/a&gt;, and you can change models in the middle of a thread.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pull requests in the app&lt;/strong&gt;: &lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/source-control.md" rel="noopener noreferrer"&gt;create, review and merge them&lt;/a&gt;, on GitHub, GitLab, Bitbucket and more.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Other machines&lt;/strong&gt;, through &lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/remote-access.md" rel="noopener noreferrer"&gt;T3 Connect, pairing on your network, Tailscale or SSH&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open source&lt;/strong&gt;, under MIT, and it moves fast.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What to know:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Reaching your computer through its relay needs a &lt;a href="https://github.com/pingdotgg/t3code/blob/main/docs/user/remote-access.md" rel="noopener noreferrer"&gt;T3 Connect account&lt;/a&gt;, which links up to 3 machines, more on a paid T3 Chat plan.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Herdr
&lt;/h2&gt;

&lt;p&gt;Herdr is &lt;a href="https://github.com/herdrdev/herdr-app" rel="noopener noreferrer"&gt;a desktop app for terminals and coding agents&lt;/a&gt;: it keeps your work in spaces, shows which agents need you, and leaves terminals running when you close the window.&lt;/p&gt;

&lt;p&gt;What it does well:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Any program&lt;/strong&gt;: an agent, a test watcher or a shell, side by side.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A status without vendor code&lt;/strong&gt;: agents that speak &lt;a href="https://control-codes.page/proposals/osc-7501/" rel="noopener noreferrer"&gt;OSC 7501&lt;/a&gt;, as Claude Code, Codex and Grok Build now do, show as working, blocked or done.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Other machines over SSH&lt;/strong&gt;, in the same sidebar.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Shaped by asking&lt;/strong&gt;: its layout, keys, themes and &lt;a href="https://github.com/herdrdev/herdr-app/blob/master/docs/mods.md" rel="noopener noreferrer"&gt;mods&lt;/a&gt; are plain files your agent edits for you, with &lt;a href="https://herdr.dev/gallery/" rel="noopener noreferrer"&gt;a gallery&lt;/a&gt; of other people's.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Open source&lt;/strong&gt;, under Apache 2.0.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What to know:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;It shows a status only for &lt;a href="https://github.com/herdrdev/herdr-app/blob/master/defaults/AGENTS.md" rel="noopener noreferrer"&gt;programs that report one&lt;/a&gt;, and Gemini CLI &lt;a href="https://github.com/google-gemini/gemini-cli/issues/29710" rel="noopener noreferrer"&gt;doesn't report one yet&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;After a reboot, its kept panes &lt;a href="https://github.com/herdrdev/herdr-app" rel="noopener noreferrer"&gt;get fresh shells in their last folders, not their old programs&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;Its macOS builds are for Apple silicon; on an Intel Mac you build it from source.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Moontors
&lt;/h2&gt;

&lt;p&gt;Moontors drives Claude Code, Codex, Gemini CLI, Grok Build and OpenCode through each one's own protocol, so their answers, plans and questions show in Moontors' own views rather than a terminal.&lt;/p&gt;

&lt;p&gt;What it does well:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Claude Code itself&lt;/strong&gt;, the command-line agent as Anthropic ships it, on your Claude subscription, with your own settings, skills and MCP servers; Codex on your ChatGPT sign-in, and Gemini CLI, Grok Build and OpenCode each on its own.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Agents that work together&lt;/strong&gt;: a Claude Code session can hand a task to a Codex or Gemini CLI session and hear back.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Every session a live card on one board&lt;/strong&gt;, its transcript streaming as it works, in sections you make, with one count in the header for every session waiting on you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keeps working with the app closed&lt;/strong&gt;, in a background service on macOS, Windows and Linux, and after a restart the next prompt carries the conversation on.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A phone app&lt;/strong&gt; that follows and answers your sessions: on Android now, with the iPhone app coming soon.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A Git view and an editor&lt;/strong&gt; beside the agents: history, staging by hunk, conflicts side by side.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Free, with no Moontors account.&lt;/strong&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What to know:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Five agents for now. Cursor is planned, but not there yet.&lt;/li&gt;
&lt;li&gt;It runs agents on your own computer, not on other machines.&lt;/li&gt;
&lt;li&gt;It's closed source.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Which one fits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;T3 Code&lt;/strong&gt;, if you want pull requests in the same app, agents beyond these five, a web app, or agents on other machines.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Herdr&lt;/strong&gt;, if you live in the terminal, run all kinds of programs beside your agents, or want to reshape the app by asking.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Moontors&lt;/strong&gt;, if you want Claude Code, Codex, Gemini CLI, Grok Build and OpenCode on the plans you already pay for, in a real interface, every session live on one board, and answered from your phone.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The full comparisons, with every source: &lt;a href="https://moontors.com/docs/compare/t3-code/" rel="noopener noreferrer"&gt;T3 Code and Moontors&lt;/a&gt; and &lt;a href="https://moontors.com/docs/compare/herdr/" rel="noopener noreferrer"&gt;Herdr and Moontors&lt;/a&gt;.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claudecode</category>
      <category>codex</category>
      <category>productivity</category>
    </item>
    <item>
      <title>OSC 7501: your coding agent can now tell your terminal it needs you</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Tue, 17 Nov 2026 13:13:22 +0000</pubDate>
      <link>https://dev.to/roidayan/osc-7501-your-coding-agent-can-now-tell-your-terminal-it-needs-you-29a</link>
      <guid>https://dev.to/roidayan/osc-7501-your-coding-agent-can-now-tell-your-terminal-it-needs-you-29a</guid>
      <description>&lt;p&gt;Run three coding agents in three terminal tabs, and starting them is the easy part. The hard part is knowing which one has stopped and is waiting for you. Until October, every agent had its own answer: a bell, a notification hook, a changed tab title. Now there's a shared one.&lt;/p&gt;

&lt;h2&gt;
  
  
  What OSC 7501 is
&lt;/h2&gt;

&lt;p&gt;OSC 7501 is the &lt;a href="https://control-codes.page/proposals/osc-7501/" rel="noopener noreferrer"&gt;Program Status Protocol&lt;/a&gt;, a proposal by Mitchell Hashimoto (revision 0.4, October 10, 2026). It "lets a program tell its terminal what state it is in": at rest, working, waiting on you, finished or failed, and why.&lt;/p&gt;

&lt;p&gt;A program reports its state with an escape sequence, the same family as the one that sets a window title:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;OSC 7501 ; state=working:app=make:msg=QnVpbGRpbmcgZG9jcw== ST
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;state&lt;/code&gt;&lt;/strong&gt; is &lt;code&gt;idle&lt;/code&gt;, &lt;code&gt;working&lt;/code&gt;, &lt;code&gt;blocked&lt;/code&gt;, &lt;code&gt;done&lt;/code&gt; or &lt;code&gt;error&lt;/code&gt;, or &lt;code&gt;clear&lt;/code&gt; to remove it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;kind&lt;/code&gt;&lt;/strong&gt; says what a &lt;code&gt;blocked&lt;/code&gt; program needs: &lt;code&gt;permission&lt;/code&gt;, &lt;code&gt;question&lt;/code&gt; or &lt;code&gt;auth&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;progress&lt;/code&gt;&lt;/strong&gt; is 0 to 100, with &lt;code&gt;working&lt;/code&gt; or &lt;code&gt;blocked&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;app&lt;/code&gt;&lt;/strong&gt; is a stable program name, like &lt;code&gt;cargo&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;msg&lt;/code&gt;&lt;/strong&gt; and &lt;strong&gt;&lt;code&gt;title&lt;/code&gt;&lt;/strong&gt; are base64 of UTF-8 text: a line on what it's doing or waiting for, and a short label.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A few rules make it hold up in practice:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Each report replaces the last one whole.&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;working&lt;/code&gt; and &lt;code&gt;blocked&lt;/code&gt; disappear when the process exits or the next shell prompt appears, so a crashed program doesn't leave a stale "working" behind. &lt;code&gt;done&lt;/code&gt; and &lt;code&gt;error&lt;/code&gt; stay until you've seen them.&lt;/li&gt;
&lt;li&gt;Everything in a report is untrusted: the terminal refuses control characters in the text, and never sends titles or messages back to the program.&lt;/li&gt;
&lt;li&gt;How it looks is the terminal's choice. The spec says a terminal may notify you when a status changes, and should rate-limit any notification or sound.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Which agents send it
&lt;/h2&gt;

&lt;p&gt;From each project's release notes, and the community tracker &lt;a href="https://areweosc7501yet.com/" rel="noopener noreferrer"&gt;Are we OSC 7501 yet?&lt;/a&gt; (updated November 4):&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Agent&lt;/th&gt;
&lt;th&gt;Sends OSC 7501&lt;/th&gt;
&lt;th&gt;Since&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;Yes&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md" rel="noopener noreferrer"&gt;2.1.295&lt;/a&gt;, October 8&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Codex&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://github.com/openai/codex/releases/tag/rust-v0.163.0" rel="noopener noreferrer"&gt;rust-v0.163.0&lt;/a&gt;, October 15&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Grok Build&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;1.0.52, October 11&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;OpenCode&lt;/td&gt;
&lt;td&gt;OpenCode 2 only&lt;/td&gt;
&lt;td&gt;2.2.0, October 27&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gemini CLI&lt;/td&gt;
&lt;td&gt;Not yet&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://github.com/google-gemini/gemini-cli/issues/29710" rel="noopener noreferrer"&gt;an open issue&lt;/a&gt; asks for it; a pull request closed unmerged on November 8&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Claude Code's changelog puts it plainly: "terminals that implement it can show whether Claude Code is working, waiting on you, or done."&lt;/p&gt;

&lt;h2&gt;
  
  
  Which terminals show it
&lt;/h2&gt;

&lt;p&gt;That's the other half, and it's earlier. Per the tracker and kitty's &lt;a href="https://sw.kovidgoyal.net/kitty/changelog/" rel="noopener noreferrer"&gt;changelog&lt;/a&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Shipped:&lt;/strong&gt; iTerm2 3.7.5 (October 15), kitty 0.50.0 (November 3), and a few smaller terminals.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not yet:&lt;/strong&gt; Ghostty (its library parses the reports, but the app drops them), WezTerm, Windows Terminal, Alacritty, VS Code's terminal and Warp.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Some apps built around terminals read it too. &lt;a href="https://herdr.dev/" rel="noopener noreferrer"&gt;Herdr&lt;/a&gt;, for one, lists an agent only once it reports a status this way.&lt;/p&gt;

&lt;p&gt;So on iTerm2 3.7.5 or kitty 0.50.0 or later, with a recent Claude Code or Codex, the status can already show. Elsewhere, the agent sends it and the terminal ignores it, which is harmless.&lt;/p&gt;

&lt;h2&gt;
  
  
  Send it from your own scripts
&lt;/h2&gt;

&lt;p&gt;Nothing about it is specific to agents. A long build, a test run or a deploy script can report the same way. In bash or zsh:&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;# osc_status &amp;lt;state&amp;gt; [message]&lt;/span&gt;
osc_status&lt;span class="o"&gt;(&lt;/span&gt;&lt;span class="o"&gt;)&lt;/span&gt; &lt;span class="o"&gt;{&lt;/span&gt;
  &lt;span class="kd"&gt;local&lt;/span&gt; &lt;span class="nv"&gt;msg&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;""&lt;/span&gt;
  &lt;span class="o"&gt;[&lt;/span&gt; -n &lt;span class="s2"&gt;"$2"&lt;/span&gt; &lt;span class="o"&gt;]&lt;/span&gt; &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nv"&gt;msg&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;":msg=$(printf '%s' "&lt;/span&gt;&lt;span class="nv"&gt;$2&lt;/span&gt;&lt;span class="s2"&gt;" | base64 | tr -d '\n')"&lt;/span&gt;
  &lt;span class="nb"&gt;printf&lt;/span&gt; &lt;span class="s1"&gt;'\033]7501;state=%s:app=deploy%s\033\\'&lt;/span&gt; &lt;span class="s2"&gt;"$1"&lt;/span&gt; &lt;span class="s2"&gt;"$msg"&lt;/span&gt;
&lt;span class="o"&gt;}&lt;/span&gt;

osc_status working &lt;span class="s2"&gt;"Deploying to staging"&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; ./deploy.sh staging&lt;span class="o"&gt;;&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt;
  osc_status &lt;span class="k"&gt;done&lt;/span&gt; &lt;span class="s2"&gt;"Staging is up"&lt;/span&gt;
&lt;span class="k"&gt;else&lt;/span&gt;
  osc_status error &lt;span class="s2"&gt;"Deploy failed"&lt;/span&gt;
&lt;span class="k"&gt;fi&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;\033]&lt;/code&gt; opens the sequence and &lt;code&gt;\033\&lt;/code&gt; ends it. Two more from the spec:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Waiting for input?&lt;/strong&gt; Say why with a &lt;code&gt;kind&lt;/code&gt;, so a terminal can tell a question from a password prompt: &lt;code&gt;printf '\033]7501;state=blocked:kind=question:app=deploy\033\\'&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Done with it?&lt;/strong&gt; &lt;code&gt;printf '\033]7501;state=clear\033\\'&lt;/code&gt; removes every status the program set.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  A few things to know
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Multiplexers:&lt;/strong&gt; the report has to get through them, and the spec doesn't cover tmux. &lt;a href="https://herdr.dev/" rel="noopener noreferrer"&gt;Herdr's FAQ&lt;/a&gt; notes that agents inside tmux within Herdr can't report their status.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No report, no status:&lt;/strong&gt; a program that doesn't send one shows nothing, which today includes Gemini CLI.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It's a proposal:&lt;/strong&gt; revision 0.4, so details can still change.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Doing this in Moontors
&lt;/h2&gt;

&lt;p&gt;I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, a free desktop app that runs Claude Code, Codex, Gemini CLI, Grok Build and OpenCode side by side, so take this part as the maker's view.&lt;/p&gt;

&lt;p&gt;Moontors gets each session's status from the agent's own protocol rather than from the terminal, so it knows what a session is waiting on, not only that it's waiting, Gemini CLI's sessions included:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;One count in the header&lt;/strong&gt; for every session waiting on you, for an answer to a question, a plan to approve or a permission. Click it to go to the next one. (&lt;a href="https://moontors.com/docs/window/#the-header" rel="noopener noreferrer"&gt;The header&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Each card shows why.&lt;/strong&gt; A waiting session's card shows the question or plan it's asking. (&lt;a href="https://moontors.com/docs/questions-and-plans/" rel="noopener noreferrer"&gt;Questions and plans&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Done stays blue until you've looked.&lt;/strong&gt; (&lt;a href="https://moontors.com/docs/sessions-notifications/" rel="noopener noreferrer"&gt;Done and notifications&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Notifications&lt;/strong&gt; on your computer when a session starts waiting for you, hits an error or finishes. Away from your desk, the &lt;a href="https://moontors.com/docs/phone/" rel="noopener noreferrer"&gt;phone app&lt;/a&gt;'s board shows the same waiting and done counts, and you answer from there.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Whatever you run your agents in, OSC 7501 is worth knowing: it turns "which tab was that?" into something your terminal can answer.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>terminal</category>
      <category>claudecode</category>
      <category>codex</category>
      <category>ai</category>
    </item>
    <item>
      <title>Before you click 'don't ask again': what your coding agent saves, and how to keep secrets out of it</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Sun, 15 Nov 2026 13:13:18 +0000</pubDate>
      <link>https://dev.to/roidayan/before-you-click-dont-ask-again-what-your-coding-agent-saves-and-how-to-keep-secrets-out-of-it-j2m</link>
      <guid>https://dev.to/roidayan/before-you-click-dont-ask-again-what-your-coding-agent-saves-and-how-to-keep-secrets-out-of-it-j2m</guid>
      <description>&lt;p&gt;On November 5, Apiiro's researchers &lt;a href="https://apiiro.com/blog/coding-agents-leaking-secrets-github-dont-ask-again" rel="noopener noreferrer"&gt;reported&lt;/a&gt; what they found in public GitHub repositories: 104,185 copies of Claude Code's &lt;code&gt;.claude/settings.local.json&lt;/code&gt;, and 1,122 of them held a real secret, from database passwords to AWS keys. Counting git history and other agent files too, it came to 2,539 distinct secrets across 1,994 repositories (&lt;a href="https://itwire.com/business-it-news/security/claude-code-approval-files-found-exposing-company-secrets-on-github-says-apirro-researchers" rel="noopener noreferrer"&gt;as iTWire reported it&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;How the secrets got there is simple. A saved approval is a rule, the rule is the command, and a password typed into the command is saved with it. Here's what each agent saves when you say "don't ask again", and the habits that keep secrets out of it, whichever agent you use.&lt;/p&gt;

&lt;h2&gt;
  
  
  What each agent saves
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Claude Code.&lt;/strong&gt; "Yes, and don't ask again" on a shell command saves a rule permanently, per repository and command, "to &lt;code&gt;.claude/settings.local.json&lt;/code&gt; at the root of the git repository" (&lt;a href="https://code.claude.com/docs/en/permissions" rel="noopener noreferrer"&gt;Permissions&lt;/a&gt;). A rule with no &lt;code&gt;*&lt;/code&gt; matches one exact command; for a prefix, the dialog writes a rule such as &lt;code&gt;npm test *&lt;/code&gt;. The first time Claude Code writes that file in a repository that doesn't ignore it, it adds &lt;code&gt;**/.claude/settings.local.json&lt;/code&gt; to your global git excludes (&lt;a href="https://code.claude.com/docs/en/settings" rel="noopener noreferrer"&gt;Settings&lt;/a&gt;). That protects your machine, not the repository: a teammate, another laptop or a script that commits everything doesn't have your excludes. &lt;code&gt;/permissions&lt;/code&gt; lists every rule and the file it came from.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Codex.&lt;/strong&gt; Its approval offers "Yes, and don't ask again for commands that start with" a prefix, and "Yes, and don't ask again for this command in this session" (&lt;a href="https://github.com/openai/codex/blob/main/codex-rs/tui/src/bottom_pane/approval_overlay.rs" rel="noopener noreferrer"&gt;source&lt;/a&gt;). The first writes a prefix rule to &lt;code&gt;~/.codex/rules/default.rules&lt;/code&gt;, in your home folder rather than the repository (&lt;a href="https://learn.chatgpt.com/docs/agent-configuration/rules" rel="noopener noreferrer"&gt;Rules&lt;/a&gt;). Codex calls rules experimental, and asks you to review a suggested prefix carefully before accepting it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Gemini CLI.&lt;/strong&gt; It offers "Allow once" and "Allow for this session". "Allow for all future sessions" appears only once you turn on &lt;code&gt;security.enablePermanentToolApproval&lt;/code&gt;, which is off by default (&lt;a href="https://geminicli.com/docs/reference/configuration/" rel="noopener noreferrer"&gt;Configuration&lt;/a&gt;). The saved policy goes to your user policies folder, and for a shell command it's keyed on the program, so approving one &lt;code&gt;npm&lt;/code&gt; command can let later ones through too (&lt;a href="https://github.com/google-gemini/gemini-cli/issues/30195" rel="noopener noreferrer"&gt;issue #30195&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;The pattern: a session approval ends with the session, and a saved one is a file that outlives the task. Know where yours live.&lt;/p&gt;

&lt;h2&gt;
  
  
  Seven habits that keep secrets out
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Never put a secret in a command an agent runs.&lt;/strong&gt; Not &lt;code&gt;PGPASSWORD=… psql&lt;/code&gt;, not &lt;code&gt;curl -H "Authorization: Bearer …"&lt;/code&gt;. An inline variable still puts the value in the command text, which ends up in the saved rule, the transcript and your shell's process list (&lt;a href="https://cwe.mitre.org/data/definitions/214.html" rel="noopener noreferrer"&gt;CWE-214&lt;/a&gt;). Let the program read it from a secret manager or a config file instead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deny reads of your secret files.&lt;/strong&gt; In Claude Code, deny rules hold in every permission mode: &lt;code&gt;"deny": ["Read(./.env)", "Read(./.env.local)", "Read(./secrets/**)"]&lt;/code&gt; (&lt;a href="https://code.claude.com/docs/en/best-practices" rel="noopener noreferrer"&gt;Best practices&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ignore local settings in the repository itself.&lt;/strong&gt; Add &lt;code&gt;.claude/settings.local.json&lt;/code&gt; to the project's &lt;code&gt;.gitignore&lt;/code&gt;, so the rule travels with the repository. If it's already tracked, &lt;code&gt;git rm --cached&lt;/code&gt; it first (&lt;a href="https://docs.github.com/en/get-started/git-basics/ignoring-files" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prefer session approvals&lt;/strong&gt; while you're unsure. You can always save the rule later.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Review saved rules now and then:&lt;/strong&gt; &lt;code&gt;/permissions&lt;/code&gt; in Claude Code, &lt;code&gt;~/.codex/rules/default.rules&lt;/code&gt; for Codex, and Gemini CLI's policies folder.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Turn on push protection for the repository.&lt;/strong&gt; It blocks a push that carries a detected secret, and it's off by default at the repository level (&lt;a href="https://docs.github.com/en/code-security/secret-scanning/introduction/about-push-protection" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;). Don't rely on scanners alone: in Apiiro's set, Gitleaks and TruffleHog both missed 215 of the files that held a secret.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If a secret leaked, rotate it first.&lt;/strong&gt; Rewriting history and force-pushing doesn't reach clones, forks or cached views (&lt;a href="https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/removing-sensitive-data-from-a-repository" rel="noopener noreferrer"&gt;GitHub&lt;/a&gt;).&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  The other direction: instruction files you didn't write
&lt;/h2&gt;

&lt;p&gt;Agents read &lt;code&gt;AGENTS.md&lt;/code&gt; and &lt;code&gt;CLAUDE.md&lt;/code&gt; as instructions, so a file you didn't write can steer one:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Backslash Security showed Codex following attacker instructions in a project's &lt;code&gt;AGENTS.md&lt;/code&gt;, copying &lt;code&gt;~/.aws/credentials&lt;/code&gt;; that payload has since been patched (&lt;a href="https://www.backslash.security/blog/openai-codex-injection-in-agents-md-exfiltrating-credentials" rel="noopener noreferrer"&gt;Backslash&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;Forever Security found an &lt;code&gt;AGENTS.md&lt;/code&gt; planted in a user's Downloads folder that made the agent rewrite its own configuration, and reproduced it on Codex and Claude Code (&lt;a href="https://forever.security/blog/hijacking-ai-agents-via-downloads-folder" rel="noopener noreferrer"&gt;Forever Security&lt;/a&gt;).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What helps: read the instruction files and the project's agent config in a repository you didn't write before you start an agent in it, and use your agents' folder trust. Codex loads a project's &lt;code&gt;.codex/&lt;/code&gt; layer only when you trust the project (&lt;a href="https://learn.chatgpt.com/docs/config-file/config-basic" rel="noopener noreferrer"&gt;Config&lt;/a&gt;), Claude Code holds a tracked settings file's rules until you trust the folder, and Gemini CLI skips an untrusted folder's settings, &lt;code&gt;.env&lt;/code&gt; files and MCP servers (&lt;a href="https://geminicli.com/docs/cli/trusted-folders/" rel="noopener noreferrer"&gt;Trusted folders&lt;/a&gt;). And don't start an agent with no approvals in a repository you've just cloned.&lt;/p&gt;

&lt;h2&gt;
  
  
  Doing this in Moontors
&lt;/h2&gt;

&lt;p&gt;I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, a free desktop app that runs Claude Code, Codex, Gemini CLI and Grok Build side by side, so take this part as the maker's view.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Every ask shows exactly what will run&lt;/strong&gt;, such as "Run this command?" with the command in a code font, and &lt;strong&gt;Allow&lt;/strong&gt; or &lt;strong&gt;Deny&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Allow for this session&lt;/strong&gt; adds a rule for that session only. The session's rules are listed in Configuration, under This session, where you can remove one. (&lt;a href="https://moontors.com/docs/permission-modes/#allow-for-this-session" rel="noopener noreferrer"&gt;Allow for this session&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A permission mode per session&lt;/strong&gt;, from Manual, which asks before each edit and command, to Bypass.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Trust this folder&lt;/strong&gt; comes last in Create session and Add project, a switch that's on by default, and its ⓘ names some of what the folder's config runs. Turn it off, and the session starts untrusted: do that for a repository you've just cloned. Moontors never writes your agents' own trust files. (&lt;a href="https://moontors.com/docs/configuration/#a-folders-trust" rel="noopener noreferrer"&gt;A folder's trust&lt;/a&gt;)&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Whatever you use, the rule of thumb is the same: approve for the session, keep secrets out of commands, and treat a saved approval as a file you're responsible for.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>security</category>
      <category>ai</category>
      <category>claudecode</category>
      <category>codex</category>
    </item>
    <item>
      <title>Claude Code hit its usage limit? How the limits work, and how to keep working</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Fri, 13 Nov 2026 13:13:15 +0000</pubDate>
      <link>https://dev.to/roidayan/claude-code-hit-its-usage-limit-how-the-limits-work-and-how-to-keep-working-3hdb</link>
      <guid>https://dev.to/roidayan/claude-code-hit-its-usage-limit-how-the-limits-work-and-how-to-keep-working-3hdb</guid>
      <description>&lt;p&gt;"You've hit your session limit" is the line nobody wants mid-task. Here's how Claude Code's limits actually work on a Claude plan, how to see one coming, what your options are when you hit it, and how to make your usage last longer. Every fact is from Anthropic's own pages, checked on November 8, 2026.&lt;/p&gt;

&lt;h2&gt;
  
  
  How the limits work
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Two windows on Pro and Max.&lt;/strong&gt; "Pro and Max plans have a five-hour session limit and a weekly limit. Ultra plans have a weekly limit, and verified Ultra accounts don't have a five-hour session limit." (&lt;a href="https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan" rel="noopener noreferrer"&gt;Anthropic&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The five-hour window starts with your first message.&lt;/strong&gt; "The period starts when you send your first message, and your limit resets in full five hours later." (&lt;a href="https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work" rel="noopener noreferrer"&gt;Anthropic&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The weekly limit resets at a fixed time&lt;/strong&gt; assigned to your account, whenever you started using Claude. It applies across all models, and some plans also have a separate weekly limit for Fable.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Chat and Claude Code share one pool.&lt;/strong&gt; "All activity in both tools counts against the same limits." A long chat on claude.ai in the morning is less Claude Code in the afternoon.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The plans:&lt;/strong&gt; Pro, Max at five times Pro's usage, and Ultra at ten times Pro's weekly usage. Ultra is what was called Max 20x. (&lt;a href="https://support.claude.com/en/articles/11049741-what-is-the-max-plan" rel="noopener noreferrer"&gt;Max&lt;/a&gt;, &lt;a href="https://support.claude.com/en/articles/17728952-what-is-the-ultra-plan" rel="noopener noreferrer"&gt;Ultra&lt;/a&gt;)&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  See it coming
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;/usage&lt;/code&gt; in Claude Code&lt;/strong&gt; shows your plan's usage limits, with session cost and activity stats. (&lt;a href="https://code.claude.com/docs/en/commands" rel="noopener noreferrer"&gt;Commands&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;On the web,&lt;/strong&gt; Settings &amp;gt; Usage shows your current session and when your weekly limits reset.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A warning before the wall:&lt;/strong&gt; Claude Code can tell you when you've used most of a limit, with a line such as &lt;code&gt;You've used 85% of your session limit · resets 3:45pm&lt;/code&gt;. (&lt;a href="https://code.claude.com/docs/en/errors" rel="noopener noreferrer"&gt;Errors&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your status line&lt;/strong&gt; can show both windows: for subscribers, its &lt;code&gt;rate_limits&lt;/code&gt; object carries a five-hour and a seven-day window. (&lt;a href="https://code.claude.com/docs/en/statusline" rel="noopener noreferrer"&gt;Status line&lt;/a&gt;)&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  When you hit it
&lt;/h2&gt;

&lt;p&gt;At the limit, Claude Code says which one, such as &lt;code&gt;You've hit your session limit · resets 3:45pm&lt;/code&gt;, and blocks further requests until that time. Your options:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Wait, and let it continue on its own.&lt;/strong&gt; With a claude.ai subscription, automatic continue is on by default: Claude Code shows &lt;code&gt;Continuing automatically at 3:45pm · Esc to cancel&lt;/code&gt; and picks up at the reset. Turn it off in &lt;code&gt;/config&lt;/code&gt; with &lt;strong&gt;Continue automatically at usage limit&lt;/strong&gt;. (&lt;a href="https://code.claude.com/docs/en/interactive-mode" rel="noopener noreferrer"&gt;Interactive mode&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;See every option in one place:&lt;/strong&gt; &lt;code&gt;/rate-limit-options&lt;/code&gt; lists waiting and continuing at the reset, adding usage credits, and upgrading your plan.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Turn on usage credits&lt;/strong&gt; in Settings &amp;gt; Usage, billed at standard API rates, with a monthly spend limit you set. In Claude Code, &lt;code&gt;/usage-credits&lt;/code&gt; does it (it used to be &lt;code&gt;/extra-usage&lt;/code&gt;). (&lt;a href="https://support.claude.com/en/articles/12429409-manage-usage-credits-for-paid-claude-plans" rel="noopener noreferrer"&gt;Usage credits&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Use an API key from the Claude Console&lt;/strong&gt; for a heavy sprint, billed at API rates. Mind the reverse too: if &lt;code&gt;ANTHROPIC_API_KEY&lt;/code&gt; is set in your environment, Claude Code uses it instead of your subscription. Check with &lt;code&gt;/status&lt;/code&gt;. (&lt;a href="https://code.claude.com/docs/en/authentication" rel="noopener noreferrer"&gt;Authentication&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Switching models won't help with the session or weekly limit,&lt;/strong&gt; which are shared across all models. It does help with a model's own limit, such as Fable's weekly one.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Make it last longer
&lt;/h2&gt;

&lt;p&gt;From Anthropic's own &lt;a href="https://code.claude.com/docs/en/costs" rel="noopener noreferrer"&gt;cost guide&lt;/a&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;/clear&lt;/code&gt; between unrelated tasks.&lt;/strong&gt; "Stale context wastes tokens on every subsequent message." It costs nothing.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;/compact&lt;/code&gt; with a focus,&lt;/strong&gt; such as &lt;code&gt;/compact Focus on the API changes&lt;/code&gt;. It reads the whole conversation to summarize it, so it's a big request itself: use &lt;code&gt;/clear&lt;/code&gt; when you don't need the history.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pick the model for the job.&lt;/strong&gt; The default on Pro, Max and Ultra is Opus. "Sonnet handles most coding tasks well and costs less than Opus." &lt;code&gt;/model opusplan&lt;/code&gt; plans with Opus and builds with Sonnet.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep CLAUDE.md under 200 lines.&lt;/strong&gt; It's read into every session; move workflow-specific instructions into skills.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Subagents count too.&lt;/strong&gt; Their requests draw on the same limits. Give simple ones &lt;code&gt;model: haiku&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Lower the effort&lt;/strong&gt; with &lt;code&gt;/effort&lt;/code&gt; for routine work, and &lt;strong&gt;turn off MCP servers&lt;/strong&gt; you aren't using (&lt;code&gt;/mcp&lt;/code&gt;). &lt;code&gt;/context&lt;/code&gt; shows what's filling the context.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Keep the work going with another agent
&lt;/h2&gt;

&lt;p&gt;The limits are per plan, so a second agent on a different plan keeps the work moving. Codex on a ChatGPT plan has its own limits, with a five-hour window on Plus, and Pro has none for now. Its pricing page says &lt;code&gt;/status&lt;/code&gt; shows what's left during a session, and there's a usage page at chatgpt.com/codex/settings/usage. (&lt;a href="https://developers.openai.com/codex/pricing" rel="noopener noreferrer"&gt;Codex pricing&lt;/a&gt;)&lt;/p&gt;

&lt;p&gt;Handing work across is easy if the work is already set up for it: a short &lt;code&gt;AGENTS.md&lt;/code&gt; that both agents read, one task per branch, and a handback that names the commit, so the next agent starts from exactly where the last one stopped.&lt;/p&gt;

&lt;h2&gt;
  
  
  Doing this in Moontors
&lt;/h2&gt;

&lt;p&gt;I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, a free desktop app that runs Claude Code, Codex, Gemini CLI and Grok Build side by side, so take this part as the maker's view.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Your limits where you can see them.&lt;/strong&gt; With a Claude subscription, a ChatGPT sign-in for Codex or a paid grok.com plan for Grok Build, Moontors shows each limit's window, how much is used and when it resets. From 75% a ring appears in the header, amber, then red from 90%. (&lt;a href="https://moontors.com/docs/insight-and-code-workflows/#context-usage-and-limits" rel="noopener noreferrer"&gt;Context, usage and limits&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude Code and Codex on your own plans,&lt;/strong&gt; Claude Code on your Claude subscription and Codex on your ChatGPT sign-in.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Agents that hand work to each other.&lt;/strong&gt; A Claude Code session can &lt;a href="https://moontors.com/docs/sessions-messaging/" rel="noopener noreferrer"&gt;hand a task to a Codex session&lt;/a&gt; and hear back, so a limit on one plan doesn't stop the work.&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>ai</category>
      <category>productivity</category>
      <category>codex</category>
    </item>
    <item>
      <title>Parallel coding agents, one human: how to stop being the bottleneck</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Wed, 11 Nov 2026 14:32:20 +0000</pubDate>
      <link>https://dev.to/roidayan/parallel-coding-agents-one-human-how-to-stop-being-the-bottleneck-805</link>
      <guid>https://dev.to/roidayan/parallel-coding-agents-one-human-how-to-stop-being-the-bottleneck-805</guid>
      <description>&lt;p&gt;Starting four coding agents is the easy part. Then the first one asks to run the tests, the second has a question about the cache, the third says it's done, and the fourth has been waiting on you for ten minutes in a tab you haven't looked at. Run agents in parallel without a plan for your own attention, and it mostly means waiting in parallel.&lt;/p&gt;

&lt;p&gt;You are the one part of the setup that doesn't scale, so the work is in asking less of yourself. Here's what does that, in plain tools first, then the same thing in a desktop app.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the waiting comes from
&lt;/h2&gt;

&lt;p&gt;Four things, roughly in order of how much time they cost:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Routine permission prompts.&lt;/strong&gt; The agent asks to run the test suite for the twentieth time.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Questions the brief should have answered.&lt;/strong&gt; Which cache? Which branch? Keep the old API?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;"Done" you can't trust.&lt;/strong&gt; It says it's finished, and you have to read everything to know.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not knowing who's waiting.&lt;/strong&gt; An agent stopped for you, and nothing told you.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  1. Answer the routine asks once
&lt;/h2&gt;

&lt;p&gt;Every agent CLI lets you decide what it may run without asking: allow rules, an approval mode, or a permission mode per session. Use them for the commands you'd approve every time anyway, like the test runner, the linter and the type checker, and keep everything else asking.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Allow the boring, repeatable commands:&lt;/strong&gt; &lt;code&gt;npm test&lt;/code&gt;, &lt;code&gt;yarn lint&lt;/code&gt;, &lt;code&gt;cargo check&lt;/code&gt;, a formatter.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Keep asking for anything you can't undo:&lt;/strong&gt; pushing, deploying, deleting, publishing, anything that reads a secret or calls a paid API.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prefer a rule for this session over a saved one&lt;/strong&gt; while you're unsure. A rule saved in the project outlives the task, and you'll forget it's there.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Each allowed command is one fewer interruption, and across four agents in a day that adds up to hours.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Write the brief so the agent doesn't have to ask
&lt;/h2&gt;

&lt;p&gt;Most questions an agent asks are decisions the brief left open. A few lines up front save a round trip later:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Goal: rate-limit POST /quotes to 60 requests a minute per customer.
Files: src/quotes/routes.ts, a new src/quotes/limit.ts, their tests.
Decided: in-memory token bucket, keyed by the x-customer-key header. No Redis.
Done means: yarn test src/quotes passes and yarn typecheck is clean.
If something is unclear: pick the simpler option, say which in your summary, and go on.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That last line matters most. "Pick the simpler option and tell me" turns a blocking question into a note you read later.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Make "done" something you can check
&lt;/h2&gt;

&lt;p&gt;"Done" from an agent is a claim. Make it a claim you can check in a minute:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Put the check in the brief.&lt;/strong&gt; Name the command that must pass, and ask the agent to run it before it says it's finished.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ask for a short handback:&lt;/strong&gt; the branch, its head commit, &lt;code&gt;git diff --stat&lt;/code&gt; against the base, what it changed, and the commands it ran with their results.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Read the stat before the diff.&lt;/strong&gt; Twelve files changed when you expected two tells you where to look before you read a line.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;A handback that points at a commit is reproducible. You review exactly what the agent checked, not whatever is in its folder at that moment.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. One queue for everything waiting on you
&lt;/h2&gt;

&lt;p&gt;Don't react to each agent as it pings. Work through them in rounds, in the order that unblocks the most work:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Blocked agents first.&lt;/strong&gt; A question or a permission prompt is usually seconds to answer, and the agent is idle until you do.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Finished ones next.&lt;/strong&gt; Check the handback, then review, merge or send it back.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Running ones last.&lt;/strong&gt; Leave them alone unless something looks wrong.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;For that you need one place that shows which agents are waiting and why, and a notification when one stops for you. With terminals, that means a notification hook per agent and a habit of checking every tab. Without it, an agent that stopped for you can sit idle all afternoon.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Keep each agent out of the others' way
&lt;/h2&gt;

&lt;p&gt;Interruptions also come from agents colliding: two of them editing the same file, or one's test run picking up the other's half-finished change. Give each task its own git worktree and branch, and those collisions stop, along with the questions about them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Doing this in Moontors
&lt;/h2&gt;

&lt;p&gt;I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, a free desktop app for running Claude Code, Codex, Gemini CLI and Grok Build side by side, so take this part as the maker's view. It's built around that queue:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;One count in the header&lt;/strong&gt; for every session waiting on you, for an answer to a question, a plan to approve or a permission. Click it to go to the next one. (&lt;a href="https://moontors.com/docs/window/#the-header" rel="noopener noreferrer"&gt;The header&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Each card shows why.&lt;/strong&gt; A waiting session's card shows the question or plan it's asking, so you can see what it needs before you open it. (&lt;a href="https://moontors.com/docs/questions-and-plans/" rel="noopener noreferrer"&gt;Questions and plans&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Done stays blue until you've looked&lt;/strong&gt;, and a pill in the header counts the finished sessions you haven't seen yet. Click it for the next one. (&lt;a href="https://moontors.com/docs/sessions-notifications/" rel="noopener noreferrer"&gt;Done and notifications&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Notifications&lt;/strong&gt; on your computer when a session starts waiting for you, hits an error or finishes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A permission mode per session&lt;/strong&gt;, from Manual to Bypass, and &lt;strong&gt;Allow for this session&lt;/strong&gt;, which adds a rule for that session only. You can remove it in Configuration. (&lt;a href="https://moontors.com/docs/permission-modes/" rel="noopener noreferrer"&gt;Permission modes&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Answer from your phone.&lt;/strong&gt; The &lt;a href="https://moontors.com/docs/phone/" rel="noopener noreferrer"&gt;phone app&lt;/a&gt; pairs with your computer to follow your sessions and answer their questions and permissions.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Whatever you run them in, the rule is the same: let the agents work in parallel, and keep yourself to one queue.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claudecode</category>
      <category>codex</category>
      <category>productivity</category>
    </item>
    <item>
      <title>Never coded? Build your first app with Claude Code this afternoon (no terminal, no jargon)</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Tue, 10 Nov 2026 13:15:37 +0000</pubDate>
      <link>https://dev.to/roidayan/never-coded-build-your-first-app-with-claude-code-this-afternoon-no-terminal-no-jargon-2daj</link>
      <guid>https://dev.to/roidayan/never-coded-build-your-first-app-with-claude-code-this-afternoon-no-terminal-no-jargon-2daj</guid>
      <description>&lt;p&gt;If you've never written code, and the idea of starting makes you a little nervous, this guide is for you. You don't need to know anything about programming, and you don't need to be "a computer person". If you can write a text message, you can do this.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Vibe coding&lt;/strong&gt; means you describe what you want in plain words, and an AI writes the code, runs it and fixes it. You never have to read a line of code. It's like texting a very patient friend who happens to know everything about computers.&lt;/p&gt;

&lt;p&gt;By the end of this guide you'll have:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;an AI helper that answers questions about your computer and makes files for you;&lt;/li&gt;
&lt;li&gt;your first small app, which you made just by describing it.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;What you need:&lt;/strong&gt; a computer (Mac, Windows or Linux), an internet connection, about an hour, and $20 for the first month.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, the free app this guide uses. It gives Claude Code, Anthropic's AI coding helper, a friendly window, so there's no black screen and no commands to type.&lt;/p&gt;

&lt;h2&gt;
  
  
  First: you can't break anything
&lt;/h2&gt;

&lt;p&gt;This is the part people worry about most, so let's get it out of the way.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Nothing happens without your click.&lt;/strong&gt; Before Claude runs anything on your computer or changes a file, it tells you what it wants to do and waits for you to click &lt;strong&gt;Allow&lt;/strong&gt;. If you're not sure, click &lt;strong&gt;Deny&lt;/strong&gt;, and nothing happens.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You choose where it works.&lt;/strong&gt; Claude works inside a folder you pick. Everything you practise with in this guide goes into folders you make yourself, and you can delete them whenever you like.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your apps stay on your computer.&lt;/strong&gt; The apps you build here are files on your own computer. Nobody sees them unless you share them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No question is too basic.&lt;/strong&gt; Didn't understand an answer? Type "Explain that like I'm 10." Claude doesn't judge, and it doesn't get tired of questions.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Words you'll see
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Prompt:&lt;/strong&gt; the message you type to Claude.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Session:&lt;/strong&gt; one conversation with Claude, working in one folder.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Project:&lt;/strong&gt; a shortcut to a folder you use often.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Command:&lt;/strong&gt; an instruction Claude gives your computer, like "show me the free space". You see each one before it runs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Web page:&lt;/strong&gt; a file your browser opens. The apps you'll build here are web pages that live on your computer, not on the internet.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That's all the jargon in this guide.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: Get a Claude account
&lt;/h2&gt;

&lt;p&gt;Go to &lt;a href="https://claude.ai" rel="noopener noreferrer"&gt;claude.ai&lt;/a&gt; and sign up with your Google account or your email. That's your Claude account, and you'll use it to sign in later.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: Pick a plan
&lt;/h2&gt;

&lt;p&gt;The free plan doesn't include Claude Code, so you need a paid one. It's a monthly subscription, like a streaming service: you can cancel any time. Here are the plans on &lt;a href="https://claude.com/pricing" rel="noopener noreferrer"&gt;Anthropic's pricing page&lt;/a&gt; today:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Plan&lt;/th&gt;
&lt;th&gt;Price&lt;/th&gt;
&lt;th&gt;Pick it if you want to…&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Pro&lt;/td&gt;
&lt;td&gt;$20 a month ($17 a month if you pay yearly)&lt;/td&gt;
&lt;td&gt;try it out, play, and build small things for yourself&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Max&lt;/td&gt;
&lt;td&gt;$100 a month, 5× Pro's usage&lt;/td&gt;
&lt;td&gt;build something bigger most days of the week&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Ultra&lt;/td&gt;
&lt;td&gt;$200 a month, 10× Pro's usage&lt;/td&gt;
&lt;td&gt;build all day, every day, with no 5-hour limit&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;&lt;strong&gt;Start with Pro.&lt;/strong&gt; It's plenty for learning.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;No surprise bills.&lt;/strong&gt; Pro and Max have a usage limit that refills every 5 hours, and a weekly one. If you reach it, Claude simply pauses until it refills; nothing is broken, it's a good moment for a break. You'd only pay more if you turn on &lt;strong&gt;Extra usage&lt;/strong&gt; yourself, in claude.ai's &lt;strong&gt;Settings&lt;/strong&gt;, under &lt;strong&gt;Usage&lt;/strong&gt;, so leave it off.&lt;/p&gt;

&lt;p&gt;Choose your plan on the &lt;a href="https://claude.com/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;, or with &lt;strong&gt;Upgrade&lt;/strong&gt; inside claude.ai.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: Install Moontors
&lt;/h2&gt;

&lt;p&gt;Download Moontors from &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;moontors.com&lt;/a&gt;. It's free, and it doesn't ask you to make another account.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Mac:&lt;/strong&gt; open the file you downloaded (it ends in &lt;code&gt;.dmg&lt;/code&gt;) and drag Moontors into Applications. Then open it from Applications.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Windows:&lt;/strong&gt; open the installer you downloaded.

&lt;ul&gt;
&lt;li&gt;Windows may show a blue box saying "Windows protected your PC". Don't panic: Windows shows it for apps it hasn't seen many times yet. Click &lt;strong&gt;More info&lt;/strong&gt;, then &lt;strong&gt;Run anyway&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;If it says Smart App Control blocked the app instead, there's no Run anyway button. Moontors' &lt;a href="https://moontors.com/docs/first-launch/" rel="noopener noreferrer"&gt;first launch guide&lt;/a&gt; explains that Windows setting.&lt;/li&gt;
&lt;/ul&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Linux:&lt;/strong&gt; the &lt;a href="https://moontors.com/docs/first-launch/" rel="noopener noreferrer"&gt;first launch guide&lt;/a&gt; has the two commands.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 4: Make a folder, sign in and start your first session
&lt;/h2&gt;

&lt;p&gt;Claude works inside a folder, so your first session gets a folder of its own for everyday questions about your computer.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;In Moontors, click &lt;strong&gt;Create session&lt;/strong&gt; at the top left.&lt;/li&gt;
&lt;li&gt;Under &lt;strong&gt;Project&lt;/strong&gt;, click &lt;strong&gt;Choose folder…&lt;/strong&gt;. In the window that opens, go to your &lt;strong&gt;Desktop&lt;/strong&gt;, click &lt;strong&gt;New Folder&lt;/strong&gt;, name it &lt;code&gt;My PC helper&lt;/code&gt;, and select it.&lt;/li&gt;
&lt;li&gt;Under &lt;strong&gt;Agent&lt;/strong&gt;, pick &lt;strong&gt;Claude Code&lt;/strong&gt;. The first time, Moontors downloads Claude Code for you and shows how far along it is. There's nothing for you to install.&lt;/li&gt;
&lt;li&gt;Leave everything else as it is, and click &lt;strong&gt;Create&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Moontors asks you to sign in. Choose &lt;strong&gt;Sign in with Claude subscription&lt;/strong&gt;. Your browser opens on Claude's page: sign in with the account from Step 1, and approve. Back in Moontors, your session opens.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If your browser doesn't open, the sign-in window shows a web address and a box for a code. Open the address in your browser, sign in, copy the code it shows you, paste it into Moontors, and click &lt;strong&gt;Continue&lt;/strong&gt;.&lt;/p&gt;

&lt;h3&gt;
  
  
  Allow and Deny
&lt;/h3&gt;

&lt;p&gt;When Claude wants to run something on your computer or change a file, a box asks for your OK and tells you what it wants to do. You click &lt;strong&gt;Allow&lt;/strong&gt; or &lt;strong&gt;Deny&lt;/strong&gt;.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Not sure what it means? Click &lt;strong&gt;Deny&lt;/strong&gt;, and ask Claude: "What did you want to do, and why?" It explains in plain words, and you can ask it again.&lt;/li&gt;
&lt;li&gt;When it asks about the same thing again and again, &lt;strong&gt;Allow for this session&lt;/strong&gt; lets that one thing through for the rest of this conversation.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 5: Five first prompts
&lt;/h2&gt;

&lt;p&gt;Type each one into the box at the bottom of the session and press Enter. There's no wrong way to ask: write the way you'd text a friend, typos and all. Each one takes less than a minute.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;How full is your computer?&lt;/strong&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;   How much free space is left on this computer?
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;What computer is this, in plain words?&lt;/strong&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;   What computer is this? Tell me its model, processor, memory and operating system version, and explain what each one means in simple words.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;What's filling your Downloads?&lt;/strong&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;   Which 10 files in my Downloads folder take up the most space? Show their sizes. Don't delete anything.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Write something and save it:&lt;/strong&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;   Write a 200-word bedtime story about a robot who's afraid of the dark, and save it as robot-story.txt on my Desktop.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Your first web page:&lt;/strong&gt;
&lt;/li&gt;
&lt;/ol&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;   Turn robot-story.txt into a pretty web page with a big title and a nice font, and open it in my browser.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That last one was coding. Claude wrote a web page, and all you did was describe it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Tip:&lt;/strong&gt; if you're unsure how to say something, type it rough, then click the sparkle button at the left of the message box. Claude rewrites your words as a clear prompt before you send them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: Make a home for your projects
&lt;/h2&gt;

&lt;p&gt;Keep your projects together in one folder, away from your Desktop, which fills up fast.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Mac:&lt;/strong&gt; open Finder, press Shift+Cmd+H to go to your home folder (the one with your name), and make a new folder named &lt;code&gt;Projects&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Windows:&lt;/strong&gt; open File Explorer, click &lt;strong&gt;This PC&lt;/strong&gt;, then &lt;strong&gt;Local Disk (C:)&lt;/strong&gt;, and make a new folder named &lt;code&gt;Projects&lt;/code&gt;, so it's at &lt;code&gt;C:\Projects&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Linux:&lt;/strong&gt; make a &lt;code&gt;Projects&lt;/code&gt; folder in your home folder.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Inside &lt;code&gt;Projects&lt;/code&gt;, make one more folder for your first app: &lt;code&gt;my-first-app&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 7: Add it as a project in Moontors
&lt;/h2&gt;

&lt;p&gt;A project is a shortcut to a folder, so you can start a session there with one click.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Click &lt;strong&gt;Add project&lt;/strong&gt;, the folder with a plus beside &lt;strong&gt;Create session&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Click &lt;strong&gt;Choose folder…&lt;/strong&gt; and select the &lt;code&gt;my-first-app&lt;/code&gt; folder inside &lt;code&gt;Projects&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;Click &lt;strong&gt;Add&lt;/strong&gt;. A button for &lt;code&gt;my-first-app&lt;/code&gt; appears at the top of the window.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Step 8: Start a session in your project
&lt;/h2&gt;

&lt;p&gt;Click the &lt;strong&gt;+&lt;/strong&gt; beside &lt;code&gt;my-first-app&lt;/code&gt;'s name. A new session starts in that folder, ready for your first prompt. (If you've signed in to more than one AI helper, you'll see an arrow instead: click it, then choose &lt;strong&gt;Claude Code&lt;/strong&gt; and &lt;strong&gt;Here&lt;/strong&gt;.)&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 9: Build something tiny
&lt;/h2&gt;

&lt;p&gt;Pick one of these for your first prompt:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Make a to-do list web page. I can add tasks, tick them off, and they're still there after I close the browser. Open it when it's done.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Make a tip calculator web page with big buttons that also works on a phone screen. Open it when it's done.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Make a snake game I can play in my browser with the arrow keys. Open it when it's done.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Make a focus timer page: 25 minutes of work, then a 5-minute break, with a sound when time's up. Open it when it's done.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then comes the actual vibe coding: look at what you got and say what you'd like changed, one thing at a time.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;"Make the colours darker and the text bigger."&lt;/li&gt;
&lt;li&gt;"Add a button that clears the finished tasks."&lt;/li&gt;
&lt;li&gt;"When I click Add with an empty box, nothing should happen."&lt;/li&gt;
&lt;li&gt;"Something's wrong: the timer stops when I switch tabs. Fix it."&lt;/li&gt;
&lt;li&gt;"Explain what you just did, in simple words."&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  When something goes wrong
&lt;/h3&gt;

&lt;p&gt;It will, and that's normal. It happens to professional programmers every day.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Tell Claude what you see, in your own words: "The button doesn't do anything." If there's an error message, copy it and paste it in. Claude reads it and fixes the code.&lt;/li&gt;
&lt;li&gt;Not happy with where it's going? Make a new folder, such as &lt;code&gt;my-second-try&lt;/code&gt;, add it as a project, and start fresh. Your first try stays where it was.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Where to go next
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Let it work while you're away.&lt;/strong&gt; Moontors' sessions keep working when you close its window. Give Claude a bigger job, go make dinner, and come back to it done.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Answer from your phone.&lt;/strong&gt; Moontors' phone app follows your sessions and lets you answer them from the couch. It's on Google Play now, and the iPhone app is coming soon.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Try other AI helpers.&lt;/strong&gt; The same steps work with Codex, Gemini CLI and Grok Build, each with its own account, all in one window.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Everyone who builds software today started by not knowing how. You just went from no account to an app you built by describing it. That's vibe coding, and you're doing it.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>beginners</category>
      <category>ai</category>
      <category>claudecode</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Moving from Conductor: the Conductor alternatives, and why Moontors is the one to pick</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Sat, 07 Nov 2026 16:04:05 +0000</pubDate>
      <link>https://dev.to/roidayan/moving-from-conductor-the-conductor-alternatives-and-why-moontors-is-the-one-to-pick-4og3</link>
      <guid>https://dev.to/roidayan/moving-from-conductor-the-conductor-alternatives-and-why-moontors-is-the-one-to-pick-4og3</guid>
      <description>&lt;p&gt;&lt;a href="https://www.conductor.build/announcement" rel="noopener noreferrer"&gt;Conductor is joining Meta&lt;/a&gt;. If you run your coding agents in it, here's what changes and when, what to save first, the alternatives, and why &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt; is the one to pick.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; I build Moontors. Everything about the other tools comes from their own pages, checked on November 7, 2026.&lt;/p&gt;

&lt;h2&gt;
  
  
  What happens to Conductor, and when
&lt;/h2&gt;

&lt;p&gt;From &lt;a href="https://www.conductor.build/announcement" rel="noopener noreferrer"&gt;Conductor's announcement&lt;/a&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;November 4:&lt;/strong&gt; no new sign-ups or downloads. Free users can't upgrade, and Pro users keep Pro's features for free.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;November 13:&lt;/strong&gt; Conductor Cloud goes read-only. You can read old chats and export data, but not send messages. Active subscribers get a refund.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;November 20:&lt;/strong&gt; Conductor Cloud shuts down, and its user data starts being deleted.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Early 2027:&lt;/strong&gt; the Mac app, which gets new models and critical bug fixes until then, winds down, and its user data is deleted.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Before November 13: save your work
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Cloud workspaces:&lt;/strong&gt; right-click each one and choose &lt;strong&gt;Transfer to local&lt;/strong&gt;, which moves it and its chats to your Mac.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your chats:&lt;/strong&gt; Settings → Account → &lt;strong&gt;Export data&lt;/strong&gt; downloads all of it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Local workspaces&lt;/strong&gt; are &lt;a href="https://www.conductor.build/docs/concepts/git-worktrees" rel="noopener noreferrer"&gt;git worktrees&lt;/a&gt;, each on its own branch, usually under &lt;code&gt;~/conductor/workspaces/&amp;lt;repo&amp;gt;/&amp;lt;workspace&amp;gt;&lt;/code&gt;. Their commits are already in your repository: commit what you want to keep, and push each branch (&lt;code&gt;git push -u origin &amp;lt;branch&amp;gt;&lt;/code&gt;). &lt;code&gt;git worktree list&lt;/code&gt; shows them all.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Anything built on &lt;a href="https://www.conductor.build/docs/api" rel="noopener noreferrer"&gt;Conductor's API&lt;/a&gt;:&lt;/strong&gt; it manages cloud workspaces, so it goes with the cloud.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  The alternatives
&lt;/h2&gt;

&lt;p&gt;Every one of these runs on a Mac too.&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;Windows&lt;/th&gt;
&lt;th&gt;Linux&lt;/th&gt;
&lt;th&gt;Agents&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Claude Code, Codex, Gemini CLI and Grok Build, in its own interface, never a terminal&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;a href="https://docs.paradev.com/install" rel="noopener noreferrer"&gt;Paradev&lt;/a&gt; (was Superset)&lt;/td&gt;
&lt;td&gt;Not yet&lt;/td&gt;
&lt;td&gt;Experimental&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/ParadevAI/Paradev#supported-agents" rel="noopener noreferrer"&gt;21 listed&lt;/a&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/stablyai/orca" rel="noopener noreferrer"&gt;Orca&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://onorca.dev/docs/agents/supported" rel="noopener noreferrer"&gt;Any CLI agent, ~40 set up&lt;/a&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/generalaction/emdash" rel="noopener noreferrer"&gt;Emdash&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://emdash.com/docs/providers" rel="noopener noreferrer"&gt;35 agent CLIs&lt;/a&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://code.claude.com/docs/en/desktop" rel="noopener noreferrer"&gt;Claude Code Desktop&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://code.claude.com/docs/en/desktop-linux" rel="noopener noreferrer"&gt;Beta&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Claude Code&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://learn.chatgpt.com/docs/quickstart" rel="noopener noreferrer"&gt;ChatGPT app (Codex)&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://learn.chatgpt.com/docs/linux/linux-app" rel="noopener noreferrer"&gt;Preview&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Codex&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/overview/" rel="noopener noreferrer"&gt;Google Antigravity&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;Yes&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/models" rel="noopener noreferrer"&gt;Its own agent&lt;/a&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Paradev&lt;/strong&gt; is the closest in shape, a workspace per task, and &lt;a href="https://paradev.com/blog/paradev-is-now-mit" rel="noopener noreferrer"&gt;open source under MIT&lt;/a&gt; since November 4. It &lt;a href="https://github.com/ParadevAI/Paradev/blob/desktop-v2.0.5/apps/desktop/src/renderer/routes/sign-in/page.tsx" rel="noopener noreferrer"&gt;needs an account&lt;/a&gt;, isn't on Windows yet, and its &lt;a href="https://docs.paradev.com/install" rel="noopener noreferrer"&gt;iPhone app needs Paradev Pro&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Orca&lt;/strong&gt; and &lt;strong&gt;Emdash&lt;/strong&gt; are open source and run dozens of agent CLIs, Cursor and OpenCode among them.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude Code Desktop&lt;/strong&gt; and &lt;strong&gt;the ChatGPT app&lt;/strong&gt; each run one vendor's agent, with that vendor's cloud.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Google Antigravity&lt;/strong&gt; runs Google's own agent, rather than Claude Code or Codex.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why Moontors
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A real interface, not a terminal.&lt;/strong&gt; Conductor gave you chats rather than a command line, and so does Moontors, for every agent it runs: it drives each one through the agent's own protocol, so you read its replies, answer its permission requests and pick its model in Moontors' own views, each session a live card on one board.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your agents keep working when you close the app.&lt;/strong&gt; In Conductor, that took its cloud, and &lt;a href="https://www.conductor.build/docs/cloud/faq" rel="noopener noreferrer"&gt;local sessions end&lt;/a&gt; when you close the app. Moontors keeps every session working in a background service on your own computer, with no cloud to pay for or to lose.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Answer them from your phone.&lt;/strong&gt; The &lt;a href="https://moontors.com/docs/phone/" rel="noopener noreferrer"&gt;phone app&lt;/a&gt; pairs with your computer directly to follow and answer your sessions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Claude Code and Codex on your own plans&lt;/strong&gt;, Claude Code on your Claude subscription and Codex on your ChatGPT sign-in, beside Gemini CLI and Grok Build.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Agents that work together.&lt;/strong&gt; A Claude Code session can &lt;a href="https://moontors.com/docs/sessions-messaging/" rel="noopener noreferrer"&gt;hand work to a Codex session&lt;/a&gt; and hear back.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Free, with no Moontors account.&lt;/strong&gt; No Moontors server can read your sessions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mac, Windows and Linux&lt;/strong&gt;, the same app on all three.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your worktrees come with you.&lt;/strong&gt; Add your repository as a project, and the worktrees Conductor made are in its &lt;a href="https://moontors.com/docs/diff-and-files/#worktrees-and-sessions" rel="noopener noreferrer"&gt;Git view&lt;/a&gt;, each ready to start a session in.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Moving to Moontors, step by step
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Install it&lt;/strong&gt; from &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;moontors.com&lt;/a&gt; and sign in to your agents. Moontors installs an agent the first time you sign in to it. (&lt;a href="https://moontors.com/docs/sign-in/" rel="noopener noreferrer"&gt;Signing in&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Add your repository as a project&lt;/strong&gt;, the same folder Conductor used. A session in any worktree of that repository counts under the project, wherever the worktree is, &lt;code&gt;~/conductor/workspaces/&lt;/code&gt; included. (&lt;a href="https://moontors.com/docs/projects/" rel="noopener noreferrer"&gt;Projects&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Pick up a Conductor workspace:&lt;/strong&gt; open the project's Git view. Every worktree of the repository is in its left column; its &lt;strong&gt;+&lt;/strong&gt;, or &lt;strong&gt;Start a session here&lt;/strong&gt;, starts a session in that worktree's folder.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Start new work&lt;/strong&gt; with &lt;strong&gt;Run in a new worktree&lt;/strong&gt; in Create session, and name a branch if you want one. Moontors puts the worktree in &lt;code&gt;~/.moontors/worktrees/&lt;/code&gt; and copies in the files your project lists, such as &lt;code&gt;.env&lt;/code&gt;. (&lt;a href="https://moontors.com/docs/worktrees/" rel="noopener noreferrer"&gt;Worktrees&lt;/a&gt;)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Leave Conductor's folders alone&lt;/strong&gt; until every workspace you need has been picked up and its branch pushed.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Good to know
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Your Conductor chats stay in Conductor, so export them before you go.&lt;/li&gt;
&lt;li&gt;Cursor and OpenCode are coming soon; today Moontors runs Claude Code, Codex, Gemini CLI and Grok Build.&lt;/li&gt;
&lt;li&gt;Setup scripts in &lt;code&gt;conductor.json&lt;/code&gt; don't carry over: list the files a new worktree needs, and Moontors copies them in.&lt;/li&gt;
&lt;li&gt;Moontors' Git view commits, pushes and pulls, and your agent opens the pull request when you ask it to.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you're moving from Conductor and something here doesn't fit how you worked, tell me in the comments.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claudecode</category>
      <category>git</category>
      <category>productivity</category>
    </item>
    <item>
      <title>Conductor alternatives for Windows and Linux (November 2026)</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Sat, 07 Nov 2026 15:33:06 +0000</pubDate>
      <link>https://dev.to/roidayan/conductor-alternatives-for-windows-and-linux-november-2026-3c9d</link>
      <guid>https://dev.to/roidayan/conductor-alternatives-for-windows-and-linux-november-2026-3c9d</guid>
      <description>&lt;p&gt;&lt;a href="https://www.conductor.build/docs" rel="noopener noreferrer"&gt;Conductor&lt;/a&gt; popularised running several coding agents at once, each in its own git worktree. It's also &lt;a href="https://www.conductor.build/docs/installation" rel="noopener noreferrer"&gt;macOS only&lt;/a&gt;. If you're on Windows or Linux, these are the tools that do the same job, checked against each one's own pages in November 2026.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Update, November 7:&lt;/strong&gt; Conductor is &lt;a href="https://www.conductor.build/announcement" rel="noopener noreferrer"&gt;joining Meta&lt;/a&gt;. It stopped taking new users on November 4; Conductor Cloud goes read-only on November 13 and starts deleting its data on November 20, and the Mac app gets fixes until early 2027. Every tool below also runs on a Mac.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; I build one of them, &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;. I've tried to describe every tool, mine included, by what its own docs say, and to say where each is the better choice.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short version
&lt;/h2&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;Windows&lt;/th&gt;
&lt;th&gt;Linux&lt;/th&gt;
&lt;th&gt;Agents&lt;/th&gt;
&lt;th&gt;Licence&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/stablyai/orca" rel="noopener noreferrer"&gt;Orca&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://onorca.dev/docs/agents/supported" rel="noopener noreferrer"&gt;Any CLI agent, ~40 set up&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;MIT&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/generalaction/emdash" rel="noopener noreferrer"&gt;Emdash&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://emdash.com/docs/providers" rel="noopener noreferrer"&gt;35 agent CLIs&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Apache-2.0&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/BloopAI/vibe-kanban" rel="noopener noreferrer"&gt;Vibe Kanban&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/BloopAI/vibe-kanban" rel="noopener noreferrer"&gt;10 agents&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Apache-2.0&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://code.claude.com/docs/en/desktop" rel="noopener noreferrer"&gt;Claude Code Desktop&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://code.claude.com/docs/en/desktop-linux" rel="noopener noreferrer"&gt;Beta&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Claude Code&lt;/td&gt;
&lt;td&gt;Closed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://learn.chatgpt.com/docs/quickstart" rel="noopener noreferrer"&gt;ChatGPT app (Codex)&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://learn.chatgpt.com/docs/linux/linux-app" rel="noopener noreferrer"&gt;Preview&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Codex&lt;/td&gt;
&lt;td&gt;Closed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/overview/" rel="noopener noreferrer"&gt;Google Antigravity&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;Yes&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/models" rel="noopener noreferrer"&gt;Its own agent, on Gemini or Claude models&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Closed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;a href="https://docs.paradev.com/install" rel="noopener noreferrer"&gt;Paradev&lt;/a&gt; (was Superset)&lt;/td&gt;
&lt;td&gt;Not yet&lt;/td&gt;
&lt;td&gt;Experimental&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/ParadevAI/Paradev#supported-agents" rel="noopener noreferrer"&gt;21 listed&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;&lt;a href="https://github.com/ParadevAI/Paradev/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;MIT&lt;/a&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Claude Code, Codex, Gemini CLI, Grok Build&lt;/td&gt;
&lt;td&gt;Closed, free&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  Orca
&lt;/h2&gt;

&lt;p&gt;Open source (MIT), on &lt;a href="https://github.com/stablyai/orca" rel="noopener noreferrer"&gt;macOS, Windows and Linux&lt;/a&gt;, with a &lt;a href="https://github.com/stablyai/orca" rel="noopener noreferrer"&gt;git worktree per agent&lt;/a&gt;. It runs any command-line agent in a terminal, with &lt;a href="https://onorca.dev/docs/agents/supported" rel="noopener noreferrer"&gt;about 40 set up already&lt;/a&gt;, and a &lt;a href="https://onorca.dev/docs/model/session-restore" rel="noopener noreferrer"&gt;background daemon&lt;/a&gt; keeps agents running when you close the app.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you want open source and the widest choice of agents.&lt;/p&gt;

&lt;h2&gt;
  
  
  Emdash
&lt;/h2&gt;

&lt;p&gt;Open source (Apache-2.0), on &lt;a href="https://emdash.com/docs/installation" rel="noopener noreferrer"&gt;macOS, Windows and Linux&lt;/a&gt;, a &lt;a href="https://github.com/generalaction/emdash" rel="noopener noreferrer"&gt;git worktree per task&lt;/a&gt;, and &lt;a href="https://emdash.com/docs/providers" rel="noopener noreferrer"&gt;35 agent CLIs&lt;/a&gt;, Claude Code and Codex among them. It &lt;a href="https://github.com/generalaction/emdash/releases/tag/v1.1.38" rel="noopener noreferrer"&gt;removed Gemini CLI&lt;/a&gt; in July.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you want open source with a polished app and don't need Gemini CLI.&lt;/p&gt;

&lt;h2&gt;
  
  
  Vibe Kanban
&lt;/h2&gt;

&lt;p&gt;A kanban board for coding agents, open source (Apache-2.0), run with &lt;code&gt;npx&lt;/code&gt; in your browser or as a &lt;a href="https://github.com/BloopAI/vibe-kanban/releases/tag/v0.1.44-20260424091429" rel="noopener noreferrer"&gt;desktop app on all three systems&lt;/a&gt;. Its site says it &lt;a href="https://www.vibekanban.com" rel="noopener noreferrer"&gt;is sunsetting&lt;/a&gt; and will continue as community-maintained open source.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you think in boards and tickets, and are comfortable with a community-maintained tool.&lt;/p&gt;

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

&lt;p&gt;Anthropic's own app, on &lt;a href="https://code.claude.com/docs/en/desktop" rel="noopener noreferrer"&gt;macOS and Windows&lt;/a&gt;, with Linux &lt;a href="https://code.claude.com/docs/en/desktop-linux" rel="noopener noreferrer"&gt;in beta&lt;/a&gt; for Ubuntu and Debian. It runs Claude Code only, and &lt;a href="https://code.claude.com/docs/en/desktop#cloud-sessions" rel="noopener noreferrer"&gt;cloud sessions keep going&lt;/a&gt; after you close it. It's included with a Claude plan.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you only use Claude Code and want cloud sessions from the vendor itself.&lt;/p&gt;

&lt;h2&gt;
  
  
  The ChatGPT app (Codex)
&lt;/h2&gt;

&lt;p&gt;OpenAI's app, on &lt;a href="https://learn.chatgpt.com/docs/quickstart" rel="noopener noreferrer"&gt;macOS, Windows and Linux&lt;/a&gt; (Linux in preview), with &lt;a href="https://learn.chatgpt.com/docs/environments/modes" rel="noopener noreferrer"&gt;local, worktree or cloud&lt;/a&gt; environments per chat. It runs Codex only, and &lt;a href="https://learn.chatgpt.com/docs/cloud" rel="noopener noreferrer"&gt;cloud tasks keep working&lt;/a&gt; even while your computer sleeps.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you only use Codex, especially if you want tasks to run in the cloud.&lt;/p&gt;

&lt;h2&gt;
  
  
  Google Antigravity
&lt;/h2&gt;

&lt;p&gt;Google's platform, on &lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;macOS, Windows and Linux&lt;/a&gt;, with a desktop app, a command-line client and editor extensions. It has a &lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;worktree mode&lt;/a&gt; and &lt;a href="https://antigravity.google/docs/subagents" rel="noopener noreferrer"&gt;subagents that run in parallel&lt;/a&gt;, and &lt;a href="https://antigravity.google/docs/remote-agents" rel="noopener noreferrer"&gt;Remote Agents&lt;/a&gt; keep working in a Google-managed environment on its paid plans. It runs Google's own agent on &lt;a href="https://antigravity.google/docs/models" rel="noopener noreferrer"&gt;Gemini or Claude models&lt;/a&gt;, rather than Claude Code or Codex.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you want Google's agent with a choice of Gemini and Claude models, and work that carries on in the cloud.&lt;/p&gt;

&lt;h2&gt;
  
  
  Paradev (formerly Superset)
&lt;/h2&gt;

&lt;p&gt;Worth knowing about, but not for Windows yet: it's open source (&lt;a href="https://github.com/ParadevAI/Paradev/blob/main/LICENSE.md" rel="noopener noreferrer"&gt;MIT since 4 November&lt;/a&gt;), on macOS with an &lt;a href="https://docs.paradev.com/install" rel="noopener noreferrer"&gt;experimental Linux build&lt;/a&gt;. It &lt;a href="https://github.com/ParadevAI/Paradev/blob/desktop-v2.0.0/apps/desktop/src/renderer/routes/sign-in/page.tsx" rel="noopener noreferrer"&gt;needs an account&lt;/a&gt;, has a browser pane, automations, an &lt;a href="https://docs.paradev.com/install#iphone-and-ipad" rel="noopener noreferrer"&gt;iPhone and iPad app&lt;/a&gt;, and remote and cloud workspaces on &lt;a href="https://paradev.com/pricing" rel="noopener noreferrer"&gt;paid plans&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you're on a Mac or comfortable with an experimental Linux build, and want automations and team features.&lt;/p&gt;

&lt;h2&gt;
  
  
  &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;
&lt;/h2&gt;

&lt;p&gt;Mine: a free desktop app on macOS, Windows and Linux that runs Claude Code, Codex, Gemini CLI and Grok Build side by side, each session a live card on one board, in its own git worktree when you want one. Sessions run in a background service, so closing the window doesn't stop them; its phone app pairs with your computer directly to follow and answer them (on Google Play now, the iPhone app coming soon). Sessions can message each other, so a Claude Code session can hand work to a Codex session. It needs no Moontors account.&lt;/p&gt;

&lt;p&gt;Its trade-offs: it's closed source, it runs those four agents and no others, it has no cloud runs, and its Windows build isn't signed yet, so SmartScreen warns on first launch.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pick it if&lt;/strong&gt; you want Claude Code, Codex, Gemini CLI and Grok Build on one board, a phone app, and no Moontors account.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to choose
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Open source matters most: &lt;strong&gt;Orca&lt;/strong&gt; or &lt;strong&gt;Emdash&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;One vendor only, with its cloud: &lt;strong&gt;Claude Code Desktop&lt;/strong&gt;, &lt;strong&gt;the ChatGPT app&lt;/strong&gt; or &lt;strong&gt;Google Antigravity&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;Claude Code, Codex, Gemini CLI and Grok Build together, with a phone app: &lt;strong&gt;&lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;On a Mac anyway: &lt;strong&gt;Paradev&lt;/strong&gt; is worth a look too; Conductor no longer takes new users.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Each of these changes fast; if something here is out of date, tell me and I'll fix it.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claudecode</category>
      <category>productivity</category>
      <category>linux</category>
    </item>
    <item>
      <title>What is an agentic development environment (ADE), and do you need one?</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Thu, 05 Nov 2026 13:13:23 +0000</pubDate>
      <link>https://dev.to/roidayan/what-is-an-agentic-development-environment-ade-and-do-you-need-one-50p0</link>
      <guid>https://dev.to/roidayan/what-is-an-agentic-development-environment-ade-and-do-you-need-one-50p0</guid>
      <description>&lt;p&gt;For thirty years the developer's main tool has been built around files: an editor in the middle, a file tree on the left, a terminal at the bottom. Coding agents change what the day looks like. You spend less time typing code and more time handing out tasks, answering an agent's questions, and reading what it changed.&lt;/p&gt;

&lt;p&gt;An &lt;strong&gt;agentic development environment (ADE)&lt;/strong&gt; is a tool built around that work instead of around the editor. The term appeared in 2025, and a dozen tools now use it. Here's what it means in practice, and how to tell whether you need one.&lt;/p&gt;

&lt;h2&gt;
  
  
  The job has changed shape
&lt;/h2&gt;

&lt;p&gt;Working with one agent in a terminal, the loop looks like this:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Give it a task.&lt;/li&gt;
&lt;li&gt;Wait. Answer its questions, allow the commands it wants to run.&lt;/li&gt;
&lt;li&gt;Read what it changed, keep it or throw it away.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That loop fits in one terminal. The trouble starts when you run several at once, which is where the real gain is: while one agent writes tests, another fixes a bug and a third explores a refactor. Now you're juggling terminals, agents are editing the same files, and a session that stopped to ask you something sits idle in a tab you forgot about.&lt;/p&gt;

&lt;h2&gt;
  
  
  What an ADE does
&lt;/h2&gt;

&lt;p&gt;Every tool in the category handles the same handful of problems:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Several agents at once, visible together.&lt;/strong&gt; Each task gets a session, and you can see every session's state, running, waiting for you, done, without switching windows.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Work kept apart.&lt;/strong&gt; Each session can get its own git worktree, a separate folder on its own branch, so one agent's edits never land in another's.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You stay in control.&lt;/strong&gt; Agents ask before running commands or editing files, as much as you choose, and you answer where you're already looking.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Review before you keep.&lt;/strong&gt; Each session's changes are shown as a diff, hunk by hunk, before anything is merged.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Work that doesn't need you watching.&lt;/strong&gt; Agents carry on while you do something else.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How it differs from what you have
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;From an IDE with an agent panel.&lt;/strong&gt; VS Code, Cursor and JetBrains put an agent beside the editor: the editor stays at the centre and the agent works in the folder you have open. An ADE puts the agents at the centre, each in its own folder, whatever editor you use for the bits you still type.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;From a vendor's own app.&lt;/strong&gt; Claude Code Desktop runs Claude Code; the ChatGPT app runs Codex. An ADE that's not tied to a vendor can run several vendors' agents side by side, so you can use each where it suits you and compare them on the same task.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;From agents in a terminal.&lt;/strong&gt; The agents are the same programs. The ADE adds what's around them: the overview, worktrees, review, notifications, and often a background process so closing the window doesn't stop anything.&lt;/p&gt;

&lt;h2&gt;
  
  
  Do you need one?
&lt;/h2&gt;

&lt;p&gt;Probably not if you run one agent at a time on one task. A terminal is fine, and the agent's own tools are enough.&lt;/p&gt;

&lt;p&gt;Probably yes if any of these sound familiar:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;you regularly run two or more agents at once;&lt;/li&gt;
&lt;li&gt;you've had two agents edit the same file;&lt;/li&gt;
&lt;li&gt;you've come back to find an agent was waiting for you for half an hour;&lt;/li&gt;
&lt;li&gt;you use more than one vendor's agent;&lt;/li&gt;
&lt;li&gt;you want to follow or answer agents away from your desk.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Where to start
&lt;/h2&gt;

&lt;p&gt;The category is young and moves fast. Open-source options include Orca and Emdash; Claude Code Desktop and the ChatGPT app cover their own vendor's agent; Conductor and Paradev are Mac-first.&lt;/p&gt;

&lt;p&gt;I build one too, &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;: free, on macOS, Windows and Linux, running Claude Code, Codex, Gemini CLI and Grok Build side by side on one board, with a phone app to answer them away from your desk (on Google Play now, the iPhone app coming soon), and no Moontors account needed. Its own explainer goes deeper on the definition: &lt;a href="https://moontors.com/docs/what-is-an-ade/" rel="noopener noreferrer"&gt;What is an agentic development environment?&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Whichever you pick, the shift is the same: your job is less "write the code" and more "run the team that writes it". It's worth having a tool built for that job.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>devtools</category>
      <category>claudecode</category>
    </item>
    <item>
      <title>Google Antigravity vs Moontors, two ways to run coding agents</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Thu, 05 Nov 2026 12:07:59 +0000</pubDate>
      <link>https://dev.to/roidayan/google-antigravity-vs-moontors-two-ways-to-run-coding-agents-5f2m</link>
      <guid>https://dev.to/roidayan/google-antigravity-vs-moontors-two-ways-to-run-coding-agents-5f2m</guid>
      <description>&lt;p&gt;Running several coding agents at once has gone from a trick to a product category, and &lt;a href="https://antigravity.google/docs/overview/" rel="noopener noreferrer"&gt;Google Antigravity&lt;/a&gt; is now the biggest name in it. It's a good moment to compare it with something built differently.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; I build the other one, &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;. What this says about Antigravity comes from Google's own docs, and each claim links its page.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short version
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Google Antigravity&lt;/th&gt;
&lt;th&gt;Moontors&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;What it is&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;An app, a command-line client, an IDE, editor extensions and an SDK&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;A desktop app with one board, and phone apps&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Agents&lt;/td&gt;
&lt;td&gt;Google's own agent, on &lt;a href="https://antigravity.google/docs/models" rel="noopener noreferrer"&gt;a choice of models&lt;/a&gt;
&lt;/td&gt;
&lt;td&gt;Claude Code, Codex, Gemini CLI and Grok Build themselves&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Systems&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;macOS, Windows, Linux&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;macOS, Windows, Linux&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Worktrees&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/getting-started" rel="noopener noreferrer"&gt;A worktree mode&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;A git worktree per session, when you want one&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;On your phone&lt;/td&gt;
&lt;td&gt;&lt;a href="https://antigravity.google/docs/remote-control" rel="noopener noreferrer"&gt;A browser dashboard&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;A phone app: Android now, iPhone coming soon&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;App closed&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://antigravity.google/docs/remote-agents" rel="noopener noreferrer"&gt;Remote Agents&lt;/a&gt; carry on in Google's environment, on Google AI Pro or Ultra&lt;/td&gt;
&lt;td&gt;Sessions keep working in a background service&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Account&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://antigravity.google/docs/install/" rel="noopener noreferrer"&gt;A Google account&lt;/a&gt;, or a Gemini API key with its command-line client&lt;/td&gt;
&lt;td&gt;No Moontors account needed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Price&lt;/td&gt;
&lt;td&gt;
&lt;a href="https://antigravity.google/docs/plans" rel="noopener noreferrer"&gt;Free with a weekly quota&lt;/a&gt;, more on Google's paid plans&lt;/td&gt;
&lt;td&gt;Free&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  The agents
&lt;/h2&gt;

&lt;p&gt;Antigravity is built around Google's own agent, and you pick the model it runs on: &lt;a href="https://antigravity.google/docs/models" rel="noopener noreferrer"&gt;Gemini or Claude, depending on your plan&lt;/a&gt;. Moontors takes the other road. It runs each vendor's own agent as that vendor ships it, with your own settings, skills and MCP servers: Claude Code, Codex, Gemini CLI and Grok Build, side by side.&lt;/p&gt;

&lt;p&gt;It also changes who you pay. Antigravity's models come with &lt;a href="https://antigravity.google/docs/plans" rel="noopener noreferrer"&gt;Google's plans&lt;/a&gt;, and it takes &lt;a href="https://antigravity.google/docs/install/" rel="noopener noreferrer"&gt;no other vendor's key&lt;/a&gt;, so past its free quota, that's an extra bill if you already pay for Claude or ChatGPT. In Moontors, Claude Code runs on your Claude subscription and Codex on your ChatGPT sign-in.&lt;/p&gt;

&lt;h2&gt;
  
  
  Working together
&lt;/h2&gt;

&lt;p&gt;Both have an answer to "more than one agent on a job", and they're different answers.&lt;/p&gt;

&lt;p&gt;Antigravity's &lt;a href="https://antigravity.google/docs/subagents" rel="noopener noreferrer"&gt;subagents&lt;/a&gt; are started and coordinated by its agent, and message each other among Antigravity's own agents. Its &lt;a href="https://antigravity.google/docs/subagents" rel="noopener noreferrer"&gt;Teamwork&lt;/a&gt; command, on paid plans, coordinates a team of them on a large project.&lt;/p&gt;

&lt;p&gt;In Moontors, sessions can message each other across vendors. You tell your Claude Code session "ask the Codex session to run its tests" in plain words, the message arrives in that session as its next prompt, and the answer comes back the same way.&lt;/p&gt;

&lt;h2&gt;
  
  
  On your phone
&lt;/h2&gt;

&lt;p&gt;Antigravity's &lt;a href="https://antigravity.google/docs/remote-control" rel="noopener noreferrer"&gt;Remote Control&lt;/a&gt; drives your desktop and command-line sessions from any browser, including a phone's, and a phone can keep it on its home screen. Moontors has a native app for your phone, on Google Play now with the iPhone app coming soon: you pair once by QR code, then follow your sessions and answer them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Your sessions
&lt;/h2&gt;

&lt;p&gt;Antigravity's &lt;a href="https://antigravity.google/terms/" rel="noopener noreferrer"&gt;terms&lt;/a&gt; say it records your interactions and uses them to improve Google's products, services and machine learning technologies, and that you can change that in its settings.&lt;/p&gt;

&lt;p&gt;With Moontors, sessions run on your own computer and no Moontors server can read them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Moontors
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Your agents are the vendors' own: Claude Code, Codex, Gemini CLI and Grok Build, each as it ships, with your settings, skills and MCP servers.&lt;/li&gt;
&lt;li&gt;It's free, and your Claude and ChatGPT plans work.&lt;/li&gt;
&lt;li&gt;They work together across vendors, every session a live card on one board.&lt;/li&gt;
&lt;li&gt;You answer them from a native app on your phone, and your sessions stay on your own computer.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The &lt;a href="https://moontors.com/docs/compare/antigravity/" rel="noopener noreferrer"&gt;full comparison&lt;/a&gt; has every source. These tools change fast, so if something here is out of date, tell me and I'll fix it.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>tooling</category>
      <category>claudecode</category>
    </item>
    <item>
      <title>Keep Claude Code running after you close the terminal (and answer it from your phone)</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Tue, 03 Nov 2026 13:13:28 +0000</pubDate>
      <link>https://dev.to/roidayan/keep-claude-code-running-after-you-close-the-terminal-and-answer-it-from-your-phone-a5d</link>
      <guid>https://dev.to/roidayan/keep-claude-code-running-after-you-close-the-terminal-and-answer-it-from-your-phone-a5d</guid>
      <description>&lt;p&gt;The best tasks to give a coding agent are the long ones: a migration, a test suite to fix, a refactor across fifty files. They're also the ones a closed terminal or a sleeping laptop kills halfway through.&lt;/p&gt;

&lt;p&gt;Four things keep a long task alive: something that outlives the terminal, a computer that stays awake, a way to resume, and a way to answer the agent when you're not at your desk.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Outlive the terminal: tmux
&lt;/h2&gt;

&lt;p&gt;Claude Code, Codex and Gemini CLI are interactive programs. They need a terminal, so &lt;code&gt;nohup&lt;/code&gt; and &lt;code&gt;&amp;amp;&lt;/code&gt; don't help. tmux does: it holds the terminal for you.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;tmux new &lt;span class="nt"&gt;-s&lt;/span&gt; refactor          &lt;span class="c"&gt;# a named session&lt;/span&gt;
claude                        &lt;span class="c"&gt;# start the agent inside it&lt;/span&gt;
&lt;span class="c"&gt;# Ctrl-b d                    # detach; the agent keeps working&lt;/span&gt;
tmux attach &lt;span class="nt"&gt;-t&lt;/span&gt; refactor       &lt;span class="c"&gt;# come back later, from any terminal&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Close the terminal window, log out of SSH, quit your terminal app: the agent keeps going. One tmux window per agent keeps several organised (&lt;code&gt;Ctrl-b c&lt;/code&gt; for a new window, &lt;code&gt;Ctrl-b ,&lt;/code&gt; to name it).&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Stay awake
&lt;/h2&gt;

&lt;p&gt;A sleeping computer stops every agent, and a turn in flight can fail when its network connection drops. Keep the machine awake while agents work:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;macOS:&lt;/strong&gt; &lt;code&gt;caffeinate -i&lt;/code&gt; in another tmux window, or &lt;code&gt;caffeinate -i claude&lt;/code&gt; to keep it awake exactly as long as the agent runs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Linux:&lt;/strong&gt; &lt;code&gt;systemd-inhibit --what=idle:sleep claude&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Windows:&lt;/strong&gt; the power settings, or PowerToys Awake.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Closing a laptop's lid usually sleeps it regardless; leave it open, or plug in an external display.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Resume when something does stop
&lt;/h2&gt;

&lt;p&gt;Restarts happen. The conversation isn't lost: &lt;code&gt;claude -c&lt;/code&gt; continues the most recent conversation in that folder, and &lt;code&gt;claude --resume&lt;/code&gt; lets you pick one. The work in progress is in the files and the git history, so the agent can pick up where it was.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Answer it from your phone
&lt;/h2&gt;

&lt;p&gt;A long task usually stops at least once to ask something. If you're away, it waits until you're back. The do-it-yourself fix:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Put your computer and phone on the same private network with &lt;a href="https://tailscale.com" rel="noopener noreferrer"&gt;Tailscale&lt;/a&gt;, so the computer is reachable from anywhere without opening ports.&lt;/li&gt;
&lt;li&gt;Install an SSH client on your phone (Termius, Blink Shell or similar).&lt;/li&gt;
&lt;li&gt;SSH in and run &lt;code&gt;tmux attach -t refactor&lt;/code&gt;. You're in the agent's session, on a phone keyboard.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;It works, and it's free. It's also a terminal UI on a five-inch screen, which is fine for "yes, run the tests" and painful for anything longer.&lt;/p&gt;

&lt;h2&gt;
  
  
  The app version
&lt;/h2&gt;

&lt;p&gt;I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, a free desktop app for coding agents, and it does this without tmux.&lt;/p&gt;

&lt;p&gt;Every session runs in a background service rather than in the window, so closing the window, or quitting the app, doesn't stop the agents; reopen it and every session is there with what it did while you were away. It doesn't keep your computer awake, so step 2 still applies.&lt;/p&gt;

&lt;p&gt;For step 4 there's a phone app, on Google Play now, with the iPhone app coming soon. You pair the phone with your computer once, by QR code, and it talks to your computer directly, on your network or through Tailscale, with no server in between. The phone shows the board, each session's transcript, and the questions and permission requests waiting for you, and you can reply from it.&lt;/p&gt;

&lt;p&gt;It runs Claude Code, Codex, Gemini CLI and Grok Build on macOS, Windows and Linux. The guide is &lt;a href="https://moontors.com/docs/how-to/keep-agents-running-when-the-window-closes/" rel="noopener noreferrer"&gt;How to keep coding agents running when the window closes&lt;/a&gt;.&lt;/p&gt;

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

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Problem&lt;/th&gt;
&lt;th&gt;Do-it-yourself fix&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Terminal closes&lt;/td&gt;
&lt;td&gt;tmux&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Laptop sleeps&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;caffeinate -i&lt;/code&gt; / &lt;code&gt;systemd-inhibit&lt;/code&gt; / power settings&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Something restarts&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;claude -c&lt;/code&gt; or &lt;code&gt;claude --resume&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Agent asks while you're out&lt;/td&gt;
&lt;td&gt;Tailscale + SSH + &lt;code&gt;tmux attach&lt;/code&gt; from your phone&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>claudecode</category>
      <category>tmux</category>
      <category>productivity</category>
      <category>ai</category>
    </item>
    <item>
      <title>Using Claude Code, Codex and Gemini CLI on the same repository</title>
      <dc:creator>Roi Dayan</dc:creator>
      <pubDate>Sun, 01 Nov 2026 14:22:15 +0000</pubDate>
      <link>https://dev.to/roidayan/using-claude-code-codex-and-gemini-cli-on-the-same-repository-1p81</link>
      <guid>https://dev.to/roidayan/using-claude-code-codex-and-gemini-cli-on-the-same-repository-1p81</guid>
      <description>&lt;p&gt;Each coding agent has days when it's the right tool. Running all three on one repository is easy; keeping them consistent is the work. This is how to share instructions and tools between Claude Code, Codex and Gemini CLI, keep their work apart, and hand tasks between them.&lt;/p&gt;

&lt;h2&gt;
  
  
  One set of instructions, three readers
&lt;/h2&gt;

&lt;p&gt;Each agent looks for its own instructions file:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Agent&lt;/th&gt;
&lt;th&gt;Reads by default&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Codex&lt;/td&gt;
&lt;td&gt;&lt;code&gt;AGENTS.md&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Claude Code&lt;/td&gt;
&lt;td&gt;&lt;code&gt;CLAUDE.md&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Gemini CLI&lt;/td&gt;
&lt;td&gt;&lt;code&gt;GEMINI.md&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Keep the real content in one file, &lt;code&gt;AGENTS.md&lt;/code&gt;, and point the others at it:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Claude Code:&lt;/strong&gt; &lt;code&gt;CLAUDE.md&lt;/code&gt; can import another file with an &lt;code&gt;@&lt;/code&gt; line, so a one-line &lt;code&gt;CLAUDE.md&lt;/code&gt; works:
&lt;/li&gt;
&lt;/ul&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight markdown"&gt;&lt;code&gt;  @AGENTS.md
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Gemini CLI:&lt;/strong&gt; its settings can name the files it reads. In &lt;code&gt;.gemini/settings.json&lt;/code&gt;:
&lt;/li&gt;
&lt;/ul&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&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;"context"&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;"fileName"&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;"AGENTS.md"&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.md"&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;ul&gt;
&lt;li&gt;
&lt;strong&gt;Codex:&lt;/strong&gt; reads &lt;code&gt;AGENTS.md&lt;/code&gt; as it is.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Now a change to your conventions lands for all three at once. Agent-specific notes, if you have any, go in that agent's own file below the import.&lt;/p&gt;

&lt;h2&gt;
  
  
  The same tools for each
&lt;/h2&gt;

&lt;p&gt;MCP servers are configured per agent, each in its own format:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Claude Code: &lt;code&gt;claude mcp add &amp;lt;name&amp;gt; -- &amp;lt;command&amp;gt;&lt;/code&gt;, or a &lt;code&gt;.mcp.json&lt;/code&gt; in the project;&lt;/li&gt;
&lt;li&gt;Codex: a &lt;code&gt;[mcp_servers.&amp;lt;name&amp;gt;]&lt;/code&gt; table in &lt;code&gt;~/.codex/config.toml&lt;/code&gt;;&lt;/li&gt;
&lt;li&gt;Gemini CLI: an &lt;code&gt;mcpServers&lt;/code&gt; object in its &lt;code&gt;settings.json&lt;/code&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;It's the same server and the same command in three places. Write it once in a short script or a README section, so adding a server to one agent reminds you to add it to the others.&lt;/p&gt;

&lt;h2&gt;
  
  
  Keep their work apart
&lt;/h2&gt;

&lt;p&gt;Three agents in one folder will collide. Give each task its own git worktree, a separate folder on its own branch:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git worktree add &lt;span class="nt"&gt;-b&lt;/span&gt; claude/refactor-auth ../app-refactor-auth
git worktree add &lt;span class="nt"&gt;-b&lt;/span&gt; codex/api-tests ../app-api-tests
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Prefixing branches with the agent's name makes the history readable later, and makes comparing agents trivial: give two of them the same prompt in two worktrees, then diff the branches.&lt;/p&gt;

&lt;h2&gt;
  
  
  Give each agent the work it suits
&lt;/h2&gt;

&lt;p&gt;Which agent suits which task changes with every release, so measure rather than trust a blog post (including this one). The worktree trick above is the cheapest benchmark you'll ever run: same prompt, same commit, two branches, one diff. After a week you'll know your own split.&lt;/p&gt;

&lt;h2&gt;
  
  
  Hand work between them
&lt;/h2&gt;

&lt;p&gt;The awkward part of a multi-agent setup is the handoff: the Claude Code session finished the refactor, and now you want Codex to write the tests. Copying context between terminals by hand works, but it's where details get lost.&lt;/p&gt;

&lt;p&gt;I build &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;, a free desktop app for running these agents side by side, and this is the part it adds. Every session's agent gets an MCP server of &lt;a href="https://moontors.com" rel="noopener noreferrer"&gt;Moontors&lt;/a&gt;' own with two tools: &lt;code&gt;list_sessions&lt;/code&gt; and &lt;code&gt;send_to_session&lt;/code&gt;. So you can tell your Claude Code session "ask the Codex session on the API to run its tests" in plain words; the message arrives in that session as its next prompt, and the answer comes back the same way. Each session still runs under its own permission mode.&lt;/p&gt;

&lt;p&gt;It also runs each agent as its vendor ships it, with your own settings, skills and MCP servers. It's on macOS, Windows and Linux; the walkthrough is &lt;a href="https://moontors.com/docs/how-to/claude-code-codex-and-gemini-cli-together/" rel="noopener noreferrer"&gt;How to use Claude Code, Codex and Gemini CLI together&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Checklist
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;[ ] &lt;code&gt;AGENTS.md&lt;/code&gt; holds the instructions; &lt;code&gt;CLAUDE.md&lt;/code&gt; imports it; Gemini CLI's settings list it.&lt;/li&gt;
&lt;li&gt;[ ] Every MCP server is added to each agent, from one script or one README section.&lt;/li&gt;
&lt;li&gt;[ ] Every task gets a worktree, and branch names say which agent did the work.&lt;/li&gt;
&lt;li&gt;[ ] You compare agents on your own code, not on benchmarks.&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;AI did the writing, I did the reading, and we both took credit.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claudecode</category>
      <category>codex</category>
      <category>gemini</category>
    </item>
  </channel>
</rss>
