<?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: Mustafa</title>
    <description>The latest articles on DEV Community by Mustafa (@mergisi).</description>
    <link>https://dev.to/mergisi</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%2F4217270%2F2c5665fd-119e-4e47-84fd-b5b59380587e.jpg</url>
      <title>DEV Community: Mustafa</title>
      <link>https://dev.to/mergisi</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/mergisi"/>
    <language>en</language>
    <item>
      <title>AGENTS.md vs CLAUDE.md vs .cursor/rules: one rules file for Cursor, Claude Code and Codex</title>
      <dc:creator>Mustafa</dc:creator>
      <pubDate>Sat, 31 Oct 2026 15:33:23 +0000</pubDate>
      <link>https://dev.to/mergisi/agentsmd-vs-claudemd-vs-cursorrules-one-rules-file-for-cursor-claude-code-and-codex-aj3</link>
      <guid>https://dev.to/mergisi/agentsmd-vs-claudemd-vs-cursorrules-one-rules-file-for-cursor-claude-code-and-codex-aj3</guid>
      <description>&lt;p&gt;If your team uses more than one coding agent, you've probably got three files saying the same thing slightly differently. Here's what each tool actually reads, according to its own docs, and a setup that keeps one source of truth.&lt;/p&gt;

&lt;h2&gt;
  
  
  What each tool reads
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Codex&lt;/strong&gt; (&lt;a href="https://developers.openai.com/codex/guides/agents-md" rel="noopener noreferrer"&gt;docs&lt;/a&gt;) reads &lt;code&gt;AGENTS.md&lt;/code&gt;. It checks &lt;code&gt;~/.codex&lt;/code&gt; for a global file, then walks from the repo root down to your working directory, taking at most one file per directory (&lt;code&gt;AGENTS.override.md&lt;/code&gt; wins over &lt;code&gt;AGENTS.md&lt;/code&gt;). Files closer to you appear later and override earlier ones. Combined size stops at &lt;code&gt;project_doc_max_bytes&lt;/code&gt;, 32 KiB by default.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Cursor&lt;/strong&gt; (&lt;a href="https://cursor.com/docs/context/rules" rel="noopener noreferrer"&gt;docs&lt;/a&gt;) supports project rules in &lt;code&gt;.cursor/rules/&lt;/code&gt; as &lt;code&gt;.mdc&lt;/code&gt; files with &lt;code&gt;description&lt;/code&gt;, &lt;code&gt;globs&lt;/code&gt; and &lt;code&gt;alwaysApply&lt;/code&gt; frontmatter. A plain &lt;code&gt;.md&lt;/code&gt; file in that folder is ignored. It also reads &lt;code&gt;AGENTS.md&lt;/code&gt; in the root and in subdirectories, as a simpler alternative.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Claude Code&lt;/strong&gt; (&lt;a href="https://code.claude.com/docs/en/memory" rel="noopener noreferrer"&gt;docs&lt;/a&gt;) reads &lt;code&gt;CLAUDE.md&lt;/code&gt; (or &lt;code&gt;.claude/CLAUDE.md&lt;/code&gt;), plus &lt;code&gt;.claude/rules/*.md&lt;/code&gt; with optional &lt;code&gt;paths:&lt;/code&gt; frontmatter. Recent versions (v2.1.277+) read &lt;code&gt;AGENTS.md&lt;/code&gt; directly, &lt;strong&gt;but only if there's no &lt;code&gt;CLAUDE.md&lt;/code&gt; or &lt;code&gt;CLAUDE.local.md&lt;/code&gt;&lt;/strong&gt; in your working directory or above it. Adding a &lt;code&gt;CLAUDE.md&lt;/code&gt; silently stops it reading &lt;code&gt;AGENTS.md&lt;/code&gt;, unless you change the Project instructions setting.&lt;/p&gt;

&lt;h2&gt;
  
  
  The setup
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Put shared rules in &lt;code&gt;AGENTS.md&lt;/code&gt;.&lt;/strong&gt; All three tools read it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;If you need Claude-only rules, import rather than copy.&lt;/strong&gt; Claude's docs recommend a &lt;code&gt;CLAUDE.md&lt;/code&gt; like:
&lt;/li&gt;
&lt;/ol&gt;

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

   ## Claude Code
   Use plan mode for changes under &lt;span class="sb"&gt;`src/billing/`&lt;/span&gt;.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It loads &lt;code&gt;AGENTS.md&lt;/code&gt; once, then the extra lines. (A symlink &lt;code&gt;ln -s AGENTS.md CLAUDE.md&lt;/code&gt; also works, but Claude's docs warn it's awkward on Windows.)&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Use &lt;code&gt;.cursor/rules/*.mdc&lt;/code&gt; only for scoped rules&lt;/strong&gt;, e.g. a rule with &lt;code&gt;globs: src/components/**/*.tsx&lt;/code&gt;. Don't duplicate your general rules there.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Nest for monorepos.&lt;/strong&gt; &lt;code&gt;services/payments/AGENTS.md&lt;/code&gt; is picked up by Codex and Cursor for that folder.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  What to put in it
&lt;/h2&gt;

&lt;p&gt;Both Cursor's and Claude's docs say the same thing: short and concrete beats long and vague. Claude's docs suggest under 200 lines per &lt;code&gt;CLAUDE.md&lt;/code&gt;; Cursor's say under 500 per rule.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Build, test and lint commands, exactly as typed.&lt;/li&gt;
&lt;li&gt;Conventions that differ from defaults ("we use pnpm", "no default exports").&lt;/li&gt;
&lt;li&gt;Where things live ("API handlers are in &lt;code&gt;src/api/handlers/&lt;/code&gt;").&lt;/li&gt;
&lt;li&gt;Things the agent got wrong twice.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Skip style guides your linter already enforces.&lt;/p&gt;

&lt;h2&gt;
  
  
  A starter AGENTS.md
&lt;/h2&gt;



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

&lt;span class="gu"&gt;## Commands&lt;/span&gt;
&lt;span class="p"&gt;-&lt;/span&gt; Install: &lt;span class="sb"&gt;`pnpm install`&lt;/span&gt;
&lt;span class="p"&gt;-&lt;/span&gt; Test: &lt;span class="sb"&gt;`pnpm test`&lt;/span&gt; (run before every commit)
&lt;span class="p"&gt;-&lt;/span&gt; Lint: &lt;span class="sb"&gt;`pnpm lint`&lt;/span&gt;

&lt;span class="gu"&gt;## Conventions&lt;/span&gt;
&lt;span class="p"&gt;-&lt;/span&gt; TypeScript strict. Named exports only.
&lt;span class="p"&gt;-&lt;/span&gt; Never edit files in &lt;span class="sb"&gt;`dist/`&lt;/span&gt; or &lt;span class="sb"&gt;`generated/`&lt;/span&gt;.

&lt;span class="gu"&gt;## When unsure&lt;/span&gt;
&lt;span class="p"&gt;-&lt;/span&gt; Read the nearest existing example before writing new code.
&lt;span class="p"&gt;-&lt;/span&gt; Ask before adding a production dependency.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Check it loaded
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Codex: &lt;code&gt;codex --ask-for-approval never "Summarize the current instructions."&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Claude Code: &lt;code&gt;/context&lt;/code&gt; or &lt;code&gt;/memory&lt;/code&gt; and look for the file.&lt;/li&gt;
&lt;li&gt;Cursor: open &lt;strong&gt;Customize â†’ Rules&lt;/strong&gt; and see what's active.&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;The rules file tells the agent how your repo works. The prompt still decides what it does. I put together 15 free prompts for debugging, refactors and tests that work across all three tools: &lt;a href="https://mustafae.gumroad.com/l/ipdvih" rel="noopener noreferrer"&gt;grab them here&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>cursor</category>
      <category>claudecode</category>
      <category>codex</category>
    </item>
    <item>
      <title>Prism shuts down Jan 1, 2027: export your projects to Overleaf without losing work</title>
      <dc:creator>Mustafa</dc:creator>
      <pubDate>Sat, 31 Oct 2026 15:22:59 +0000</pubDate>
      <link>https://dev.to/mergisi/prism-shuts-down-jan-1-2027-export-your-projects-to-overleaf-without-losing-work-38oh</link>
      <guid>https://dev.to/mergisi/prism-shuts-down-jan-1-2027-export-your-projects-to-overleaf-without-losing-work-38oh</guid>
      <description>&lt;p&gt;OpenAI's &lt;a href="https://help.openai.com/en/articles/20001615-prism-to-codex-migration-guide" rel="noopener noreferrer"&gt;Prism to Codex migration guide&lt;/a&gt; is blunt: "Prism will shut down on January 1, 2027." And: "After shutdown, all Prism-related data will be deleted... Once the data is deleted, it cannot be recovered."&lt;/p&gt;

&lt;p&gt;The export is easy. What it leaves out is the part people lose. Here's how to get everything out and onto Overleaf (or a local folder) safely.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the export does and doesn't keep
&lt;/h2&gt;

&lt;p&gt;According to OpenAI's guide (checked Oct 31, 2026):&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Kept:&lt;/strong&gt; your project files, as a zip.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Not kept:&lt;/strong&gt; project history, inline comments, collaborator permissions, and the live Zotero connection.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Maybe kept:&lt;/strong&gt; the all-projects archive can include a &lt;code&gt;conversations&lt;/code&gt; folder (JSON) for some projects. It's for reference only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Shared projects you don't own:&lt;/strong&gt; the zip has the files but not your Prism conversation history.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;So the order is: save the things the zip drops, then export.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: Save what the zip leaves out
&lt;/h2&gt;

&lt;p&gt;OpenAI's advice is to "copy important context and save milestone versions before shutdown."&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Comments:&lt;/strong&gt; open each project and copy the comments you need into a &lt;code&gt;NOTES.md&lt;/code&gt;, or into &lt;code&gt;%&lt;/code&gt; comments next to the text they refer to.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;History:&lt;/strong&gt; if an older version matters, save it as a dated copy, e.g. &lt;code&gt;paper-2026-10-31/&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Zotero:&lt;/strong&gt; the live link stops working. Export a fresh &lt;code&gt;.bib&lt;/code&gt; from Zotero into the project folder. Your &lt;code&gt;\cite{}&lt;/code&gt; keys keep working as long as they're in the &lt;code&gt;.bib&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Collaborators:&lt;/strong&gt; write down who had access so you can invite them again.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 2: Export
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;All projects: &lt;strong&gt;Settings â†’ Data Controls â†’ Export all projects (zip) â†’ Download zip&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;One project: the project-title menu â†’ &lt;strong&gt;Export (zip)&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;For each important project, open its last good PDF preview and click &lt;strong&gt;Download PDF&lt;/strong&gt;. That PDF is your reference for checking the new build.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If the archive contains &lt;code&gt;export-errors.json&lt;/code&gt;, the guide says to retry with the project menu's &lt;strong&gt;Export (zip)&lt;/strong&gt;, and to email &lt;a href="mailto:support@openai.com"&gt;support@openai.com&lt;/a&gt; if a project is still missing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: Check the zip before you trust it
&lt;/h2&gt;

&lt;p&gt;Unzip it and check, per project:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;unzip &lt;span class="nt"&gt;-l&lt;/span&gt; prism-export.zip | less          &lt;span class="c"&gt;# is every project there?&lt;/span&gt;
find &lt;span class="nb"&gt;.&lt;/span&gt; &lt;span class="nt"&gt;-name&lt;/span&gt; &lt;span class="s1"&gt;'*.tex'&lt;/span&gt; &lt;span class="nt"&gt;-exec&lt;/span&gt; &lt;span class="nb"&gt;grep&lt;/span&gt; &lt;span class="nt"&gt;-l&lt;/span&gt; &lt;span class="s1"&gt;'\\documentclass'&lt;/span&gt; &lt;span class="o"&gt;{}&lt;/span&gt; +   &lt;span class="c"&gt;# root documents&lt;/span&gt;
find &lt;span class="nb"&gt;.&lt;/span&gt; &lt;span class="nt"&gt;-name&lt;/span&gt; &lt;span class="s1"&gt;'*.bib'&lt;/span&gt;                       &lt;span class="c"&gt;# bibliographies made it?&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Compare file counts against what you see in Prism. Keep a SHA-256 list (&lt;code&gt;find . -type f -exec sha256sum {} + &amp;gt; MANIFEST.txt&lt;/code&gt;) so you can prove later that nothing changed.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: Upload to Overleaf
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;In Overleaf: &lt;strong&gt;New project â†’ Upload project&lt;/strong&gt;, and pick a zip of &lt;strong&gt;one&lt;/strong&gt; project (zip the project folder, not the whole archive).&lt;/li&gt;
&lt;li&gt;Mind Overleaf's limits from its &lt;a href="https://www.overleaf.com/learn/how-to/Uploading_a_project" rel="noopener noreferrer"&gt;upload docs&lt;/a&gt;: 50 MB per file, 180 files per upload, 2,000 files per project, 7 MB of editable text.&lt;/li&gt;
&lt;li&gt;Overleaf compiles with pdfLaTeX by default. If your project used XeLaTeX or LuaLaTeX, change the compiler in the menu and check the main document is set.&lt;/li&gt;
&lt;li&gt;Compile and compare against the PDF you saved in Step 2.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Or move to Codex (OpenAI's own path)
&lt;/h2&gt;

&lt;p&gt;The guide's route is to open the extracted folder as a project in Codex in the ChatGPT desktop app, run it locally with approvals on, and start with an audit prompt that inspects the project without changing it. Either way, keep the original zip untouched.&lt;/p&gt;

&lt;h2&gt;
  
  
  Don't leave it for December
&lt;/h2&gt;

&lt;p&gt;Exporting takes minutes. Rebuilding lost comments from memory doesn't. Do the export this week and keep the zip in two places.&lt;/p&gt;




&lt;p&gt;I wrote a small offline Python script that checks a Prism export (manifest, root &lt;code&gt;.tex&lt;/code&gt;, &lt;code&gt;.bib&lt;/code&gt;, export errors) and builds one Overleaf-ready zip per project, plus a checklist PDF. It's &lt;a href="https://mustafae.gumroad.com/l/gnslhl" rel="noopener noreferrer"&gt;here&lt;/a&gt; if it helps. The steps above work without it.&lt;/p&gt;

</description>
      <category>latex</category>
      <category>openai</category>
      <category>overleaf</category>
      <category>writing</category>
    </item>
    <item>
      <title>OpenAI is retiring GPT-4o, 4.1 and 5-nano: how to find every model ID in your repo</title>
      <dc:creator>Mustafa</dc:creator>
      <pubDate>Sat, 31 Oct 2026 15:22:47 +0000</pubDate>
      <link>https://dev.to/mergisi/openai-is-retiring-gpt-4o-41-and-5-nano-how-to-find-every-model-id-in-your-repo-2n79</link>
      <guid>https://dev.to/mergisi/openai-is-retiring-gpt-4o-41-and-5-nano-how-to-find-every-model-id-in-your-repo-2n79</guid>
      <description>&lt;p&gt;OpenAI's &lt;a href="https://developers.openai.com/api/docs/deprecations" rel="noopener noreferrer"&gt;deprecations page&lt;/a&gt; has a busy few months coming up. Some model IDs already stopped working on Oct 23, 2026. More shut down on Dec 11, then Jan 14, then Apr 1. If a model string is hard-coded in a config file, a &lt;code&gt;.env&lt;/code&gt;, or a test fixture, you won't find out until requests start failing.&lt;/p&gt;

&lt;p&gt;Here are the dates that matter (checked against the page on Oct 31, 2026) and a free one-liner that finds them.&lt;/p&gt;

&lt;h2&gt;
  
  
  The dates
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Shutdown&lt;/th&gt;
&lt;th&gt;Model IDs&lt;/th&gt;
&lt;th&gt;Replacement listed by OpenAI&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Already gone (Oct 23, 2026)&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-4.1-nano&lt;/code&gt;, &lt;code&gt;gpt-4o-2024-05-13&lt;/code&gt;, &lt;code&gt;gpt-4-turbo&lt;/code&gt;, &lt;code&gt;gpt-3.5-turbo&lt;/code&gt;, &lt;code&gt;o1&lt;/code&gt;, &lt;code&gt;o3-mini&lt;/code&gt;, &lt;code&gt;o4-mini&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-5.6-luna&lt;/code&gt; / &lt;code&gt;gpt-5.6-sol&lt;/code&gt; / &lt;code&gt;gpt-5.6-terra&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Dec 1, 2026&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-image-1-mini&lt;/code&gt;, &lt;code&gt;gpt-image-1.5&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-image-2.5-sunburst&lt;/code&gt; or &lt;code&gt;-flare&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Dec 11, 2026&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-5&lt;/code&gt;, &lt;code&gt;gpt-5-mini&lt;/code&gt;, &lt;code&gt;gpt-5-nano&lt;/code&gt;, &lt;code&gt;gpt-5-pro&lt;/code&gt;, &lt;code&gt;o3&lt;/code&gt;, &lt;code&gt;o3-pro&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-5.6-sol&lt;/code&gt; / &lt;code&gt;-terra&lt;/code&gt; / &lt;code&gt;-luna&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Jan 6, 2027&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;tts-1&lt;/code&gt;, &lt;code&gt;tts-1-hd&lt;/code&gt;, &lt;code&gt;gpt-4o-mini-tts&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;&lt;code&gt;gpt-realtime-2.1-mini&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Jan 14, 2027&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-4o-2024-08-06&lt;/code&gt;, &lt;code&gt;gpt-4o-2024-11-20&lt;/code&gt;, &lt;code&gt;gpt-4o-mini-2024-07-18&lt;/code&gt;, &lt;code&gt;gpt-4.1-2025-04-14&lt;/code&gt;, &lt;code&gt;gpt-4.1-mini-2025-04-14&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;&lt;code&gt;gpt-6-luna&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Jan 20, 2027&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-realtime&lt;/code&gt;, &lt;code&gt;gpt-audio&lt;/code&gt;, &lt;code&gt;gpt-4o-realtime&lt;/code&gt;, &lt;code&gt;gpt-4o-audio&lt;/code&gt; and mini variants&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-realtime-2.1&lt;/code&gt; / &lt;code&gt;gpt-audio-1.5&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Feb 26, 2027&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;whisper-1&lt;/code&gt;, &lt;code&gt;gpt-4o-transcribe&lt;/code&gt;, &lt;code&gt;gpt-4o-mini-transcribe&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-live-transcribe&lt;/code&gt; or &lt;code&gt;gpt-transcribe&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Apr 1, 2027&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-5.4-nano&lt;/code&gt;, &lt;code&gt;gpt-5.1&lt;/code&gt;, &lt;code&gt;gpt-5.3-codex&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-6-luna&lt;/code&gt; / &lt;code&gt;gpt-6-sol&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;This is a summary. The page lists dated snapshots for most rows, so check it for the exact ID you use. OpenAI also notes that fine-tuned models follow their base model's schedule.&lt;/p&gt;

&lt;p&gt;One catch: the bare aliases &lt;code&gt;gpt-4o&lt;/code&gt;, &lt;code&gt;gpt-4o-mini&lt;/code&gt;, &lt;code&gt;gpt-4.1&lt;/code&gt; and &lt;code&gt;gpt-4.1-mini&lt;/code&gt; aren't rows on the page. They point at snapshots that are on it, so treat them as Jan 14 and confirm which snapshot your account resolves them to.&lt;/p&gt;

&lt;h2&gt;
  
  
  The one-liner
&lt;/h2&gt;

&lt;p&gt;Run this from your repo root. It needs GNU &lt;code&gt;grep&lt;/code&gt; (for &lt;code&gt;-P&lt;/code&gt;; on macOS, &lt;code&gt;brew install grep&lt;/code&gt; and use &lt;code&gt;ggrep&lt;/code&gt;):&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="nb"&gt;grep&lt;/span&gt; &lt;span class="nt"&gt;-rnIoP&lt;/span&gt; &lt;span class="nt"&gt;--exclude-dir&lt;/span&gt;&lt;span class="o"&gt;={&lt;/span&gt;.git,node_modules,venv,.venv,dist,build&lt;span class="o"&gt;}&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="s1"&gt;'(?&amp;lt;![\w.-])(gpt-4o(-mini)?(-(2024-[0-9]{2}-[0-9]{2}|tts|realtime|audio|transcribe))?|gpt-4\.1(-mini|-nano)?(-2025-04-14)?|gpt-5(-mini|-nano|-pro)?(-20[0-9]{2}-[0-9]{2}-[0-9]{2})?|gpt-5\.4-nano|gpt-5\.1|o1|o3(-mini|-pro)?|o4-mini|gpt-3\.5-turbo|gpt-4-turbo|tts-1(-hd)?|whisper-1)(?![\w.-])'&lt;/span&gt; &lt;span class="nb"&gt;.&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  | &lt;span class="nb"&gt;sort&lt;/span&gt; | &lt;span class="nb"&gt;uniq&lt;/span&gt; &lt;span class="nt"&gt;-c&lt;/span&gt; | &lt;span class="nb"&gt;sort&lt;/span&gt; &lt;span class="nt"&gt;-rn&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It covers &lt;code&gt;.env&lt;/code&gt;, YAML and JSON as well as code, because &lt;code&gt;-r&lt;/code&gt; reads every text file. The lookarounds stop it matching &lt;code&gt;gpt-5&lt;/code&gt; inside &lt;code&gt;gpt-5.6-luna&lt;/code&gt;. You get &lt;code&gt;count path:line:match&lt;/code&gt;, which is enough to open a ticket per service.&lt;/p&gt;

&lt;p&gt;Watch out for false positives. &lt;code&gt;o1&lt;/code&gt; and &lt;code&gt;o3&lt;/code&gt; are short, so the regex can catch things that aren't model names. Read the matches before you change anything.&lt;/p&gt;

&lt;h2&gt;
  
  
  What grep can't see
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;IDs built at runtime&lt;/strong&gt;, like &lt;code&gt;f"gpt-{version}"&lt;/code&gt;. Search for &lt;code&gt;"gpt-"&lt;/code&gt; string fragments too.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;IDs stored in a database or a feature flag service.&lt;/strong&gt; Query those directly.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Third-party SDK defaults.&lt;/strong&gt; A library may pick a model for you. Check its config.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Other repos.&lt;/strong&gt; Infra-as-code, notebooks and Zapier-style automations often have their own copies.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  A safe swap order
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Fix anything that's &lt;strong&gt;already broken&lt;/strong&gt; first. Those calls are failing now.&lt;/li&gt;
&lt;li&gt;Change the model ID in &lt;strong&gt;one place&lt;/strong&gt; (an env var or a config constant), not across 40 files.&lt;/li&gt;
&lt;li&gt;Run your evals or a small golden set on the replacement before you switch traffic. Newer models can behave differently on structured output and tool calls.&lt;/li&gt;
&lt;li&gt;Add a CI check so the old IDs don't come back in a copy-pasted snippet.&lt;/li&gt;
&lt;/ol&gt;




&lt;p&gt;I packaged this into a small scanner that reads the dates from a JSON file, marks rows past their date as BROKEN NOW, writes a CSV/markdown swap list and can fail a CI job. If that saves you time, it's &lt;a href="https://mustafae.gumroad.com/l/hqpcu" rel="noopener noreferrer"&gt;here&lt;/a&gt;. The grep above works fine on its own.&lt;/p&gt;

</description>
      <category>openai</category>
      <category>ai</category>
      <category>devops</category>
      <category>python</category>
    </item>
    <item>
      <title>How to export your Google Jules data before November 11 (and move to Claude Code or Codex)</title>
      <dc:creator>Mustafa</dc:creator>
      <pubDate>Fri, 30 Oct 2026 12:38:19 +0000</pubDate>
      <link>https://dev.to/mergisi/how-to-export-your-google-jules-data-before-november-11-and-move-to-claude-code-or-codex-72</link>
      <guid>https://dev.to/mergisi/how-to-export-your-google-jules-data-before-november-11-and-move-to-claude-code-or-codex-72</guid>
      <description>&lt;p&gt;Google announced on October 29 that Jules is winding down. The dates, from the &lt;a href="https://jules.google/docs/changelog/2026-10-29/" rel="noopener noreferrer"&gt;official changelog&lt;/a&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Until November 11, 2026:&lt;/strong&gt; you can keep working and finish sessions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;After November 11:&lt;/strong&gt; Jules turns read-only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;December 31, 2026:&lt;/strong&gt; Jules shuts down.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Your code, branches and pull requests already live in GitHub, so they're safe. What isn't in GitHub is everything Jules kept on its side: settings, environment variables, scheduled tasks, per-repo setup scripts, memories, and the full history of every session. Google published an export for exactly that. Here's how I'd run it, what you get, and where each piece goes in Claude Code or Codex.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 0: clean up open work first
&lt;/h2&gt;

&lt;p&gt;Before exporting, go through your open Jules pull requests and merge or close them. Any session you still want to finish has to be done before November 11. After that you can read it, but you can't continue it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: install Jules Tools and sign in
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npm &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="nt"&gt;-g&lt;/span&gt; @google/jules
jules login
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;jules login&lt;/code&gt; opens a browser sign-in, so run it yourself in a terminal.&lt;/p&gt;

&lt;p&gt;If you used Jules with GitHub repos, check that your GitHub account is still connected to Jules. Google's note is blunt: if it isn't, the export won't include repository data. A quick test is:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;jules remote list &lt;span class="nt"&gt;--repo&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If that fails with a 401 that mentions GitHub, connect the account at &lt;a href="https://github.com/apps/google-labs-jules/installations/select_target" rel="noopener noreferrer"&gt;https://github.com/apps/google-labs-jules/installations/select_target&lt;/a&gt; and try again.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: run the official export prompt
&lt;/h2&gt;

&lt;p&gt;The changelog page has an "export prompt" (click "Show the export prompt"). Open a CLI coding agent such as Claude Code &lt;strong&gt;from your home directory&lt;/strong&gt;, paste the whole prompt, and let it work. The prompt tells the agent to check your setup, save a Python script, run it, and report counts. It also tells the agent not to print your token or env var values.&lt;/p&gt;

&lt;p&gt;A few things the prompt itself warns about, worth knowing up front:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The access token expires about an hour after login. If every call returns 401, run &lt;code&gt;jules login&lt;/code&gt; again.&lt;/li&gt;
&lt;li&gt;The script uses undocumented endpoints. If a call starts returning 404 or 400, note which one rather than guessing at a fix.&lt;/li&gt;
&lt;li&gt;Seeing a few more sessions in the export than &lt;code&gt;jules remote list --session&lt;/code&gt; shows is expected.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 3: know what you got
&lt;/h2&gt;

&lt;p&gt;Everything lands in &lt;code&gt;~/jules-export&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;~/jules-export/
  user-settings.json
  env-variables.json
  scheduled-tasks.json
  sources.json            # connected repositories
  sessions.json
  repos/&amp;lt;repo&amp;gt;/
    setup.sh              # environment setup script
    test-command.txt
    memories.json
    memories.md
  sessions/&amp;lt;id&amp;gt;/
    session.json
    activities.json       # full activity log
    summary.md            # readable: prompt, plan, outputs
    patch-1.diff ...
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A repo folder only gets &lt;code&gt;setup.sh&lt;/code&gt; and &lt;code&gt;test-command.txt&lt;/code&gt; if a session recorded them, and only gets memories if it had any. An empty folder isn't necessarily a failed export.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: treat env-variables.json as a secret
&lt;/h2&gt;

&lt;p&gt;This is the part I'd not skip. &lt;code&gt;env-variables.json&lt;/code&gt; contains your environment variable &lt;strong&gt;values&lt;/strong&gt;, not just names. Google says to keep the export folder private and out of your repositories.&lt;/p&gt;

&lt;p&gt;What I do:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Add &lt;code&gt;jules-export/&lt;/code&gt; to my global gitignore so it can't sneak into a commit.&lt;/li&gt;
&lt;li&gt;Write down the variable &lt;em&gt;names&lt;/em&gt; I need, then put the values into the new home (GitHub Actions secrets, a local &lt;code&gt;.env&lt;/code&gt; that's ignored, your secret manager).&lt;/li&gt;
&lt;li&gt;Rotate anything that's an actual credential. Those values sat with a third party and now sit on disk in plain text.&lt;/li&gt;
&lt;li&gt;Delete &lt;code&gt;env-variables.json&lt;/code&gt; once that's done.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Step 5: map each file to its new home
&lt;/h2&gt;

&lt;p&gt;Claude Code reads &lt;code&gt;CLAUDE.md&lt;/code&gt; in the repo; Codex reads &lt;code&gt;AGENTS.md&lt;/code&gt;. Both are plain Markdown instructions, so the move is mostly copy and trim.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Jules export&lt;/th&gt;
&lt;th&gt;Where it goes&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;repos/&amp;lt;repo&amp;gt;/setup.sh&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A "Setup" section in &lt;code&gt;CLAUDE.md&lt;/code&gt; / &lt;code&gt;AGENTS.md&lt;/code&gt;, and the install step of your CI workflow&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;repos/&amp;lt;repo&amp;gt;/test-command.txt&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A "How to test" line in the same file, and the test step in CI&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;repos/&amp;lt;repo&amp;gt;/memories.md&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Paste into &lt;code&gt;CLAUDE.md&lt;/code&gt; / &lt;code&gt;AGENTS.md&lt;/code&gt;, then cut anything stale&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;env-variables.json&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Names into docs/CI, values into secrets (see step 4)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;scheduled-tasks.json&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A cron job or GitHub Actions &lt;code&gt;schedule:&lt;/code&gt; trigger (below)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;sessions/&amp;lt;id&amp;gt;/patch-N.diff&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;git apply&lt;/code&gt; if the work never got pushed&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;sessions/&amp;lt;id&amp;gt;/summary.md&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Reference only. Useful when you forget why something was done&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;A minimal &lt;code&gt;AGENTS.md&lt;/code&gt; built from the export might be:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight markdown"&gt;&lt;code&gt;&lt;span class="gh"&gt;# Agent notes&lt;/span&gt;

&lt;span class="gu"&gt;## Setup&lt;/span&gt;
(paste setup.sh here)

&lt;span class="gu"&gt;## Tests&lt;/span&gt;
Run: (paste test-command.txt here)

&lt;span class="gu"&gt;## Things to remember&lt;/span&gt;
(trimmed memories.md)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Keep it short. A long, outdated memory list gets followed less well than ten lines that are true.&lt;/p&gt;

&lt;h3&gt;
  
  
  Scheduled tasks
&lt;/h3&gt;

&lt;p&gt;Each Jules scheduled task is basically a prompt plus a schedule. Both replacements have a non-interactive mode: &lt;code&gt;claude -p "&amp;lt;prompt&amp;gt;"&lt;/code&gt; for Claude Code, &lt;code&gt;codex exec "&amp;lt;prompt&amp;gt;"&lt;/code&gt; for Codex. You can call either from a local cron, or from a GitHub Actions workflow with an &lt;code&gt;on: schedule:&lt;/code&gt; cron trigger and the API key stored as a repository secret.&lt;/p&gt;

&lt;p&gt;The exact JSON shape of &lt;code&gt;scheduled-tasks.json&lt;/code&gt; isn't documented, so open it and read each task's prompt and schedule by hand. Then run each new job manually once before you trust it on a timer.&lt;/p&gt;

&lt;h3&gt;
  
  
  Unpushed patches
&lt;/h3&gt;

&lt;p&gt;If a session produced code that never became a branch, the diff is still in the export:&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="nb"&gt;cd &lt;/span&gt;your-repo
git checkout &lt;span class="nt"&gt;-b&lt;/span&gt; recover-jules-session
git apply ~/jules-export/sessions/&amp;lt;&lt;span class="nb"&gt;id&lt;/span&gt;&lt;span class="o"&gt;&amp;gt;&lt;/span&gt;/patch-1.diff
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Use &lt;code&gt;git apply --check&lt;/code&gt; first if the repo has moved on since; old patches won't always apply cleanly.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: verify while you still can
&lt;/h2&gt;

&lt;p&gt;Between November 11 and December 31 Jules is read-only, which is a good window to compare. Open a few sessions in the Jules UI next to their &lt;code&gt;summary.md&lt;/code&gt; files, check the repo count in &lt;code&gt;sources.json&lt;/code&gt; matches what you had connected, and make sure every scheduled task has a replacement. After December 31 there's nothing left to compare against.&lt;/p&gt;

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

&lt;ul&gt;
&lt;li&gt;[ ] Merge or close open Jules PRs&lt;/li&gt;
&lt;li&gt;[ ] Finish any sessions you need before Nov 11&lt;/li&gt;
&lt;li&gt;[ ] &lt;code&gt;npm install -g @google/jules&lt;/code&gt; and &lt;code&gt;jules login&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;[ ] Confirm GitHub is connected (&lt;code&gt;jules remote list --repo&lt;/code&gt;)&lt;/li&gt;
&lt;li&gt;[ ] Run the official export prompt from your home directory&lt;/li&gt;
&lt;li&gt;[ ] Keep &lt;code&gt;~/jules-export&lt;/code&gt; out of git; move secrets, rotate credentials&lt;/li&gt;
&lt;li&gt;[ ] Write &lt;code&gt;CLAUDE.md&lt;/code&gt; / &lt;code&gt;AGENTS.md&lt;/code&gt; per repo from setup, test command and memories&lt;/li&gt;
&lt;li&gt;[ ] Rebuild scheduled tasks with &lt;code&gt;claude -p&lt;/code&gt; or &lt;code&gt;codex exec&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;[ ] &lt;code&gt;git apply&lt;/code&gt; any unpushed patches&lt;/li&gt;
&lt;li&gt;[ ] Spot-check the export against Jules before Dec 31&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;Disclosure: I made a small paid script that does step 5 automatically (CLAUDE.md, AGENTS.md and CI workflows per repo, plus a key-rotation list), $19 at &lt;a href="https://mustafae.gumroad.com/l/zxjwzd" rel="noopener noreferrer"&gt;https://mustafae.gumroad.com/l/zxjwzd&lt;/a&gt;. The manual steps above work fine without it.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claudecode</category>
      <category>githubactions</category>
      <category>jules</category>
    </item>
  </channel>
</rss>
