# Agent Snacks > Fresh snacks from the AI agent world: new agents, tools and tricks for first-timers and builders alike. Every bite links to its source. Every Thursday. Bite-sized, published weekly on Thursdays (around 7:00 AM ET). Every item links to an official changelog or release note. Independent. Sponsored items are labelled. Subscribe: https://agentsnacks.beehiiv.com/subscribe. ## Issues - [Snack Pack #5: Mark the dates before the switches flip](https://agentsnacks.ai/packs/005/): Mark the dates before the switches flip. Copilot in Teams needs a new app, Cursor and OpenAI add agent powers, and Claude tightens access. All source-linked. - [Snack Pack #4: Check what your editor let in](https://agentsnacks.ai/packs/004/): Check what your editor let in. Cursor pulled a tampered Claude Code extension, and Claude Code, Gemini and OpenAI set deadlines. All source-linked. - [Snack Pack #3: Let your agents do the clicking](https://agentsnacks.ai/packs/003/): Let your agents do the clicking: OpenAI’s GPT-6.1 Terra handles screen chores, plus OpenCode, Cline, Claude, Copilot, Replit and Grok Bot changes to check this week. All source-linked. - [Snack Pack #2: Hand off the chores, keep the final say](https://agentsnacks.ai/packs/002/): Hand off the chores, keep the final say: Claude Duet asks before it acts, plus PHP, Pydantic AI, Claude Code and Cline fixes to check this week. All source-linked. - [Snack Pack #1: Your agents got a babysitter](https://agentsnacks.ai/packs/001/): Your agents got a babysitter: Copilot CLI Autopilot no longer stalls on a hook’s “ask”, and unpinned Cline setups get new default models. All source-linked. ## Tool hubs - [Claude](https://agentsnacks.ai/tools/claude/) - [ChatGPT](https://agentsnacks.ai/tools/chatgpt/) - [Gemini](https://agentsnacks.ai/tools/gemini/) - [Microsoft Copilot](https://agentsnacks.ai/tools/microsoft-copilot/) - [Claude Code](https://agentsnacks.ai/tools/claude-code/) - [GitHub Copilot](https://agentsnacks.ai/tools/copilot/) - [Cursor](https://agentsnacks.ai/tools/cursor/) - [Gemini CLI](https://agentsnacks.ai/tools/gemini-cli/) - [Codex CLI](https://agentsnacks.ai/tools/codex/) - [Cline](https://agentsnacks.ai/tools/cline/) - [Devin Desktop (Windsurf)](https://agentsnacks.ai/tools/devin-desktop/) - [OpenCode](https://agentsnacks.ai/tools/opencode/) - [Pydantic AI](https://agentsnacks.ai/tools/pydantic-ai/) ## Pages - [Home](https://agentsnacks.ai/) - [Archive](https://agentsnacks.ai/packs/) - [Tools](https://agentsnacks.ai/tools/) - [Subscribe](https://agentsnacks.ai/subscribe/) — beehiiv hosted: https://agentsnacks.beehiiv.com/subscribe - [FAQ](https://agentsnacks.ai/faq/) - [Sponsor](https://agentsnacks.ai/sponsor/) — inquire at sponsors@agentsnacks.ai - [llms-full.txt](https://agentsnacks.ai/llms-full.txt) - [Creators shelf: image and video tools for agents](https://agentsnacks.ai/shelf/#creators) ## Sponsors - [Sponsor page](https://agentsnacks.ai/sponsor/): founding offer, formats, house rules, how to apply - [sponsor.json](https://agentsnacks.ai/sponsor.json): machine-readable terms (placements, specs, pay-what-you-want pricing, audience, cadence, payment, review policy, apply fields) - Apply by email to sponsors@agentsnacks.ai (subject "Sponsor application", fields listed in sponsor.json). AI agents may apply on behalf of a company. All applications are reviewed before acceptance. - Payment: PayPal invoice from billing@agentsnacks.ai (USD). Audience: new publication, launching 2026. ## For agents - [Start here (beginners)](https://agentsnacks.ai/start/) - [Agents guide](https://agentsnacks.ai/agents/) - [Video Tasting](https://agentsnacks.ai/videos/) · [First Bite](https://agentsnacks.ai/first-bite/) · [Secret Sauce](https://agentsnacks.ai/secret-sauce/) · [Before you go](https://agentsnacks.ai/before-you-go/) · [Shelf directory](https://agentsnacks.ai/shelf/) · [Main Course index](https://agentsnacks.ai/main-course/) · [Search](https://agentsnacks.ai/search/) - [Latest JSON](https://agentsnacks.ai/api/latest.json) · [All snacks JSON](https://agentsnacks.ai/api/snacks.json) · [Source-checked only](https://agentsnacks.ai/api/source-checked.json) · [Schema](https://agentsnacks.ai/api/schema.json) - Per tool: [chatgpt](https://agentsnacks.ai/api/tools/chatgpt.json), [claude](https://agentsnacks.ai/api/tools/claude.json), [claude-code](https://agentsnacks.ai/api/tools/claude-code.json), [cline](https://agentsnacks.ai/api/tools/cline.json), [codex](https://agentsnacks.ai/api/tools/codex.json), [copilot](https://agentsnacks.ai/api/tools/copilot.json), [cursor](https://agentsnacks.ai/api/tools/cursor.json), [devin-desktop](https://agentsnacks.ai/api/tools/devin-desktop.json), [gemini](https://agentsnacks.ai/api/tools/gemini.json), [gemini-cli](https://agentsnacks.ai/api/tools/gemini-cli.json), [mcp](https://agentsnacks.ai/api/tools/mcp.json), [microsoft-copilot](https://agentsnacks.ai/api/tools/microsoft-copilot.json), [opencode](https://agentsnacks.ai/api/tools/opencode.json), [other](https://agentsnacks.ai/api/tools/other.json), [pydantic-ai](https://agentsnacks.ai/api/tools/pydantic-ai.json), [skills](https://agentsnacks.ai/api/tools/skills.json) - Feeds: [RSS](https://agentsnacks.ai/feed.xml) · [Atom](https://agentsnacks.ai/atom.xml) · [JSON Feed](https://agentsnacks.ai/feed.json) - Raw markdown: [Snack Pack #5](https://agentsnacks.ai/packs/005/index.md), [Snack Pack #4](https://agentsnacks.ai/packs/004/index.md), [Snack Pack #3](https://agentsnacks.ai/packs/003/index.md), [Snack Pack #2](https://agentsnacks.ai/packs/002/index.md), [Snack Pack #1](https://agentsnacks.ai/packs/001/index.md), [Daily](https://agentsnacks.ai/daily/index.md) ## FAQ ### Do you track me? Only with cookieless analytics via Umami: page views and clicks, counted in aggregate. Our pages set no cookies of their own, run no ad trackers and collect no personal data, and search records only how long your query was, never what you typed. The beehiiv signup form on /subscribe/ sets cookies and loads beehiiv's scripts, including Google Tag Manager. Videos are click to play: nothing loads from YouTube until you press play, then the video plays from YouTube (youtube-nocookie.com). ### What is Agent Snacks? A free weekly newsletter of bite-sized, source-linked AI agent news: what shipped in ChatGPT, Claude, Gemini, Copilot, Cursor, Codex and more, and why it might matter to you. It lands every Thursday. ### Who is it for? People who use AI agents at work or at home, from curious first-timers to developers, who want what changed without hype or rumors. ### How do you write each snack? The first sentence leads with the concrete change for the reader, with exact versions, settings and dates right after it. Question hooks stay rare. Security fixes, outages, billing changes, data loss and breaking changes skip the jokes and open with a plain action, like Update now. Each card bolds one phrase in its first sentence, never the title, and we never claim to have tested something we only read about. ### How do you pick items? We collect official GitHub releases, changelog RSS and vendor changelog pages, then write each snack only from primary-source text. If we can't link a claim to its official source, it doesn't run. ### When does it send? Thursdays, around 7:00 AM ET. Subscribe at https://agentsnacks.ai/subscribe/. ### How do I subscribe? Hit Subscribe on any page. Unsubscribe is one click from any email. ### Is this affiliated with any vendor? No. Independent. Sponsored items are always labelled Sponsored and never mixed into the news snacks. ### How do I sponsor? Email sponsors@agentsnacks.ai with your product, target dates and preferred slot. Founding sponsors pay what they want (suggested $0–$49 per issue); see the Sponsor page. ### How do I contact you? Say hi at hello@agentsnacks.ai. Sponsorship goes to sponsors@agentsnacks.ai. ### Where are past Snack Packs? On the Packs page (/packs/). Each issue also appears as a static HTML page with JSON-LD NewsArticle markup. ### What do the spice peppers mean? Each snack gets a spice level. 🌶️ Mild: nice to know. 🌶️🌶️ Medium: worth a look. 🌶️🌶️🌶️ Spicy: act now — security, a deadline, a break, or a big launch. Hover or focus the peppers for the short why on that item. ## Tool update summaries (source-linked) ### Claude Official: https://support.claude.com/en/articles/12138966-release-notes - 2026-10-16 — Cowork and chat become one Claude: No more picking a mode: Cowork now lives inside ordinary Claude chats. Its get-it-done powers roll into chats on web, desktop and phone for Pro and Max plans. Chats, projects, connectors, and skills carry over. Source: https://support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude - 2026-10-01 — Cowork on web and mobile: Close the laptop and your Cowork session keeps cooking. Sessions can run in the cloud and show up on phone as well as desktop. Source: https://support.claude.com/en/articles/15520349-use-claude-cowork-on-web-desktop-and-mobile ### ChatGPT Official: https://learn.chatgpt.com/docs/changelog - 2026-10-16 — ChatGPT Work for longer multi-step jobs: ChatGPT Work takes on the long, many-step jobs. It handles research, connected apps and files, and finished docs, sheets, decks and sites, with you steering as it goes. Source: https://learn.chatgpt.com/docs/changelog ### Gemini Official: https://gemini.google/release-notes/ - 2026-10-13 — Gemini 4 Argon in the Gemini app: Gemini 4 Argon lands in the Gemini app for tougher jobs. It rolls out for Google AI Pro and Ultra, built for start-to-finish tasks, with Antigravity coding built into the Mac app. Source: https://gemini.google/release-notes/ - 2026-09-01 — Gemini Spark personal agent: Gemini Spark wants to do your busywork, not just answer. Spark works on tasks across Google apps and local files (Mac app for Ultra). Source: https://gemini.google/release-notes/ ### Microsoft Copilot Official: https://support.microsoft.com/en-us/microsoft-365-copilot/learning/changes-microsoft-copilot-app - 2026-08-18 — One Copilot app for personal, work, and school: One Copilot app for all your accounts. Microsoft's updates bring personal, work and school sign-ins together, with Home combining chat and Cowork-style help. Source: https://support.microsoft.com/en-us/microsoft-365-copilot/learning/changes-microsoft-copilot-app ### Claude Code Official: https://github.com/anthropics/claude-code/releases - 2026-10-13 — Claude Code tidies up long chats you leave idle: Walk away from a long chat and Claude Code tidies it up. Terminal sessions over 200K tokens compact after ~54 minutes idle, so the next message does not reprocess everything. CLAUDE_CODE_DISABLE_FILE_MENTIONS passes @-mention paths instead of file contents. OpenTelemetry now splits cache writes into 5m and 1h buckets. Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.299 (Snack Pack #1) - 2026-10-09 — Claude Code subagents can each pick their own model and memory budget: Subagents get their own memory budget and model. Each subagent can set its own auto-compact window, workflow subagents can pick a different model, and overloaded retries wait longer before trying again. Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.296 ### GitHub Copilot Official: https://github.blog/changelog/label/copilot/ - 2026-10-15 — Copilot can now fix your code-scanning alerts for everyone on the right plan: Copilot can now fix your security alerts and prove it. Assign a CodeQL alert; it explores, fixes, re-runs CodeQL, then opens a draft PR. Batch up to 25 alerts. Needs the Agentic autofix enterprise policy. Source: https://github.blog/changelog/2026-10-15-agentic-autofix-for-code-scanning-alerts-general-availability (Snack Pack #1) - 2026-10-14 — HydraFusion picks from more models and adds a Max tier for hard tasks: HydraFusion gets a bigger model pool and a Max tier. New in the pool: Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna. Medium effort for drafts, high for review; Max handles harder tasks. It is still an early test. Source: https://github.blog/changelog/2026-10-14-hydrafusion-updates-and-new-max-option (Snack Pack #1) - 2026-10-13 — Gemini 4 Argon is now open to every paid Copilot plan: Gemini 4 Argon is now generally available in Copilot. Use it in VS Code, Visual Studio, Copilot CLI, coding agent, Copilot app, JetBrains, Xcode and Eclipse for Pro+/Max/Business/Enterprise. Gradual rollout. Source: https://github.blog/changelog/2026-10-13-gemini-4-argon-in-github-copilot - 2026-10-09 — Copilot's local sandbox is ready for everyone, and Copilot CLI finds your Ollama models: Copilot's local sandbox is now generally available. It covers Copilot CLI, the Copilot app and VS Code Agent Host. In Copilot CLI, /model discovers local Ollama models. Source: https://github.blog/changelog/2026-10-09-github-copilot-weekly-releases-october-5/ - 2026-10-14 — Copilot CLI lets you expand and copy long /ask answers: Long /ask answers in Copilot CLI are easier to read and reuse. Press Ctrl+E to expand one and Ctrl+Y to copy it. Source: https://github.com/github/copilot-cli/releases/tag/v1.0.98 (Snack Pack #1) - 2026-10-13 — Copilot CLI Autopilot no longer stalls on a hook's “ask” when nobody is watching: Autopilot no longer stalls on a hook's “ask” when nobody is watching. It now denies a preToolUse “ask” instead of stopping on a permission dialog. Custom agents also get declarative hooks. Source: https://github.com/github/copilot-cli/releases/tag/v1.0.97 (Snack Pack #1) - 2026-10-12 — Copilot CLI can open one more folder just for this session: Let Copilot CLI into one more folder, just for this session. The new /add-dir for session-only sandbox access; interactive settings suggest secret-masking hosts. Source: https://github.com/github/copilot-cli/releases/tag/v1.0.96 (Snack Pack #1) ### Cursor Official: https://cursor.com/changelog - 2026-10-13 — Cursor Origin now connects to 20+ apps you already use: Cursor's Origin now plugs into 20+ apps you already use. They cover deploy, CI, review, security, monitoring and project management. Public submissions open. Apps use JWT + installation tokens and write check runs / PR comments. Source: https://cursor.com/changelog/origin-app-marketplace (Snack Pack #1) ### Gemini CLI Official: https://github.com/google-gemini/gemini-cli/releases - 2026-10-13 — Gemini CLI now checks folder trust in headless runs too: Update now: Gemini CLI tightens folder trust in headless mode. Update to 0.64.0 or later: folder-trust now carries into headless mode, and untrusted folders get read-only workspace settings next. Source: https://github.com/google-gemini/gemini-cli/releases/tag/v0.64.0 (Snack Pack #1) ### Codex CLI Official: https://github.com/openai/codex/releases ### Cline Official: https://github.com/cline/cline/releases - 2026-10-13 — Cline now defaults unpinned setups to Claude Sonnet 5.5: **Check your model**: unpinned Cline setups now default to Sonnet 5.5. The switch covers Cline and OpenRouter providers with no pinned model. Catalog adds Haiku 5.5; ~20 other providers change defaults. Reasoning-off 400s fixed on several frontier models. Source: https://github.com/cline/cline/releases/tag/v4.1.24 (Snack Pack #1) - 2026-10-13 — Cline Desktop adds a clickable outline for long chats: Jump around long Cline Desktop chats with a clickable outline. The outline sits along the conversation and makes plugin failure status clearer. Source: https://github.com/cline/cline/releases/tag/desktop-v0.0.46 (Snack Pack #1) ### Devin Desktop (Windsurf) Official: https://windsurf.com/changelog - 2026-10-12 — Devin Desktop lets admins lock the extension marketplace: Devin Desktop speaks a newer MCP and lets admins lock the marketplace. It also speaks the newest MCP revision, and enterprise admins can pin a single extension marketplace URL. Faster @-mentions; code-context category removed. Source: https://windsurf.com/changelog (Snack Pack #1) ### OpenCode Official: https://github.com/anomalyco/opencode/releases - 2026-10-10 — OpenCode 2 is here, but you have to upgrade on purpose: **Action needed**: OpenCode 2 is here, and updates are no longer automatic. Run opencode upgrade --major (or /update in the TUI) to move up. It also allows OpenAI's ultrafast service tier. Source: https://github.com/anomalyco/opencode/releases/tag/v1.19.0 (Snack Pack #1) ### Pydantic AI Official: https://github.com/pydantic/pydantic-ai/releases ## Start Here (full text) We picked five quick wins below. Each takes about 10–15 minutes and links to the official how-to. ## 5 things AI agents can do for you (real examples)

No code needed

### 1. Turn a spreadsheet into clear charts **Upload a spreadsheet nobody wants to read** and ChatGPT returns the key insights and clear charts. Agent: [ChatGPT](/tools/chatgpt/) · about 10 min

Official guide: data analysis with ChatGPT

### 2. Turn hours of web research into one report **Gemini Deep Research browses the web for you** and returns a report with sources, so market research stops eating an afternoon of tabs. Agent: [Gemini Deep Research](/tools/gemini/) · about 15 min

Official guide: use Deep Research in Gemini Apps

If you write code

### 3. Hand off a coding task and come back to the result **Connect your GitHub repository and Codex works on it in the cloud** while you do something else. Agent: [OpenAI Codex (cloud)](/tools/codex/) · about 15 min

Official guide: Codex cloud

### 4. Tidy up old code in your project **Claude Code finds outdated parts, updates them and runs your checks** so nothing broke. Agent: [Claude Code](/tools/claude-code/) · about 15 min

Official guide: clean up code with Claude Code

### 5. Ask your project how it works, then fix a bug **Claude Code explains how unfamiliar code fits together**, then fixes that one annoying bug. Agent: [Claude Code](/tools/claude-code/) · about 15 min

Official guide: get started with Claude Code

## Keep snacking See what agents shipped today in [{LATEST_DAILY_TITLE}]({LATEST_DAILY_URL}) or browse [all Daily snacks](/daily/). Get the best of the week every Thursday: [subscribe to the Snack Pack](/subscribe/). Want the bigger picture? Read [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) from Anthropic, or browse the [Shelf](/shelf/) for more tools. ## Video Tasting (official tutorials) ### Claude’s new mobile app updates are here (2:32, beginner) https://www.youtube.com/watch?v=SkpwzfR4A1E Channel: Claude · Published 2026-10-29 After watching you can: Keep your work and personal Claude accounts signed in on one phone, ask Claude something through CarPlay on a drive, and switch the app’s menus to your own language. - You can stay signed in to two Claude accounts on your phone. Switch between work and personal without signing out, and photos also show up in the chat as soon as you pick them. - On iPhone Duo and iPad, a chart or anything else Claude makes opens in its own column beside the chat, so you can read it and ask for changes at the same time. Chromebooks and foldables get a resizable sidebar, right-click and the Claude Desktop keyboard shortcuts. - Voice mode answers sooner, hears you better in a noisy room and works with Apple CarPlay on iPhone. Menus and settings now come in 33 languages, and pull-down search is on the way for iPhone and iPad (it’s already on web and desktop). - Watch just 0:11 Two accounts in one app: https://www.youtube.com/watch?v=SkpwzfR4A1E&t=11s - Watch just 0:30 What Claude makes opens beside the chat: https://www.youtube.com/watch?v=SkpwzfR4A1E&t=30s - Watch just 1:26 Faster voice mode and CarPlay: https://www.youtube.com/watch?v=SkpwzfR4A1E&t=86s (Summary source: description + captions + visual analysis) ### A new way to use GitHub Copilot in JetBrains IDEs (1:40, intermediate) https://www.youtube.com/watch?v=uhQ-kFaFIFc Channel: GitHub · Published 2026-11-03 After watching you can: Add Air to your JetBrains IDE, connect the Copilot account you already have, and hand it a bug fix with a test while you watch the diff. - Copilot now runs as an agent in JetBrains Air. Install the Air plugin from the JetBrains Marketplace, choose GitHub Copilot and connect your existing Copilot account. - The demo gives Copilot a local task to fix a bug and add a test. It works with your project context and IDE tools, and you follow the agent session and inspect the diff right in the IDE. - With Air Teams, a task can run against a repository in an isolated cloud environment, and you review the changes and open a pull request from a browser or your phone. The regular GitHub Copilot plugin stays available too. - Watch just 0:18 Install Air and connect Copilot: https://www.youtube.com/watch?v=uhQ-kFaFIFc&t=18s - Watch just 0:32 Local task to fix a bug and add a test: https://www.youtube.com/watch?v=uhQ-kFaFIFc&t=32s - Watch just 1:03 Cloud tasks with Air Teams: https://www.youtube.com/watch?v=uhQ-kFaFIFc&t=63s (Summary source: description + captions + visual analysis) ### How an Anthropic engineer uses Claude Docs (1:23, beginner) https://www.youtube.com/watch?v=SkFmiMHDN38 Channel: Claude · Published 2026-10-20 After watching you can: Bring Claude into a shared team doc to pull the numbers, check a claim and answer teammates’ comments. - Claude Docs is a shared doc where Claude works alongside your team. In standup, everyone types updates and Claude suggests topics worth talking about. - For a launch write-up, Doug writes the words himself and asks Claude for the numbers. Claude drops a chart right into the doc. - Highlight a sentence and ask Claude “is this right?” to get it confirmed or corrected. Teammates ask questions in the comments, and Claude answers some of them. - Watch just 0:14 Team standups: https://www.youtube.com/watch?v=SkFmiMHDN38&t=14s - Watch just 0:36 Launch review: chart, fact-check and comments: https://www.youtube.com/watch?v=SkFmiMHDN38&t=36s (Summary source: description + captions + visual analysis) ### Track tokens and models with the My AI Usage Canvas | GitHub Copilot (4:39, intermediate) https://www.youtube.com/watch?v=2xiyBnQFM20 Channel: GitHub · Published 2026-10-17 After watching you can: See where your Copilot tokens go, which models you lean on, and try one small habit change to use fewer. - Copilot’s /chronicle commands look back at your past sessions. The My AI Usage Canvas, a custom dashboard, turns that history into charts in the Copilot app. - The Usage tab shows tokens (the chunks of text AI models read and write), model calls, your model mix, top repos and the hours you usually start sessions. - The Insights tab writes a tip from your own history and suggests experiments. Start one, note what you tried, mark it done, and revisit saved insights later. - Watch just 0:54 Exploring the usage tab and model mix: https://www.youtube.com/watch?v=2xiyBnQFM20&t=54s - Watch just 2:38 Generating personalized tips in the insights tab: https://www.youtube.com/watch?v=2xiyBnQFM20&t=158s - Watch just 3:09 Running experiments to improve AI habits: https://www.youtube.com/watch?v=2xiyBnQFM20&t=189s (Summary source: chapters + captions + description + visual analysis) ### Coding on the go with Claude Code (4:02, intermediate) https://www.youtube.com/watch?v=-OH1y7CjIew Channel: Claude · Published 2026-10-13 After watching you can: Hand Claude Code a task in the cloud, close your laptop, and answer its questions from your phone. - A cloud session runs Claude Code on a virtual machine (a computer in the cloud), not your laptop. It copies your repo (your project's code on GitHub) and works on its own branch (a separate copy of the code), so it keeps going while your laptop sleeps. - Good for long jobs like a big refactor or migration, ideas that hit you away from your desk, or several tasks at once. It counts toward your plan's limits, with no extra charge for the virtual machine. - Connect GitHub and pick which repos Claude can see. Then check in from the phone app, review the changes and tap Create PR (a pull request: a proposed change you approve before it joins the main code). - Watch just 0:18 What is a cloud session?: https://www.youtube.com/watch?v=-OH1y7CjIew&t=18s - Watch just 0:41 When to use one: https://www.youtube.com/watch?v=-OH1y7CjIew&t=41s - Watch just 1:09 Connecting GitHub: https://www.youtube.com/watch?v=-OH1y7CjIew&t=69s - Watch just 2:50 From your phone: https://www.youtube.com/watch?v=-OH1y7CjIew&t=170s (Summary source: chapters + captions + visual analysis) ### Getting started with Claude Code Projects (4:33, advanced) https://www.youtube.com/watch?v=omihvjf0A2k Channel: Claude · Published 2026-10-09 After watching you can: Tell when a goal is big enough for a Claude Code project, and how Claude splits it into threads you check on. - A project is one ongoing chat where Claude splits a big goal into tasks and runs each as a thread: its own Claude Code session on its own branch (a separate copy of the code), in the cloud, so work goes on with your laptop closed. - A thread opens a pull request (a proposed change you approve before it joins the main code) when its change is ready, then fixes failed checks and review comments. To merge, just ask Claude. - Use it for goals bigger than one session or spread over several repos (project code folders). Threads use your plan's limits faster, so keep the coordinator on low effort. Projects are in beta, rolling out to Pro and Max. - Watch just 0:20 What is a project?: https://www.youtube.com/watch?v=omihvjf0A2k&t=20s - Watch just 0:44 When to use a project?: https://www.youtube.com/watch?v=omihvjf0A2k&t=44s - Watch just 2:04 Reviewing pull requests: https://www.youtube.com/watch?v=omihvjf0A2k&t=124s - Watch just 3:17 Tips from the Claude Code team: https://www.youtube.com/watch?v=omihvjf0A2k&t=197s (Summary source: chapters + captions + visual analysis) ### Getting started with Claude.ai (5:19, beginner) https://www.youtube.com/watch?v=0vZ_UVLhSQQ Channel: Anthropic · Published 2025-12-02 After watching you can: Have your first good chat with Claude: ask well, add files, and pick a model or style. - New to Claude? Here's the friendly tour of the chat screen, plus prompting basics. - How to add context and files, tweak replies (styles, model choice) and use research mode. - Wraps up with tips for working with Claude over the long haul. - Watch just 0:56 Prompting basics: https://www.youtube.com/watch?v=0vZ_UVLhSQQ&t=56s - Watch just 2:05 Adding context and files: https://www.youtube.com/watch?v=0vZ_UVLhSQQ&t=125s (Summary source: chapters + description) ### Getting started with projects in Claude.ai (7:08, beginner) https://www.youtube.com/watch?v=GJ5jTgcbRHA Channel: Anthropic · Published 2025-12-02 After watching you can: Set up a Claude Project once, so you stop re-explaining yourself every chat. - Projects stop you re-explaining yourself in every chat. Think of a lunchbox that keeps your instructions and files across chats. - Make your first project, write its instructions and add knowledge. - Share it and set permissions for your team. - Watch just 1:33 Creating your first project: https://www.youtube.com/watch?v=GJ5jTgcbRHA&t=93s - Watch just 2:21 Defining project instructions: https://www.youtube.com/watch?v=GJ5jTgcbRHA&t=141s (Summary source: chapters + description) ### Getting started with connectors in Claude.ai (3:43, beginner) https://www.youtube.com/watch?v=_jjSS0qGFbI Channel: Anthropic · Published 2025-12-11 After watching you can: Hook Claude up to the apps and files you already use, and keep those hookups tidy. - Plug Claude into your files, apps and workflows with connectors. - How to set up new connectors and manage the ones you have. - Ends with a real-world workflow you can copy. - Watch just 0:50 Setting up new connectors: https://www.youtube.com/watch?v=_jjSS0qGFbI&t=50s - Watch just 2:25 Real-world workflow example: https://www.youtube.com/watch?v=_jjSS0qGFbI&t=145s (Summary source: chapters + description) ### Claude Agent Skills Explained (3:14, intermediate) https://www.youtube.com/watch?v=fOxC44g8vig Channel: Anthropic · Published 2025-11-26 After watching you can: Tell skills, CLAUDE.md, MCP servers and subagents apart, and know when to use each. - Skills are expertise Claude loads only when it needs it. Each one is a folder that kicks in automatically when a task matches. - Only each skill's name and description load at startup; the full SKILL.md loads on demand. - Compares skills with CLAUDE.md (project facts), MCP (connects to data) and subagents (fixed roles). - Watch just 0:30 How Agent Skills work: https://www.youtube.com/watch?v=fOxC44g8vig&t=30s - Watch just 1:42 Agent Skills vs MCP Servers: https://www.youtube.com/watch?v=fOxC44g8vig&t=102s - Watch just 2:33 Stacking skills into one bigger agent: https://www.youtube.com/watch?v=fOxC44g8vig&t=153s (Summary source: transcript + chapters + visual analysis) ### Why we built—and donated—the Model Context Protocol (MCP) (35:32, intermediate) https://www.youtube.com/watch?v=PLyCki2K0Lg Channel: Anthropic · Published 2025-12-11 After watching you can: Explain what MCP is, why it exists, and why it's now an open, foundation-run standard. - Why MCP exists, straight from one of its co-creators. Stuart Ritchie talks with David Soria Parra about where the protocol came from. - Covers the problem MCP solves, the USB-C analogy, community adoption and criticisms. - Explains why Anthropic donated MCP to the Linux Foundation. - Watch just 0:00 What is MCP?: https://www.youtube.com/watch?v=PLyCki2K0Lg&t=0s - Watch just 2:46 The USB-C analogy: https://www.youtube.com/watch?v=PLyCki2K0Lg&t=166s - Watch just 32:53 Advice for non-developers: https://www.youtube.com/watch?v=PLyCki2K0Lg&t=1973s (Summary source: chapters + description) ### Getting started with research in Claude.ai (2:50, beginner) https://www.youtube.com/watch?v=R-KJgjIrh24 Channel: Anthropic · Published 2025-12-02 After watching you can: Kick off a Research task in Claude and know when it's worth the wait. - Hand Claude the digging and go sip your coffee. See how to start a Research task and what it's doing meanwhile. - The jobs it's best at: digging across many sources. - How it teams up with extended thinking. - Watch just 0:00 Initiating research: https://www.youtube.com/watch?v=R-KJgjIrh24&t=0s - Watch just 1:38 Ideal use cases: https://www.youtube.com/watch?v=R-KJgjIrh24&t=98s (Summary source: chapters + description) ### Let Claude handle work in your browser (1:35, beginner) https://www.youtube.com/watch?v=rBJnWMD0Pho Channel: Anthropic · Published 2025-12-18 After watching you can: See a browser agent in action: Claude for Chrome looking, clicking and typing for you. - Watch Claude click around the web for you in Chrome. It's a browser extension that can see, click, type and move around pages. - Three demos: dashboard data into one doc, fixing slide comments, and testing a Claude Code build in Chrome. - Basically ‘chatbot vs agent’, live. (Summary source: description) ### Introducing Cowork: Claude Code for the rest of your work (1:09, beginner) https://www.youtube.com/watch?v=UAmKyyZ-b9E Channel: Anthropic · Published 2026-01-12 After watching you can: Get Cowork: hand Claude a big task, point it at your stuff, come back to a finished file. - Cowork hands back finished work from your time-eating tasks. - It pulls from your local files, cloud tools and the web. - Out come spreadsheets, slides, documents and PDFs. (Summary source: description) ### Your tools are now interactive in Claude (0:54, beginner) https://www.youtube.com/watch?v=bluAmTHoEow Channel: Anthropic · Published 2026-01-26 After watching you can: See what ‘interactive tools inside Claude’ actually looks like. - No more tab-hopping: your connected tools now work right inside the Claude chat. - Examples: Asana projects, Slack drafts, Amplitude charts, Figma diagrams. - Everything happens in one window. (Summary source: description) ### Introducing Runway MCP (0:45, beginner) https://www.youtube.com/watch?v=KhXxcU6cgzM Channel: Runway · Published 2026-05-27 After watching you can: Picture how an agent makes images and videos for you through Runway MCP. - Make Runway images and videos from the tools you already use. Runway MCP plugs into Claude, ChatGPT, Cursor, Replit and more. - Pick models like Gen-4.5, Seedance 2.0, GPT Images 2.0 and Kling. - No app-switching needed. (Summary source: description) ### Using Workflows | Runway Academy (2:35, beginner) https://www.youtube.com/watch?v=MsVl_CKu_gk Channel: Runway · Published 2025-11-21 After watching you can: Run a featured Runway workflow and tweak it into your own little tool. - Build your own custom tools inside Runway with Workflows. - Start from a featured workflow, run it, then see how its steps connect. - Customize it and make it yours. - Watch just 0:25 Using featured workflows: https://www.youtube.com/watch?v=MsVl_CKu_gk&t=25s - Watch just 0:44 Running a workflow: https://www.youtube.com/watch?v=MsVl_CKu_gk&t=44s - Watch just 1:27 Customizing workflows: https://www.youtube.com/watch?v=MsVl_CKu_gk&t=87s (Summary source: chapters + description) ### Collaborate on artifacts in Claude (2:40, beginner) https://www.youtube.com/watch?v=kJA3fQnntpY Channel: Claude · Published 2026-11-04 After watching you can: Share a Claude artifact with teammates, set who can view, comment or edit, and ask Claude to apply their comments. - Ask Claude for a doc or tracker, then share it with teammates. - Pick an access level for each person, from viewer to editor, with commenter and contributor roles on shared trackers. - When someone leaves a comment, ask Claude to make the update, whether it is content or styling. - Watch just 0:48 Sharing and roles: https://www.youtube.com/watch?v=kJA3fQnntpY&t=48s - Watch just 1:12 Turn the doc into a tracker: https://www.youtube.com/watch?v=kJA3fQnntpY&t=72s - Watch just 2:15 Find shared artifacts: https://www.youtube.com/watch?v=kJA3fQnntpY&t=135s (Summary source: description + visual analysis) ### Your apps have moved into Claude (1:00, beginner) https://www.youtube.com/watch?v=Gy1jSeSzAhk Channel: Claude · Published 2026-11-10 After watching you can: Know what interactive connectors look like in Claude and which kinds of edits you can ask for. - Interactive connectors open the full version of a tool inside Claude. - The demo adds a summary to a Confluence launch plan, applies a dither effect to a photo in Adobe, and checks a Box floor plan against a design standard. - Make changes yourself, or ask Claude to make them with you. Rolling out on web and desktop. - Watch just 0:11 Confluence summary: https://www.youtube.com/watch?v=Gy1jSeSzAhk&t=11s - Watch just 0:22 Adobe dither effect: https://www.youtube.com/watch?v=Gy1jSeSzAhk&t=22s - Watch just 0:32 Box floor plan check: https://www.youtube.com/watch?v=Gy1jSeSzAhk&t=32s (Summary source: description + visual analysis) ## Creator corner: making images and video with agents - [Video] Using Workflows | Runway Academy (2025-11-21): Build your own custom tools inside Runway with Workflows. Start from a featured workflow, run it, then see how its steps connect. Customize it and make it yours. Source: https://www.youtube.com/watch?v=MsVl_CKu_gk - [Video] Introducing Runway MCP (2026-05-27): Make Runway images and videos from the tools you already use. Runway MCP plugs into Claude, ChatGPT, Cursor, Replit and more. Pick models like Gen-4.5, Seedance 2.0, GPT Images 2.0 and Kling. No app-switching needed. Source: https://www.youtube.com/watch?v=KhXxcU6cgzM - [Plugin] ElevenLabs plugin (2026-10-05): ElevenLabs gives your agent a voice, officially. The official plugin bundles skills for text-to-speech, music, sound effects, dubbing and more, plus a hosted MCP server. Sign in with your ElevenLabs account; no API key needed for the hosted server. Source: https://github.com/elevenlabs/plugin · Sample media: https://elevenlabs.io/docs/api-reference/voices/search (ElevenLabs: official preview of the premade voice "Roger") - [Skill] Anthropic canvas-design skill (2026-04-20): Claude gets an art-school recipe card for posters. This skill outputs posters and art as .png or .pdf. It writes a short design philosophy first, then paints it. Original designs only: the skill says never copy existing artists. Source: https://github.com/anthropics/skills/tree/main/skills/canvas-design · Sample media: https://claude.com/blog/skills (Anthropic: Introducing Agent Skills (canvas-design)) - [MCP server] MiniMax MCP (2026-08-20): MiniMax hands your agent a camera, a paintbrush and a voice. Its official server adds text_to_image and generate_video tools, plus voices. Latest update: moved to MCP SDK v2 and retired the Music tool. Needs a MiniMax API key. Source: https://github.com/MiniMax-AI/MiniMax-MCP · Sample media: https://github.com/MiniMax-AI/MiniMax-MCP#4-generate-images (MiniMax (official README)) - [Plugin] Runway plugin (Runway MCP) (2026-09-11): Make Runway images and video without leaving your agent chat. The official plugin connects to Runway's hosted MCP server, so you can ask for images, video, music or speech. Sign in with your Runway account (no API key); generations spend your plan's credits. Source: https://github.com/runwayml/runway-mcp-plugin · Sample media: https://www.youtube.com/watch?v=KhXxcU6cgzM (Runway (official YouTube): Introducing Runway MCP) - [MCP server] Luma API MCP (2025-04-18): Luma lets your MCP client make images and videos. Its official server creates images with Photon and videos with Ray. Needs a Luma API key and a quick setup script. Older pick: last updated April 2025. Source: https://github.com/lumalabs/luma-api-mcp · Sample media: https://www.youtube.com/watch?v=qfWoP7wNVJY (Luma (official YouTube): Introducing Ray 3.2) - [Skill] Anthropic algorithmic-art skill (2026-04-20): Claude can paint generative art with p5.js. The skill covers flow fields, particles and seeded randomness. Original art only, per the skill. Source: https://github.com/anthropics/skills/tree/main/skills/algorithmic-art · Sample media: https://claude.com/blog/skills (Anthropic: Introducing Agent Skills) - [Skill] OpenAI speech skill (2026-02-02): Give your scripts a voice, no recording booth needed. It uses OpenAI's built-in voices to turn text into voiceover, narration or audio prompts. One clip or a whole batch. Needs an OPENAI_API_KEY. Source: https://github.com/openai/skills/tree/main/skills/.curated/speech · Sample media: https://platform.openai.com/docs/guides/text-to-speech (OpenAI (official TTS docs): Alloy voice sample) - [Skill] OpenAI imagegen skill (2026-04-15): Codex brings its own image-making recipe card. It's a built-in skill for mockups, sprites, textures and cutouts. By default it uses Codex's own image tool, so no API key. Source: https://github.com/openai/skills/tree/main/skills/.system/imagegen · Sample media: https://platform.openai.com/docs/guides/image-generation (OpenAI (official Images API docs sample)) - [MCP server] Figma MCP server (2026-10-15): Your agent can read your Figma designs and even draw on them. Drawing on the canvas needs the official remote server. It can even make placeholder images (it asks first and uses Figma AI credits). The latest bump added figma-spec and figma-code-workflows skills. Source: https://github.com/figma/mcp-server-guide · Sample media: https://www.youtube.com/watch?v=s7BZz03icnY (Figma (official YouTube): What is the Figma MCP server?) - [Plugin] Pika plugin (2026-10-14): Pika turns your agent into a tiny media studio. The official plugin makes and edits images, video, music, sound effects and voice. Comes with task skills for ads, headshots, camera moves, podcasts and more. Uses the credits on your Pika account. Source: https://github.com/Pika-Labs/pika-create-plugin · Sample media: https://www.youtube.com/watch?v=zIyjTj2dk10 (Pika Labs (official YouTube): The new Pika platform) ## Daily snacks (full text) # Agent Snacks — Daily snacks ## 2026-11-13 · Claude: Self-hosted Claude Managed Agents, your server **Your agents’ session history can now live on your own server.** Claude Platform’s Nov 12, 2026 release notes launch self-hosted Claude Managed Agents, where you run the whole service and not only the sandboxes. Access goes through the Anthropic account team and isn’t self-serve. Spice: 2 · Source: https://platform.claude.com/docs/en/managed-agents/self-hosted ## 2026-11-12 · Grok Bot: Grok Bot can now join your Meet, Zoom or Teams call **Send your Bot to the meeting and read the recap later.** On Team, Enterprise and Ultra plans, Grok Bot’s Nov 11, 2026 update lets a Bot join a Google Meet, Zoom or Microsoft Teams call from a link, answer in the meeting chat and send a summary afterward. Spice: 2 · Source: https://x.ai/changelog/bot ## 2026-11-11 · Copilot: Dependabot alerts turn off on idle repos Dec 8, 2026 **Dependabot alerts go quiet on repos with no activity for a year**, starting Dec 8, 2026. GitHub announced the change on Nov 10, 2026, and says existing alerts are archived, not deleted. Spice: 2 · Source: https://github.blog/changelog/2026-11-10-dependabot-alerts-will-pause-on-inactive-repositories ## 2026-11-11 · Copilot: Gemini 4 Carbon is now selectable in Copilot **Google’s Gemini 4 Carbon just landed in Copilot’s model picker.** GitHub announced on Nov 10, 2026 that it is generally available, with a rollout that is gradual. Spice: 2 · Source: https://github.blog/changelog/2026-11-10-gemini-4-carbon-in-github-copilot ## 2026-11-10 · Copilot: Secret scanning now checks GitHub Actions workflow logs **A key that slips into a build log can now raise a secret scanning alert.** GitHub announced on Nov 9, 2026 that secret scanning covers GitHub Actions workflow logs, in public preview. Spice: 2 · Source: https://github.blog/changelog/2026-11-09-secret-scanning-detects-secrets-in-github-actions-workflow-logs ## 2026-11-10 · Copilot: Claude Fable 5.2 in Copilot is off until an admin flips it **Claude Fable 5.2 is now in your Copilot model picker, if your plan and admin allow it.** GitHub made it generally available on Nov 9, 2026, for complex, multistep coding tasks across large codebases. Spice: 2 · Source: https://github.blog/changelog/2026-11-09-claude-fable-5-2-in-github-copilot ## 2026-11-09 · Claude Code: Claude Code “Don’t ask again” can save your secrets **Check your project’s .claude/settings.local.json for passwords and keys.** As reported by The Hacker News, Apiiro found 1,122 public copies of that file on GitHub holding at least one secret, saved by “Yes, and don’t ask again” approvals. Spice: 2 · Source: https://apiiro.com/blog/coding-agents-leaking-secrets-github-dont-ask-again ## 2026-11-06 · OpenAI: GPT-6 Cyber joins Daybreak Red for defenders **GPT-6 Cyber is OpenAI’s new model for authorized security research.** The Nov 5, 2026 post says it finds and validates exploitable vulnerabilities, helps develop fixes, and is trained to stay inside the scope researchers set. It starts on the API and Codex for a select group of Daybreak Red customers. Spice: 3 · Source: https://openai.com/index/introducing-gpt-6-cyber ## 2026-11-06 · Cursor: Cursor cloud agents ship shareable live apps **Your cloud agent can now leave localhost with a shareable live app.** Cursor’s Nov 5, 2026 changelog says you describe the app, and it scaffolds the code, can provision Postgres, puts auth behind Cursor sign-in, and hosts it after you publish. Spice: 3 · Source: https://cursor.com/changelog/build-live-apps ## 2026-11-06 · Claude: Claude IP allowlists lock keys to your net **A leaked Claude API key now has fewer places it can run.** Claude Platform’s Nov 5, 2026 release notes add organization IP allowlists in beta, so covered requests from addresses outside your list are refused. You can add up to 1,000 IPv4 or IPv6 addresses or CIDR ranges. Spice: 2 · Source: https://platform.claude.com/docs/en/manage-claude/ip-allowlist ## 2026-11-05 · OpenAI: Images 2.6 adds GIFs and SVG in Codex **Images 2.6 brings GIFs, SVG, and faster turns to Codex.** OpenAI’s Nov 4, 2026 post also puts the release in ChatGPT and the API, with generation about 50% faster than before and GPT-Image-2.6 Flare and Sunburst for developers. Spice: 2 · Source: https://openai.com/index/introducing-images-2-6/ ## 2026-11-05 · Copilot: Legacy GitHub Teams app retires Nov 16 **Action needed**: the legacy GitHub app in Microsoft Teams retires on Nov 16, 2026. GitHub announced this on Oct 23, 2026. Install the consolidated GitHub app (formerly GitHub notifications) so @GitHub and Copilot in Teams keep working. Spice: 3 · Source: https://github.blog/changelog/2026-10-23-turn-notifications-into-code-with-github-copilot-in-teams ## 2026-11-05 · Copilot: Copilot joins Xcode’s coding assistant **GitHub Copilot now sits inside Xcode’s built-in coding assistant.** The public preview, announced Nov 4, 2026, starts with Xcode 27.3 beta, and the older Copilot for Xcode extension stays available as a separate path. Spice: 1 · Source: https://github.blog/changelog/2026-11-04-github-copilot-is-now-available-in-xcodes-coding-assistant ## 2026-11-04 · Copilot: Copilot now runs inside JetBrains Air **Your Copilot plan now doubles as an agent inside JetBrains Air.** Since Nov 3, 2026, you can pick GitHub Copilot in Air’s agent picker, hand it coding tasks in your JetBrains IDE or in the cloud, and review the changes next to your code. Spice: 1 · Source: https://github.blog/changelog/2026-11-03-github-copilot-is-now-available-in-jetbrains-air ## 2026-11-04 · Claude Code: Claude Code warns before your sign-in ends **Your headless Claude Code jobs now get a heads-up before the stored sign-in runs out.** The latest release adds a one-line warning to `claude -p` text output starting 10 days before a stored claude.ai sign-in ends, and shows the end date in `claude auth status`. Spice: 1 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.318 ## 2026-11-03 · Gemini API: Gemini 2.5 Pro and Flash reroute on Dec 1 **Action needed**: Google deprecated `gemini-2.5-pro` and `gemini-2.5-flash` in the Gemini API on Nov 2, 2026. From Dec 1, 2026, requests to them go to newer models automatically and are billed at those models’ rates. Spice: 2 · Source: https://ai.google.dev/gemini-api/docs/changelog ## 2026-11-03 · Claude Code: Claude Code flags extra connectors in /schedule **A scheduled task that quietly picks up connectors now gets called out.** Claude Code’s latest release warns in `/schedule` when a new cloud scheduled task gets connectors you didn’t ask for, and shows how to remove them. Spice: 1 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.317 ## 2026-11-03 · Claude API: Claude API message threads send only new turns **Long agent loops can stop re-uploading the whole conversation every turn.** Message threads, in beta on the Claude Messages API since Nov 2, 2026, keep the conversation on the server so each later request carries only the new messages, with a lower time to first token. Spice: 1 · Source: https://platform.claude.com/docs/en/release-notes/overview ## 2026-11-02 · Claude Code: Opus 4.8 fast mode in Claude Code ends Nov 30 **Action needed**: Claude Code deprecated fast mode for Opus 4.8 as of Oct 30, 2026, and removes it on Nov 30, 2026. After that, fast mode requests on Opus 4.8 return an error and don’t fall back to standard Opus 4.8. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.316 ## 2026-11-02 · Codex: ChatGPT and Codex will share one usage limit in December **Billing change**: starting in December, Plus, Pro and Business plans get one usage allowance shared across ChatGPT and Codex. OpenAI’s Help Center page on the change went up on Oct 31, 2026, and says usage will be measured in tokens. Spice: 2 · Source: https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits ## 2026-10-30 · Claude Code: Claude Code: block edits outside your project **Keep Claude Code’s file edits inside your project folders.** A new setting, `permissions.blockWritesOutsideWorkingDirectories`, makes the Write, Edit and NotebookEdit tools refuse paths outside your working directories. The release came out late on Oct 29, 2026. Spice: 1 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.314 ## 2026-10-30 · Cursor: Cursor Claude Code malware: check SSH keys **Action needed**: a tampered Claude Code extension in Cursor’s extension marketplace tries to plant an SSH backdoor. Cursor’s Oct 29, 2026 notice says versions 2.1.312 and 2.1.313 were not made by Anthropic and were live from about 5:27 a.m. to 4:35 p.m. ET that day. If you installed or updated either in that window, check for the `svc-telemetry` key now. Spice: 3 · Source: https://cursor.com/blog/claude-code-extension-security-notice ## 2026-10-29 · Copilot: Copilot app Projects + computer use preview **Give Copilot a goal that spans repos — or your desktop.** The GitHub Copilot app’s Oct 28, 2026 Universe release puts Projects and computer use in public preview: Projects gather multi-repo or non-repo work in one place, and computer use (macOS and Windows) can click and type in desktop apps after you approve. Spice: 3 · Source: https://github.blog/changelog/2026-10-28-github-copilot-app-universe-releases ## 2026-10-29 · Copilot: GitHub Enterprise now includes Copilot **Billing change**: GitHub Enterprise now includes Copilot for every licensed user. Announced Oct 28, 2026, it’s available now on GitHub Enterprise Cloud for enterprises that work with a GitHub or Microsoft account team, with AI usage metered from the first request against the enterprise budget. Spice: 2 · Source: https://github.blog/changelog/2026-10-28-github-enterprise-now-includes-access-to-github-copilot ## 2026-10-29 · Copilot: Copilot Memory GA: stop repeating your rules **Stop telling Copilot your test command every session.** Copilot Memory became generally available on paid Copilot plans on Oct 28, 2026, so personal preferences and repository conventions carry across sessions and clients. Dreaming, now in public preview, keeps those memories current. Spice: 2 · Source: https://github.blog/changelog/2026-10-28-github-intelligence-puts-your-code-context-and-history-to-work ## 2026-10-28 · Copilot: Copilot Fusion GA picks models for you **No more guessing which model fits the job.** Fusion went generally available on Oct 28, 2026: select Fusion in the model picker, choose Standard, High or Max, and Copilot routes and orchestrates models for each prompt. Spice: 3 · Source: https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion ## 2026-10-28 · Copilot: GitHub Agentic Workflows generally available **Your next GitHub Action can start as a Markdown file.** Agentic Workflows went generally available on Oct 28, 2026. Describe issue triage, CI failure analysis, docs updates or test writing in natural-language Markdown, and they compile into standard Actions YAML. Spice: 3 · Source: https://github.blog/changelog/2026-10-28-github-agentic-workflows-generally-available ## 2026-10-28 · Copilot: Copilot SDK agents get server-to-server auth **Your Copilot SDK agents can stop riding on someone’s login.** In public preview since Oct 27, 2026, a GitHub App installed on your organization can make Copilot requests for the org, with no signed-in user’s credentials. Usage is billed to the org. Spice: 2 · Source: https://github.blog/changelog/2026-10-27-github-copilot-supports-server-to-server-authentication ## 2026-10-28 · Codex: Codex CLI gets a built-in Vertex AI provider **Codex CLI now talks to Google Cloud on its own.** The latest release adds a built-in Google Cloud Vertex AI provider, set with `model_provider = "gcp"`, and puts GPT-6.1 Terra in the model catalogs. Spice: 2 · Source: https://github.com/openai/codex/releases/tag/rust-v0.165.0 ## 2026-10-28 · Claude Code: Claude Code: shorter prompt for custom models **Claude Code quietly shortens the prompt for custom model names.** On Bedrock, Vertex, Foundry and Mantle, names it doesn’t recognize, even ones that point at Sonnet or Haiku, now get the shorter system prompt. Set `CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=0` to keep the full prompt. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.312 ## 2026-10-27 · OpenCode: OpenCode GitHub Action removed: what to do **Action needed**: the OpenCode GitHub Action is no longer published, and the release notes say to remove it from your workflows. The same release, out Oct 27, 2026, also moves the Promise API to the root of `@opencode/ai` and makes TUI plugins require OpenTUI 0.7.0. Spice: 3 · Source: https://github.com/anomalyco/opencode/releases/tag/v2.2.0 ## 2026-10-27 · Copilot: Copilot assisted permissions: fewer prompts **Let Copilot run the clearly safe steps without asking.** Assisted permissions are now generally available in Copilot CLI, the Copilot app and VS Code. A language model checks each request to run a command, access a file path or open a URL, lets clearly safe ones proceed and asks you about the rest. Spice: 2 · Source: https://github.blog/changelog/2026-10-26-github-copilot-assisted-permissions-generally-available ## 2026-10-27 · Claude Code: Claude Code fixes “Usage credits required for 1M context” Update now: **long Claude Code sessions stop dead-ending on a credits error.** Standard-context sessions stopped with “Usage credits required for 1M context” when a request ran over 200K tokens. The release out Oct 26, 2026 compacts the conversation and keeps going instead. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.311 ## 2026-10-26 · OpenAI: GPT-6.1 Terra: OpenAI’s computer-use model at half the price **Agent builders get a cheaper computer-use model**: `gpt-6.1-terra` is in the OpenAI API at $1 per million input tokens, $0.05 cached and $5 output. OpenAI announced it on Oct 26, 2026, and says that’s 50% below GPT-5.6 Terra and GPT-6.1 Sol. It’s also in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Spice: 3 · Source: https://openai.com/index/introducing-gpt-6-1-terra ## 2026-10-26 · Cline: Cline Desktop shows what your agent changed **See every file your agent touched, right in Cline Desktop.** A new Changes button opens a pane listing the files changed in the last turn or across the whole session, with a diff of each. The update also adds a project file tree, a built-in terminal and a Local Models page for Ollama and LM Studio. Spice: 2 · Source: https://github.com/cline/cline/releases/tag/desktop-v0.0.49 ## 2026-10-26 · Cline: Cline shuffles its default models again **Check your model**: Cline’s Oct 26, 2026 updates change default models again. ClinePass now defaults to Kimi K3.1 Pro, and most other providers get new defaults, mostly Kimi K3.1 Pro, MiniMax M3.1 Flash or DeepSeek V4.1 Flash 1017. With no model chosen, the Cline provider uses Claude Sonnet 5.5 again. Spice: 3 · Source: https://github.com/cline/cline/releases/tag/v4.1.26 ## 2026-10-25 · Codex: Keep coding agents out of your Downloads **Coding agent users, stop launching agents from Downloads**: a planted AGENTS.md there told a real customer’s agent to install a malicious npm package. Forever Security caught the attempt and reproduced the attack on Codex and Claude Code. There’s no patch coming, because the agents are doing what they were built to do. Spice: 2 · Source: https://forever.security/blog/hijacking-ai-agents-via-downloads-folder/ ## 2026-10-23 · Grok Bot: Grok Bot signs in without seeing your password **Your agent can now sign in without seeing your password.** On Oct 22, 2026, 1Password launched 1Password for Agents, with Grok Bot among the first integrations. Your agent asks for one saved login, you approve it in 1Password, and 1Password fills it in, so the agent never sees the password or the one-time code. Spice: 3 · Source: https://1password.com/blog/1password-for-ai-agents ## 2026-10-23 · Replit: Replit’s new agent takes your on-call shift **Replit’s Monitoring Agent answers your app’s alarms for you.** Launched on Oct 22, 2026, it can investigate a production issue, make and validate a fix and email you a summary of what happened. Turn on Auto-republish and it can publish the fix too; leave it off to review the change first. Spice: 3 · Source: https://replit.com/blog/replit-monitoring-agent ## 2026-10-23 · OpenAI: OpenAI’s Decisions API leaves beta **Your agent’s small decisions get a fast lane.** OpenAI made the Decisions API generally available on Oct 22, 2026. Built with GPT-6 Luna, it handles questions with bounded answers, and OpenAI says it makes decisions up to 10x faster than GPT-6 Luna through the Responses API. Input costs $0.10 per million tokens, and there are no output-token charges. Spice: 2 · Source: https://openai.com/index/decisions-api ## 2026-10-23 · Copilot: Copilot’s default policy kicks in by Oct 26 **Action needed**: Copilot features you left unconfigured will follow your global default. GitHub began rolling out enforcement of the global default policy for Copilot Business and Copilot Enterprise on Oct 22, 2026, through Oct 26, so timing can differ by enterprise or organization. Choices you made yourself stay as they are. Spice: 2 · Source: https://github.blog/changelog/2026-10-22-global-default-for-copilot-features-generally-available ## 2026-10-23 · Claude: Claude apps are switching to auto mode **Check your Claude mode**: auto mode is becoming the default in most Claude apps. In an update dated Oct 22, 2026, Anthropic said it rolls out next week as the default for new sessions on Pro, Max and Team plans. Instead of asking you to approve each action, an automated check lets safe actions go ahead and blocks risky ones while Claude looks for another route. Spice: 3 · Source: https://claude.com/resources/articles/auto-mode-default-in-claude-code ## 2026-10-22 · OpenAI: Codex and OpenAI models come to Google Cloud **Google Cloud teams can soon buy OpenAI models right where they build.** OpenAI said on Oct 21, 2026 that Codex agentic workflows are coming to Google Cloud’s Model Garden alongside the models, which will be sold through Google Cloud Marketplace. General availability is planned in the coming weeks, and you can request access now. Spice: 1 · Source: https://openai.com/index/openai-models-google-cloud ## 2026-10-22 · Claude: Claude API 2026-10-21: what now returns a 400 error **Claude SDK users, check before you upgrade**: the new major SDK and `ant` CLI versions released on Oct 21, 2026 send API version `2026-10-21` by default. In that version, `temperature`, `top_p` or `top_k` in a request returns a 400 error, and so do document blocks placed directly in assistant messages. The new versions also move from the MIT License to Apache 2.0. Spice: 2 · Source: https://platform.claude.com/docs/en/release-notes/overview ## 2026-10-22 · Claude: Claude Security keeps up with your code **Claude Security now keeps scanning as your code changes.** Announced on Oct 21, 2026: after a deep first scan, it scans new commits based on their risk, marks findings fixed when the fix lands and reopens them if the bug comes back. It also adds an API, webhooks and an MCP connector. It’s in public beta for Claude Enterprise. Spice: 2 · Source: https://claude.com/resources/articles/claude-security-now-scans-continuously-and-tracks-fixes ## 2026-10-21 · Pydantic AI: Pydantic AI patches a High web_fetch bug **Pydantic AI users, update now**: a High-severity fix stops crafted web pages from tying up your app. When `web_fetch_tool` fetched attacker-controlled HTML with many elements, converting it could delay unrelated work in the same process. The maintainers published the fix on Oct 21, 2026: upgrade to 2.59.0, or 1.107.9 on v1. Spice: 3 · Source: https://github.com/pydantic/pydantic-ai/security/advisories/GHSA-4whf-38hw-pxxh ## 2026-10-21 · PHP: Patch PHP now: two critical remote-code bugs **PHP users, update now**: two critical bugs can let an attacker run code on your server without logging in. PHP’s maintainers shipped the fixes on Oct 21, 2026 in 8.5.12, 8.4.27, 8.3.36 and 8.2.35. OpenAI says researchers at Calif found one of the flaws using GPT-5.6 Sol, and Wiz independently reported both. Spice: 3 · Source: https://www.php.net/archive/2026.php#2026-10-21-5 ## 2026-10-21 · Gemini CLI: Untrusted folders keep their Gemini CLI settings **Gemini CLI users, update now**: settings in untrusted folders are now read-only. Before, `gemini mcp add` or `gemini mcp remove` in a folder you hadn’t trusted could remove all existing keys from that project’s `.gemini/settings.json`. Spice: 2 · Source: https://github.com/google-gemini/gemini-cli/releases/tag/v0.65.0 ## 2026-10-21 · Claude: Claude Duet works while you step away **Claude Duet checks your apps and tells you what needs you.** Anthropic’s new app sends a morning brief, takes on chores on its own cloud computer while you’re away, and asks before it sends, pays, signs or deletes anything. It’s a research preview by waitlist for Pro and Max subscribers. Spice: 3 · Source: https://claude.com/resources/articles/claude-duet ## 2026-10-21 · Claude Code: Claude Code fixes its Bedrock and Vertex snag **Bedrock and Vertex users, update now**: Claude Code’s failing requests are fixed. The previous release could fail every request on Vertex, Bedrock and other providers when the endpoint or a gateway rejected tool search. Claude Code now retries without it. Spice: 3 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.305 ## 2026-10-20 · Pydantic AI: Pydantic AI keeps MCP tokens out of traces **Pydantic AI users, update now**: two Medium security fixes are out, one for MCP credentials that could reach OpenTelemetry traces and one for the Vercel AI and AG-UI adapters. Upgrade to 2.58.0, or 1.107.8 on v1, then restrict access to old traces and rotate any credentials they exposed. Spice: 3 · Source: https://github.com/pydantic/pydantic-ai/releases/tag/v2.58.0 ## 2026-10-20 · Copilot: Six Copilot models just left the picker **Action needed**: six Copilot models were deprecated on Oct 19, 2026, including GPT-5.5, GPT-5.4 and Grok 4.5. Move workflows and integrations to the suggested alternatives, such as GPT-6.1 Sol, GPT-6 Luna, Gemini 3.8 Flash and Grok 4.7. Spice: 2 · Source: https://github.blog/changelog/2026-10-19-github-copilot-models-deprecated-on-october-19-2026 ## 2026-10-20 · Copilot: Script your Copilot model policies **Set Copilot model rules with a script, not a settings page.** GET and PATCH REST endpoints are now generally available to read and update Copilot model policies for enterprises, enterprise teams, enterprise apps and organizations. Spice: 1 · Source: https://github.blog/changelog/2026-10-19-copilot-model-policy-api-for-enterprise-orgs-and-teams ## 2026-10-20 · Claude Code: Claude Code warns before a costly resume **Back to an old chat after a break?** On Pro and Max plans, Claude Code now warns you when resuming an inactive conversation would use a noticeable share of your 5-hour usage limit. Resume anyway, or start a new conversation and Claude can look up the previous one if you refer to it. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.304 ## 2026-10-19 · Cline: Cline CLI and Desktop default to Sonnet again **Check your model**: the Cline provider’s default in Cline CLI and Desktop is Claude Sonnet 5.5 again. It’s back at the top of Cline’s recommended list. Also fixed: GPT-6 models now work on OpenAI and Azure OpenAI-compatible endpoints, where every request used to be rejected. Spice: 3 · Source: https://github.com/cline/cline/releases/tag/cli-v3.0.73 ## 2026-10-19 · Claude: Standalone Claude Design closes Dec 14 **Action needed**: move your design systems out of standalone Claude Design before Dec 14, 2026. That’s when claude.ai/design closes and starts sending you to Claude, where Claude Design now lives and new designs are created as artifacts. Save any chats and comments you need first, because they don’t carry over. Spice: 3 · Source: https://support.claude.com/en/articles/17440474-migrate-from-standalone-claude-design-to-claude ## 2026-10-19 · Claude Code: Claude Code: switch off idle compaction **Want long chats left alone while you’re away?** A new Idle compaction row in `/config` turns that feature off or back on, and the notice after an idle compaction now says how. Also changed: fork subagents are on by default in `claude -p`, the Agent SDK, the desktop app and VS Code, as they already were in the terminal. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.302 ## 2026-10-16 · Cline: Unpinned Cline setups got a new default model **Check your model**: unpinned Cline provider setups now default to Gemini 4 Argon. OpenRouter stays on Claude Sonnet 5.5, but over 20 other providers change defaults too, so pin the model you want to keep. Also new: Cloud sessions, rolling out gradually, run a job in a hosted sandbox on a GitHub repo and branch you pick and keep going after you close the editor. Spice: 3 · Source: https://github.com/cline/cline/releases/tag/v4.1.25 ## 2026-10-16 · Claude Code: Claude Code chats fit smaller context windows Switch to a model with a smaller context window and Claude Code makes the chat fit. A conversation too long for the new model is now summarized by the previous model first, and a fix covers a case where that summary could spend usage credits without asking. Unattended retry mode can now also give up on a usage limit that resets too far away. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.301 ## 2026-10-15 · Copilot: Copilot retires three models Nov 16 **Action needed**: three Copilot models retire Nov 16, 2026. Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra leave every Copilot experience. Move to Claude Opus 5.5 or GPT-6.1 Sol (or GPT-6 Luna) before your workflows find out the hard way. Individual annual plans keep Opus 4.8. Spice: 3 · Source: https://github.blog/changelog/2026-10-15-upcoming-deprecation-of-selected-github-copilot-models-in-mid-november ## 2026-10-15 · Copilot: Copilot will now fix your security warnings for you Copilot can now fix your security alerts and prove it. Assign a CodeQL alert and it patches the code, re-runs CodeQL to prove the alert is gone, then opens a draft PR. Up to 25 alerts per batch. Now GA. Default-branch alerts only: an enterprise admin has to turn on the “Agentic autofix” policy, and each run spends Code Security AI Credits + Actions minutes. You can also start it via the REST API (`copilot-swe-agent[bot]`). Spice: 2 · Source: https://github.blog/changelog/2026-10-15-agentic-autofix-for-code-scanning-alerts-general-availability ## 2026-10-15 · Copilot: Copilot in Slack and Teams now works across multiple repositories Ask about several repos at once without leaving Slack or Teams. Mention @GitHub to cover open PRs or issues across repos in one conversation. You can now mention it anywhere in your message, and in Slack it can read shared messages and threads as context. Spice: 1 · Source: https://github.blog/changelog/2026-10-15-copilot-in-slack-and-teams-adds-model-and-repository-updates ## 2026-10-15 · Codex: Codex lets you cut in mid-answer Change your mind mid-answer and Codex listens. In the latest Codex CLI, send a follow-up while it's still answering and it steers right away. New `/config` searches every setting, and independent read-only tools now run in parallel. Spice: 2 · Source: https://github.com/openai/codex/releases/tag/rust-v0.163.0 ## 2026-10-14 · Copilot: HydraFusion stops making you pick a model (and adds a “Max” mode) HydraFusion gets a bigger model pool and a Max tier. Its model pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. Cascade gets a judging step; Critique can tool-snoop the workspace with parallel critics. New **HydraFusion Max** is the “go deep” button. Lives in VS Code, the Copilot app, and Copilot CLI. Spice: 1 · Source: https://github.blog/changelog/2026-10-14-hydrafusion-updates-and-new-max-option ## 2026-10-14 · Claude Code: Claude Code: idle chats now compact before the cache expires Walk away from a long chat and Claude Code tidies it up. After ~54 idle minutes, a 200K+ token chat compacts before the prompt cache expires, so your next message doesn't re-chew the whole transcript. Tune the threshold with `CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS`, or opt out with `"idleCompaction": false`. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.299 ## 2026-10-13 · Gemini CLI: Gemini CLI respects folder trust in headless runs Update now: Gemini CLI 0.64.0 carries folder trust into headless mode, so scripts and CI follow the same trust rules as a person at the keyboard. File tool writes are also atomic now, and saved state recovers from a backup if it gets corrupted. Spice: 2 · Source: https://github.com/google-gemini/gemini-cli/releases/tag/v0.64.0 ## 2026-10-13 · Cursor: Cursor Origin opens the snack aisle Cursor’s Origin now plugs into the apps you already use. That’s 20+ apps: Cloudflare, Vercel, Buildkite, CircleCI, CodeRabbit, Greptile, Semgrep, Socket, Sentry, Linear, and friends. Public submissions welcome. Apps auth with app JWT + install tokens, get webhooks, write check runs / PR comments / reviews. Spice: 1 · Source: https://cursor.com/changelog/origin-app-marketplace ## 2026-10-13 · Copilot CLI: Copilot CLI: hooks that don’t hang the babysitter Copilot CLI Autopilot no longer stalls on a hook's “ask” when nobody is watching. - **One more folder:** `/add-dir` grants sandbox access for the session; interactive settings nudge you to mask likely secrets. - **No more stalled Autopilot:** agent-scoped declarative hooks; Autopilot now **denies** a preToolUse hook's “ask” instead of stopping on a permission dialog nobody is watching. - **New model, easier answers:** Gemini 4 Argon support; `/ask` expands with Ctrl+E, copies with Ctrl+Y. Spice: 3 · Source: https://github.com/github/copilot-cli/releases/tag/v1.0.97 ## 2026-10-13 · Cline: Cline swapped your default model **Check your model**: unpinned Cline setups just got a new default. Cline + OpenRouter users now land on Claude Sonnet 5.5. About 20 other providers also shuffle defaults, so unpinned users will wake up on a different model. Pin the model you want to keep. Spice: 3 · Source: https://github.com/cline/cline/releases/tag/v4.1.24 ## 2026-10-10 · OpenCode: OpenCode 1.19: the on-ramp to OpenCode 2 (and auto-update is off) **Action needed**: OpenCode 2 is here, and updates are no longer automatic. `opencode upgrade --major` (or `/update` in the TUI) walks you there, so upgrade on purpose. It also allows OpenAI’s `ultrafast` service tier. Spice: 2 · Source: https://github.com/anomalyco/opencode/releases/tag/v1.19.0 ## 2026-10-09 · Claude Code: Claude Code subagents get their own memory budget Give each subagent its own memory budget. The latest Claude Code release adds `autoCompactWindow` to subagent definitions, so a subagent can compact earlier than the main chat. A new variable runs every workflow agent on one model. Spice: 1 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.296 ## 2026-10-08 · Codex: Codex sets up side workspaces for you Run side tasks without the Git gymnastics. The new Codex CLI release can create and list managed Git worktrees from trusted local projects once the worktrees feature is on. It also lets you pin tasks in the agent Command Center with `p`. Spice: 1 · Source: https://github.com/openai/codex/releases/tag/rust-v0.162.0 ## 2026-10-08 · Claude Code: Claude Code hooks can now fail safe A broken safety hook no longer waves the action through. The latest Claude Code release adds `onFailure: "block"` for command and HTTP hooks: a hook that can't start, times out or exits with an unexpected code now blocks the action. Spice: 1 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.295 ## 2026-10-07 · Codex: Codex quietly moved you to GPT-6.1 Sol **Check your model**: Codex now defaults to GPT-6.1 Sol. The new Codex CLI release makes it the default in the bundled and Amazon Bedrock catalogs. You can also sign in to MCP servers from a session with `/mcp login `. Spice: 2 · Source: https://github.com/openai/codex/releases/tag/rust-v0.161.0 ## 2026-10-06 · Cursor: Check on your Cursor agents from your phone Step away from your desk without losing track of your agents. The Cursor iOS app now shows the local agents running on your computer and lets you reply to them. It is on by default, except for Enterprise organizations. Spice: 1 · Source: https://cursor.com/changelog/remote-control-local-agents ## 2026-10-03 · Claude Code: Claude Code closes gaps in your deny rules Your deny rules stop slipping through side doors. The latest Claude Code release applies Bash deny and ask rules behind environment variable prefixes, holds them on nested parts of compound commands on managed machines, and applies Read deny rules to files reached through a symlink. Spice: 2 · Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.289 ## 2026-10-02 · Copilot: Four Copilot models just left the menu **Action needed**: four Copilot models are gone as of Oct 2, 2026. Suggested replacements: Gemini 3.8 Flash for Gemini 3.5 Flash and Gemini 3.6 Flash, Kimi K3 for Kimi K2.7 Code, and Claude Opus 5.5 for Claude Opus 4.7. Update workflows that name them. Spice: 2 · Source: https://github.blog/changelog/2026-10-02-selected-models-in-github-copilot-deprecated ## Snack Pack #5 — full text (https://agentsnacks.ai/packs/005/) # Snack Pack #5: Mark the dates before the switches flip Week of Nov 5–11, 2026 · 7 snacks · ~8 min **Two dates are closing in on your GitHub setup.** The legacy GitHub app in Microsoft Teams retires on Nov 16, Cursor cloud agents can now publish shareable live apps and OpenAI opens GPT-6 Cyber to a select group of Daybreak Red defenders. Dependabot alerts pause on idle repositories on Dec 8, while Claude Code’s “Don’t ask again” can save secrets, Claude adds IP allowlists and Copilot offers Claude Fable 5.2. ``` NUTRITION FACTS · Serving: 1 Snack Pack Snacks ................ 7 First Bite ............ 1 Main Course ........... 1 · 11-min read Secret Sauce .......... 1 Video Tasting ......... 1 New on the Shelf ...... 3 Before you go ......... 1 (Tiny bites worth a glance) Source links .......... 18 🌶️🌶️🌶️ items .......... 3 of 7 Actually useful ~100% ... 7 of 7 Counts from this issue · 15 cards · Hype: 0 g. ``` **Here’s what changed in your agent tools.** Every bite links to its source. No funding fluff. No “AI will change everything.” Just what shipped, what changed, and why it might matter to you. A fresh pack lands every Thursday. --- ## 1. [Legacy GitHub Teams app retires Nov 16](https://agentsnacks.ai/daily/2026-11-05-github-teams-legacy-app-retires/) 🌶️🌶️🌶️ *Deadline Nov 16 for Copilot in Teams* **Action needed**: the legacy GitHub app in Microsoft Teams retires on Nov 16, 2026. GitHub announced this on Oct 23, 2026. Install the consolidated GitHub app (formerly GitHub notifications) so @GitHub and Copilot in Teams keep working. Who cares: Anyone still on the legacy GitHub app in Microsoft Teams who uses Copilot there. Source: [GitHub Changelog: Turn notifications into code with GitHub Copilot in Teams, Oct 23, 2026](https://github.blog/changelog/2026-10-23-turn-notifications-into-code-with-github-copilot-in-teams) --- ## 2. [Cursor cloud agents ship shareable live apps](https://agentsnacks.ai/daily/2026-11-06-cursor-build-live-apps/) 🌶️🌶️🌶️ *From prompt to a live link teammates can open* **Your cloud agent can now leave localhost with a shareable live app.** Cursor’s Nov 5, 2026 changelog says you describe the app, and it scaffolds the code, can provision Postgres, puts auth behind Cursor sign-in, and hosts it after you publish. Who cares: Paid Cursor users with cloud agents who need internal dashboards, portals, or microsites without a separate deploy stack. Source: [Cursor Changelog: Build Live Apps in Cursor, Nov 5, 2026](https://cursor.com/changelog/build-live-apps) · [Cursor docs: Cloud Agent Apps](https://cursor.com/docs/cloud-agent/apps) --- ## 3. [GPT-6 Cyber joins Daybreak Red for defenders](https://agentsnacks.ai/daily/2026-11-06-openai-gpt-6-cyber/) 🌶️🌶️🌶️ *A cyber model built to stay inside your scope* **GPT-6 Cyber is OpenAI’s new model for authorized security research.** The Nov 5, 2026 post says it finds and validates exploitable vulnerabilities, helps develop fixes, and is trained to stay inside the scope researchers set. It starts on the API and Codex for a select group of Daybreak Red customers. Who cares: Approved Daybreak Red security teams who already use OpenAI for authorized vulnerability research, triage, and patch work. Source: [OpenAI: Introducing GPT-6 Cyber, Nov 5, 2026](https://openai.com/index/introducing-gpt-6-cyber) · [OpenAI API changelog, Nov 5, 2026](https://developers.openai.com/api/docs/changelog) --- ## 4. [Dependabot alerts turn off on idle repos Dec 8, 2026](https://agentsnacks.ai/daily/2026-11-11-copilot-dependabot-alerts-inactive-repos/) 🌶️🌶️ *A year of silence switches alerts off* **Dependabot alerts go quiet on repos with no activity for a year**, starting Dec 8, 2026. GitHub announced the change on Nov 10, 2026, and says existing alerts are archived, not deleted. Who cares: Anyone with older repositories on github.com, including agent side projects. Source: [GitHub Changelog: Dependabot alerts will pause on inactive repositories, Nov 10, 2026](https://github.blog/changelog/2026-11-10-dependabot-alerts-will-pause-on-inactive-repositories) --- ## 5. [Claude Code “Don’t ask again” can save your secrets](https://agentsnacks.ai/daily/2026-11-09-claude-code-dont-ask-again-secrets/) 🌶️🌶️ *A saved approval can carry a password* **Check your project’s .claude/settings.local.json for passwords and keys.** As reported by The Hacker News, Apiiro found 1,122 public copies of that file on GitHub holding at least one secret, saved by “Yes, and don’t ask again” approvals. Who cares: Anyone who clicks “Yes, and don’t ask again” in Claude Code on a command that has a password, token or key typed into it. Source: [Apiiro research, Nov 6, 2026](https://apiiro.com/blog/coding-agents-leaking-secrets-github-dont-ask-again) · [The Hacker News write-up of the Apiiro research, Nov 7, 2026](https://thehackernews.com/2026/11/claude-codes-dont-ask-again-file.html) · [Claude Code docs: Settings](https://code.claude.com/docs/en/settings) --- ## 6. [Claude IP allowlists lock keys to your net](https://agentsnacks.ai/daily/2026-11-06-claude-ip-allowlist/) 🌶️🌶️ *A leaked key stops working off your IPs* **A leaked Claude API key now has fewer places it can run.** Claude Platform’s Nov 5, 2026 release notes add organization IP allowlists in beta, so covered requests from addresses outside your list are refused. You can add up to 1,000 IPv4 or IPv6 addresses or CIDR ranges. Who cares: Claude Console and Claude Enterprise admins who want to limit where API keys or stolen sign-ins still work. Source: [Claude Platform release notes, Nov 5, 2026](https://platform.claude.com/docs/en/release-notes/overview) · [Claude docs: IP allowlisting](https://platform.claude.com/docs/en/manage-claude/ip-allowlist) --- ## 7. [Claude Fable 5.2 in Copilot is off until an admin flips it](https://agentsnacks.ai/daily/2026-11-10-copilot-claude-fable-5-2/) 🌶️🌶️ *A new top model, gated by an admin switch* **Claude Fable 5.2 is now in your Copilot model picker, if your plan and admin allow it.** GitHub made it generally available on Nov 9, 2026, for complex, multistep coding tasks across large codebases. Who cares: Copilot Pro+, Max, Business and Enterprise users, and the admins who set Copilot policies. Source: [GitHub Changelog: Claude Fable 5.2 in GitHub Copilot, Nov 9, 2026](https://github.blog/changelog/2026-11-09-claude-fable-5-2-in-github-copilot) --- ## First Bite 🍪 **[What’s a secret, and what do you do when one leaks?](https://agentsnacks.ai/first-bite/#fb-005)** **A secret is any credential that proves it’s you, such as an API key, a password or a token.** GitHub’s docs say that when secret scanning finds one exposed, the advice is to rotate the affected credential right away, because scrubbing it from Git history is time-intensive and often unnecessary once the credential is revoked ([GitHub Docs: About secret scanning](https://docs.github.com/en/code-security/secret-scanning/introduction/about-secret-scanning)). Try this: if you have ever pasted a key into a command, a chat or a config file, read [our Claude Code Daily](https://agentsnacks.ai/daily/2026-11-09-claude-code-dont-ask-again-secrets/) to see one place it can end up. New here? [Start here](https://agentsnacks.ai/start/). --- ## The Main Course 🍽️ — [When Opus isn’t enough, Claude Fable 5.2 is the next step up](https://agentsnacks.ai/main-course/016/) **Claude Fable 5.2 is the model Anthropic wants you to reach for when Opus 5.5 still isn’t enough.** It shipped on Nov 9, 2026, costs less than Fable 5.1, and generates outputs up to 30% faster, with an optional fast mode on some platforms. Four API changes can turn a request that used to work into a 400 error, so don’t just swap the model id. Full Main Course: 11-min read [Read the full Main Course →](https://agentsnacks.ai/main-course/016/) Sources: [Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2](https://www.anthropic.com/claude-fable-and-mythos-5-2) · [Claude docs: Fable 5.2 overview](https://platform.claude.com/docs/en/models/fable-5-2/overview) · [Claude docs: What’s new in Claude Fable 5.2](https://platform.claude.com/docs/en/models/fable-5-2/whats-new-fable-5-2) · [system card](https://www.anthropic.com/document/claude-fable-5-2-mythos-5-2-system-card) · [Claude Help Center: Claude Fable 5.2 launch offers](https://support.claude.com/en/articles/17506494-claude-fable-5-2-launch-offers) · [@claudeai on X, Nov 9, 2026](https://x.com/claudeai/status/2119867138987176141) · [@claudeai on X, Nov 9, 2026](https://x.com/claudeai/status/2119867129751056485) --- ## Secret Sauce 🧂 — [Put the riskiest Dependabot alerts first](https://agentsnacks.ai/secret-sauce/#ss-005) **Try it now:** 1. In a repository, open the Security and quality tab, then pick Vulnerabilities under Dependabot in the Findings sidebar. 2. Keep the default Most important sort, which weighs signals such as CVSS score and dependency scope, or type `scope:development` in the search bar to see which alerts only touch development dependencies. 3. Open an alert and use the Assignees menu to give it an owner, which can be a person, a team or an AI agent. Source: [GitHub Docs: Viewing and updating Dependabot alerts](https://docs.github.com/en/code-security/dependabot/dependabot-alerts/viewing-and-updating-dependabot-alerts) --- ## Video Tasting ▶️ [Your apps have moved into Claude](https://agentsnacks.ai/videos/#v-Gy1jSeSzAhk) Beginner · 1:00 · Claude · Nov 10, 2026 Video Summary: - **Interactive connectors open the full version of a tool inside Claude.** - The demo adds a summary to a Confluence launch plan, applies a dither effect to a photo in Adobe, and checks a Box floor plan against a design standard. - Make changes yourself, or ask Claude to make them with you. Rolling out on web and desktop. On screen: Claude chat with Confluence, Adobe and Box opening as live panels next to the conversation. After watching you can: Know what interactive connectors look like in Claude and which kinds of edits you can ask for. Short on time? Watch just: - [▶ 0:11](https://www.youtube.com/watch?v=Gy1jSeSzAhk&t=11s) Confluence summary - [▶ 0:22](https://www.youtube.com/watch?v=Gy1jSeSzAhk&t=22s) Adobe dither effect - [▶ 0:32](https://www.youtube.com/watch?v=Gy1jSeSzAhk&t=32s) Box floor plan check Source: [YouTube: Your apps have moved into Claude](https://www.youtube.com/watch?v=Gy1jSeSzAhk) --- ## New on the Shelf 🛒 - [MCP reference servers](https://agentsnacks.ai/shelf/#shelf-005-1) 🌶️🌶️🌶️ · MCP server **Update before your agent reads another repository.** The latest reference servers release fixes six vulnerabilities across the filesystem, Git and fetch servers, including a repository’s own Git config making a read-only call run a program. If you run a date version of the filesystem server, install 2.3.0 by name, because package managers won’t offer it as an upgrade. Works with: Any MCP client Install: `npm install @modelcontextprotocol/server-filesystem@2.3.0` (also mcp-server-git and mcp-server-fetch 2026.11.11) Source: [MCP reference servers v2.4.0 release notes](https://github.com/modelcontextprotocol/servers/releases/tag/v2.4.0) - [Context7 MCP](https://agentsnacks.ai/shelf/#shelf-005-2) 🌶️ · MCP server **Let your agent read a library’s source at the version its docs describe.** Upstash’s Context7 adds a query-code tool and a get-upgrade-guide tool that compares two versions, both in a research preview that stays hidden until you switch it on. Works with: Any MCP client Install: `npx ctx7 setup` Source: [Context7 MCP v4.6.0 release notes](https://github.com/upstash/context7/releases/tag/%40upstash/context7-mcp%404.6.0) - [MongoDB MCP server](https://agentsnacks.ai/shelf/#shelf-005-3) 🌶️ · MCP server **If you choose read-only in setup, that choice now sticks for every client.** MongoDB’s official server writes the --readOnly flag into the configs it generates for Cursor, VS Code, Claude Code and others, dials a preconfigured connection on startup and accepts named connection strings. Works with: Any MCP client Install: `npx -y mongodb-mcp-server@latest setup` Source: [mongodb-mcp-server v3.3.0 release notes](https://github.com/mongodb-js/mongodb-mcp-server/releases/tag/v3.3.0) --- ## Before you go 🥡 **[Most of this week’s news came with a switch attached](https://agentsnacks.ai/before-you-go/#byg-005)** - **Four more Dailies are worth a catch-up.** Secret scanning now reads [GitHub Actions workflow logs](https://agentsnacks.ai/daily/2026-11-10-github-secret-scanning-actions-logs/), [Gemini 4 Carbon](https://agentsnacks.ai/daily/2026-11-11-copilot-gemini-4-carbon/) joined Copilot’s model picker, [Images 2.6](https://agentsnacks.ai/daily/2026-11-05-openai-images-2-6/) added GIFs and SVG in Codex, and [Copilot reached Xcode’s coding assistant](https://agentsnacks.ai/daily/2026-11-05-copilot-xcode-coding-assistant/) in preview. - Two Main Courses from this week are also worth your time. [Let your dot borrow a computer](https://agentsnacks.ai/main-course/017/) covers what one Allow hands over to a ChatGPT dot, and [FlashProve](https://agentsnacks.ai/main-course/013/) asks how you would prove which model answered. --- That’s the pack. Pass a snack to a friend. ## Snack Pack #4 — full text (https://agentsnacks.ai/packs/004/) # Snack Pack #4: Check what your editor let in Week of Oct 30–Nov 5, 2026 · 7 snacks · ~8 min **One bad extension can leave a key behind.** Cursor pulled a fake Claude Code build from its marketplace after it tried to plant an SSH backdoor, while Claude Code, the Gemini API and OpenAI put dates on fast mode, older models and a shared ChatGPT and Codex allowance. Claude Code also fences in file edits, and Copilot moves into JetBrains Air. ``` NUTRITION FACTS · Serving: 1 Snack Pack Snacks ................ 7 First Bite ............ 1 Main Course ........... 1 · 7-min read Secret Sauce .......... 1 Video Tasting ......... 1 New on the Shelf ...... 2 Before you go ......... 1 (Tiny bites worth a glance) Source links .......... 14 🌶️🌶️🌶️ items .......... 1 of 7 Actually useful ~57% ... 4 of 7 Counts from this issue · 14 cards · Hype: 0 g. ``` **Here’s what changed in your agent tools.** Every bite links to its source. No funding fluff. No “AI will change everything.” Just what shipped, what changed, and why it might matter to you. A fresh pack lands every Thursday. --- ## 1. [Cursor Claude Code malware: check SSH keys](https://agentsnacks.ai/daily/2026-10-30-cursor-claude-code-extension-malware/) 🌶️🌶️🌶️ *A backdoor key may still be on disk* **Action needed**: a tampered Claude Code extension in Cursor’s extension marketplace tries to plant an SSH backdoor. Cursor’s Oct 29, 2026 notice says versions 2.1.312 and 2.1.313 were not made by Anthropic and were live from about 5:27 a.m. to 4:35 p.m. ET that day. If you installed or updated either in that window, check for the `svc-telemetry` key now. Who cares: Anyone who installed or updated the Claude Code extension from Cursor’s marketplace on Oct 29, 2026, and the people who look after their machines. Source: [Cursor security notice: malicious Anthropic.claude-code 2.1.312 and 2.1.313 in the Cursor Extension Marketplace, Oct 29, 2026](https://cursor.com/blog/claude-code-extension-security-notice) --- ## 2. [Opus 4.8 fast mode in Claude Code ends Nov 30](https://agentsnacks.ai/daily/2026-11-02-claude-code-opus-4-8-fast-mode-deprecated/) 🌶️🌶️ *Fast mode on Opus 4.8 will error* **Action needed**: Claude Code deprecated fast mode for Opus 4.8 as of Oct 30, 2026, and removes it on Nov 30, 2026. After that, fast mode requests on Opus 4.8 return an error and don’t fall back to standard Opus 4.8. Who cares: Claude Code users who run Opus 4.8 in fast mode. Source: [Claude Code v2.1.316 release notes, Oct 30, 2026](https://github.com/anthropics/claude-code/releases/tag/v2.1.316) --- ## 3. [Gemini 2.5 Pro and Flash reroute on Dec 1](https://agentsnacks.ai/daily/2026-11-03-gemini-2-5-pro-flash-deprecated/) 🌶️🌶️ *Old model names reroute and rebill* **Action needed**: Google deprecated `gemini-2.5-pro` and `gemini-2.5-flash` in the Gemini API on Nov 2, 2026. From Dec 1, 2026, requests to them go to newer models automatically and are billed at those models’ rates. Who cares: Developers whose apps or agents call gemini-2.5-pro or gemini-2.5-flash in the Gemini API. Source: [Gemini API release notes, Nov 2, 2026](https://ai.google.dev/gemini-api/docs/changelog) · [Google docs: Gemini API deprecation schedule](https://ai.google.dev/gemini-api/docs/deprecations) --- ## 4. [ChatGPT and Codex will share one usage limit in December](https://agentsnacks.ai/daily/2026-11-02-chatgpt-codex-shared-usage-limits/) 🌶️🌶️ *One allowance for chat and code* **Billing change**: starting in December, Plus, Pro and Business plans get one usage allowance shared across ChatGPT and Codex. OpenAI’s Help Center page on the change went up on Oct 31, 2026, and says usage will be measured in tokens. Who cares: ChatGPT Plus, Pro and Business users who also code with Codex. Source: [OpenAI Help Center: understanding ChatGPT and Codex usage limits, Oct 31, 2026](https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits) --- ## 5. [Claude Code: block edits outside your project](https://agentsnacks.ai/daily/2026-10-30-claude-code-block-writes-outside-folders/) 🌶️ *Edits stay inside your folders* **Keep Claude Code’s file edits inside your project folders.** A new setting, `permissions.blockWritesOutsideWorkingDirectories`, makes the Write, Edit and NotebookEdit tools refuse paths outside your working directories. The release came out late on Oct 29, 2026. Who cares: Claude Code users who want file edits fenced to the project they’re working in. Source: [Claude Code v2.1.314 release notes, Oct 29, 2026](https://github.com/anthropics/claude-code/releases/tag/v2.1.314) --- ## 6. [Claude API message threads send only new turns](https://agentsnacks.ai/daily/2026-11-03-claude-api-message-threads/) 🌶️ *Smaller requests in long agent loops* **Long agent loops can stop re-uploading the whole conversation every turn.** Message threads, in beta on the Claude Messages API since Nov 2, 2026, keep the conversation on the server so each later request carries only the new messages, with a lower time to first token. Who cares: Developers who run long conversations or agent loops on the Claude Messages API. Source: [Claude Platform release notes, Nov 2, 2026](https://platform.claude.com/docs/en/release-notes/overview) · [Claude docs: how message threads work](https://platform.claude.com/docs/en/build-with-claude/message-threads) --- ## 7. [Copilot now runs inside JetBrains Air](https://agentsnacks.ai/daily/2026-11-04-copilot-jetbrains-air/) 🌶️ *Your Copilot plan as an Air agent* **Your Copilot plan now doubles as an agent inside JetBrains Air.** Since Nov 3, 2026, you can pick GitHub Copilot in Air’s agent picker, hand it coding tasks in your JetBrains IDE or in the cloud, and review the changes next to your code. Who cares: JetBrains IDE users who already pay for GitHub Copilot, plus Air Teams admins. Source: [GitHub Changelog: GitHub Copilot is now available in JetBrains Air, Nov 3, 2026](https://github.blog/changelog/2026-11-03-github-copilot-is-now-available-in-jetbrains-air) --- ## First Bite 🍪 **[What’s an authorized_keys file?](https://agentsnacks.ai/first-bite/#fb-004)** **SSH signs you in with a key pair instead of a password.** The private key stays with you and works like a password, and the public key goes on the machine you want to reach, in a plain text file called `authorized_keys` in your `.ssh` folder (administrator accounts on Windows use `administrators_authorized_keys` instead). When someone connects, the SSH server checks their private key against the public keys it has, so a key in that file lets whoever holds its private half sign in ([Microsoft Learn: key-based authentication in OpenSSH for Windows](https://learn.microsoft.com/en-us/windows-server/administration/openssh/openssh_keymanagement)). Try this: if you have an `authorized_keys` file, open it and make sure you recognize every key in it, then read [our Cursor Daily](https://agentsnacks.ai/daily/2026-10-30-cursor-claude-code-extension-malware/) for the key name to look for. New here? [Start here](https://agentsnacks.ai/start/). --- ## The Main Course 🍽️ — [Your ChatGPT chats are about to share a meter with Codex](https://agentsnacks.ai/main-course/012/) **Starting in December, ChatGPT and Codex share one allowance.** One meter will pay for your ChatGPT chats and your Codex runs, and plenty of people are nervous about running dry halfway through a task. OpenAI’s own help pages already answer most of the worry. They say what draws from that allowance, what never touches it, where to see what’s left and what you can do when it runs out. Here’s that map, with no guesses about how fast anyone’s allowance will drain. Full Main Course: 7-min read [Read the full Main Course →](https://agentsnacks.ai/main-course/012/) Sources: [OpenAI Help Center: understanding ChatGPT and Codex usage limits](https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits) · [OpenAI Help Center: using Codex with your ChatGPT plan](https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan) · [OpenAI Help Center: using credits in ChatGPT personal plans](https://help.openai.com/en/articles/12642688-using-credits-for-flexible-usage-in-chatgpt-personal-plans) · [OpenAI Help Center: paid Work and Codex rate limit resets](https://help.openai.com/en/articles/20001507-paid-weekly-work-and-codex-rate-limit-resets) --- ## Secret Sauce 🧂 — [Decide when your VS Code extensions update](https://agentsnacks.ai/secret-sauce/#ss-004) **Try it now:** 1. In a terminal, run `code --list-extensions --show-versions` to see every extension VS Code has installed and which version you’re on. 2. To update by hand instead, run the Disable Auto Update for Extensions command (or change the `extensions.autoUpdate` setting), and run Enable Auto Update for Extensions to switch it back on. 3. To go back to a release you trust, right-click the extension in the Extensions view, choose Install Another Version and pick one from the list. Source: [Visual Studio Code docs: Extension Marketplace](https://code.visualstudio.com/docs/configure/extensions/extension-marketplace) --- ## Video Tasting ▶️ [Claude’s new mobile app updates are here](https://agentsnacks.ai/videos/#v-SkpwzfR4A1E) Beginner · 2:32 · Claude · Oct 29, 2026 Video Summary: - **You can stay signed in to two Claude accounts on your phone.** Switch between work and personal without signing out, and photos also show up in the chat as soon as you pick them. - On iPhone Duo and iPad, a chart or anything else Claude makes opens in its own column beside the chat, so you can read it and ask for changes at the same time. Chromebooks and foldables get a resizable sidebar, right-click and the Claude Desktop keyboard shortcuts. - Voice mode answers sooner, hears you better in a noisy room and works with Apple CarPlay on iPhone. Menus and settings now come in 33 languages, and pull-down search is on the way for iPhone and iPad (it’s already on web and desktop). On screen: A presenter talks between screen recordings and title cards. A chart opens beside the chat on a folding iPhone, an account switcher lists Work and Personal, Android runs in a laptop-style layout on a Chromebook, voice mode listens on an Apple CarPlay screen, the app menus appear in other languages, and a preview shows pull-down search on iPhone. After watching you can: Keep your work and personal Claude accounts signed in on one phone, ask Claude something through CarPlay on a drive, and switch the app’s menus to your own language. Short on time? Watch just: - [▶ 0:11](https://www.youtube.com/watch?v=SkpwzfR4A1E&t=11s) Two accounts in one app - [▶ 0:30](https://www.youtube.com/watch?v=SkpwzfR4A1E&t=30s) What Claude makes opens beside the chat - [▶ 1:26](https://www.youtube.com/watch?v=SkpwzfR4A1E&t=86s) Faster voice mode and CarPlay Source: [YouTube: Claude’s new mobile app updates are here](https://www.youtube.com/watch?v=SkpwzfR4A1E) --- ## New on the Shelf 🛒 - [Grafana MCP server](https://agentsnacks.ai/shelf/#shelf-004-2) 🌶️🌶️ · MCP server **Attach a Grafana datasource to your chat the way you’d attach a file.** Grafana’s official server now lists datasources as MCP resources for clients with a resource picker, adds opt-in CloudWatch Logs Insights queries, and stops opening the interactive chart from every Prometheus or Tempo query (a breaking change). Works with: Any MCP client Install: `uvx mcp-grafana` Source: [mcp-grafana v2.2.0 release notes](https://github.com/grafana/mcp-grafana/releases/tag/v2.2.0) - [Playwright MCP](https://agentsnacks.ai/shelf/#shelf-004-1) 🌶️ · MCP server **Keep two sign-ins apart in one browser.** Microsoft’s browser server can now open tabs in separate named sessions that don’t share cookies or storage, blank out access tokens and other URL values you name in its replies (Microsoft calls that a convenience, not a security feature), and cap how many matches a page search returns. Works with: Any MCP client Install: `npx @playwright/mcp@latest` Source: [Playwright MCP v0.0.84 release notes](https://github.com/microsoft/playwright-mcp/releases/tag/v0.0.84) --- ## Before you go 🥡 **[Most of the pack’s news came with a date attached](https://agentsnacks.ai/before-you-go/#byg-004)** - **Claude Code now warns in `/schedule`** when a new cloud scheduled task picks up connectors you didn’t ask for ([our Daily](https://agentsnacks.ai/daily/2026-11-03-claude-code-schedule-connector-warning/)), and headless `claude -p` jobs get a heads-up before a stored sign-in runs out ([our Daily](https://agentsnacks.ai/daily/2026-11-04-claude-code-sign-in-expiry-warning/)). - Two Main Courses from Oct 30 are worth a catch-up. [Copilot Fusion’s bill](https://agentsnacks.ai/main-course/010/) explains when one prompt calls several models, and [Claude Duet, one week in](https://agentsnacks.ai/main-course/011/) shows how two early users run it. --- That’s the pack. Pass a snack to a friend. ## Snack Pack #3 — full text (https://agentsnacks.ai/packs/003/) # Snack Pack #3: Let your agents do the clicking Week of Oct 23–29, 2026 · 9 snacks · ~9 min **Your agents just took on more of the clicking.** OpenAI’s GPT-6.1 Terra takes over screen chores, while OpenCode, Cline and Claude changed things your setup may rely on. Copilot now picks models and runs workflows, Replit’s Monitoring Agent covers on-call, and Grok Bot logs in without ever seeing your password. ``` NUTRITION FACTS · Serving: 1 Snack Pack Snacks ................ 9 First Bite ............ 1 Main Course ........... 1 · 7-min read Secret Sauce .......... 1 Video Tasting ......... 1 New on the Shelf ...... 2 Before you go ......... 1 (Tiny bites worth a glance) Source links .......... 20 🌶️🌶️🌶️ items .......... 8 of 9 Actually useful ~100% ... 9 of 9 Counts from this issue · 16 cards · Hype: 0 g. ``` **Here’s what changed in your agent tools, and what to do about it.** Every bite links to its source. No funding fluff. No “AI will change everything.” Just what shipped, what changed, and why it might matter to you. A fresh pack lands every Thursday. --- ## 1. [GPT-6.1 Terra: OpenAI’s computer-use model at half the price](https://agentsnacks.ai/daily/2026-10-26-openai-gpt-6-1-terra/) 🌶️🌶️🌶️ *Major launch, lower API price* **Agent builders get a cheaper computer-use model**: `gpt-6.1-terra` is in the OpenAI API at $1 per million input tokens, $0.05 cached and $5 output. OpenAI announced it on Oct 26, 2026, and says that’s 50% below GPT-5.6 Terra and GPT-6.1 Sol. It’s also in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Who cares: Developers running computer-use or high-volume agent work on the OpenAI API, and Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Source: [OpenAI: introducing GPT-6.1 Terra, Oct 26, 2026](https://openai.com/index/introducing-gpt-6-1-terra) --- ## 2. [OpenCode GitHub Action removed: what to do](https://agentsnacks.ai/daily/2026-10-27-opencode-github-action-removed/) 🌶️🌶️🌶️ *Breaking changes on upgrade* **Action needed**: the OpenCode GitHub Action is no longer published, and the release notes say to remove it from your workflows. The same release, out Oct 27, 2026, also moves the Promise API to the root of `@opencode/ai` and makes TUI plugins require OpenTUI 0.7.0. Who cares: Teams that run OpenCode in GitHub workflows, plugin authors, and developers who import @opencode/ai. Source: [OpenCode v2.2.0 release notes, Oct 27, 2026](https://github.com/anomalyco/opencode/releases/tag/v2.2.0) --- ## 3. [Cline shuffles its default models again](https://agentsnacks.ai/daily/2026-10-26-cline-default-models-shuffle/) 🌶️🌶️🌶️ *Default model swap* **Check your model**: Cline’s Oct 26, 2026 updates change default models again. ClinePass now defaults to Kimi K3.1 Pro, and most other providers get new defaults, mostly Kimi K3.1 Pro, MiniMax M3.1 Flash or DeepSeek V4.1 Flash 1017. With no model chosen, the Cline provider uses Claude Sonnet 5.5 again. Who cares: Cline users who never pinned a model, and ClinePass subscribers. Source: [Cline v4.1.26 release notes](https://github.com/cline/cline/releases/tag/v4.1.26) · [Cline v3.0.74 release notes](https://github.com/cline/cline/releases/tag/cli-v3.0.74) · [Cline Desktop v0.0.49 release notes](https://github.com/cline/cline/releases/tag/desktop-v0.0.49) --- ## 4. [Claude apps are switching to auto mode](https://agentsnacks.ai/daily/2026-10-23-claude-apps-auto-mode-default/) 🌶️🌶️🌶️ *Default permission swap* **Check your Claude mode**: auto mode is becoming the default in most Claude apps. In an update dated Oct 22, 2026, Anthropic said it rolls out next week as the default for new sessions on Pro, Max and Team plans. Instead of asking you to approve each action, an automated check lets safe actions go ahead and blocks risky ones while Claude looks for another route. Who cares: Claude Pro, Max and Team users who let Claude take actions, and Enterprise admins. Source: [Anthropic: auto mode default update, Oct 22, 2026](https://claude.com/resources/articles/auto-mode-default-in-claude-code) · [Claude Help Center: Get started with Claude Cowork](https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork) --- ## 5. [Copilot Fusion GA picks models for you](https://agentsnacks.ai/daily/2026-10-28-copilot-fusion-ga/) 🌶️🌶️🌶️ *Stop picking a model per task* **No more guessing which model fits the job.** Fusion went generally available on Oct 28, 2026: select Fusion in the model picker, choose Standard, High or Max, and Copilot routes and orchestrates models for each prompt. Who cares: Copilot users in VS Code, the Copilot app, Copilot CLI and JetBrains, plus admins who set model policy. Source: [GitHub Changelog: Route and orchestrate models with GitHub Copilot Fusion, Oct 28, 2026](https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion) --- ## 6. [GitHub Agentic Workflows generally available](https://agentsnacks.ai/daily/2026-10-28-github-agentic-workflows-ga/) 🌶️🌶️🌶️ *Automate triage and CI with agents* **Your next GitHub Action can start as a Markdown file.** Agentic Workflows went generally available on Oct 28, 2026. Describe issue triage, CI failure analysis, docs updates or test writing in natural-language Markdown, and they compile into standard Actions YAML. Who cares: Teams that want coding agents inside GitHub Actions for triage, CI, docs or tests. Source: [GitHub Changelog: GitHub Agentic Workflows generally available, Oct 28, 2026](https://github.blog/changelog/2026-10-28-github-agentic-workflows-generally-available) --- ## 7. [Replit’s new agent takes your on-call shift](https://agentsnacks.ai/daily/2026-10-23-replit-monitoring-agent/) 🌶️🌶️🌶️ *New on-call agent* **Replit’s Monitoring Agent answers your app’s alarms for you.** Launched on Oct 22, 2026, it can investigate a production issue, make and validate a fix and email you a summary of what happened. Turn on Auto-republish and it can publish the fix too; leave it off to review the change first. Who cares: People who publish apps on Replit without an on-call engineer to call. Source: [Replit: Monitoring Agent announcement, Oct 22, 2026](https://replit.com/blog/replit-monitoring-agent) · [Replit docs: set up Monitoring Agent](https://docs.replit.com/features/publishing/monitoring-agent) --- ## 8. [Grok Bot signs in without seeing your password](https://agentsnacks.ai/daily/2026-10-23-1password-for-agents-grok-bot/) 🌶️🌶️🌶️ *New agent sign-in launch* **Your agent can now sign in without seeing your password.** On Oct 22, 2026, 1Password launched 1Password for Agents, with Grok Bot among the first integrations. Your agent asks for one saved login, you approve it in 1Password, and 1Password fills it in, so the agent never sees the password or the one-time code. Who cares: Grok Bot users with a 1Password Individual or Family account who want tasks done behind a login. Source: [1Password: 1Password for Agents announcement, Oct 22, 2026](https://1password.com/blog/1password-for-ai-agents) · [1Password: how Grok Bot uses your approved logins](https://1password.com/partners/spacexsi-grokbot) --- ## 9. [Keep coding agents out of your Downloads](https://agentsnacks.ai/daily/2026-10-25-agentsmd-downloads-hijack/) 🌶️🌶️ *Security risk, no patch coming* **Coding agent users, stop launching agents from Downloads**: a planted AGENTS.md there told a real customer’s agent to install a malicious npm package. Forever Security caught the attempt and reproduced the attack on Codex and Claude Code. There’s no patch coming, because the agents are doing what they were built to do. Who cares: Anyone who starts Codex, Claude Code or another coding agent from their Downloads folder. Source: [Forever Security: AGENTS.md hijack in the Downloads folder, Oct 22, 2026](https://forever.security/blog/hijacking-ai-agents-via-downloads-folder/) · [BleepingComputer: coding agents hijacked from Downloads, Oct 25, 2026](https://www.bleepingcomputer.com/news/security/ai-coding-agents-hijacked-using-agentsmd-file-in-downloads-folder/) --- ## First Bite 🍪 **[What’s a GitHub Action?](https://agentsnacks.ai/first-bite/#fb-003)** **A GitHub workflow is a small automated job that lives** in your repository: a YAML file in the `.github/workflows` folder that runs when something happens, like a new pull request or issue, on a schedule, or when you start it by hand. An action is a ready-made, reusable step that workflows plug in, such as pulling your code or setting up your tools, and GitHub Actions is the service that runs it all ([GitHub Docs: understand GitHub Actions](https://docs.github.com/en/actions/get-started/understand-github-actions)). Try this: open the `.github/workflows` folder in one of your repositories and see which actions it uses, then read [our OpenCode Daily](https://agentsnacks.ai/daily/2026-10-27-opencode-github-action-removed/) if one of them is OpenCode’s. New here? [Start here](https://agentsnacks.ai/start/). --- ## The Main Course 🍽️ — [GPT-6.1 Terra: hand it the screen chores](https://agentsnacks.ai/main-course/008/) **OpenAI’s new Terra model takes on your screen chores for less.** GPT-6.1 Terra, out Oct 26, 2026, reads what’s on your screen, clicks and types across your apps, and writes in a clear, direct style. On paid ChatGPT plans it lives in ChatGPT Work and Codex, where OpenAI says your usage allowance covers more tasks with it than with its bigger models. Early users are already filling in forms, making fonts and hunting bugs with it, and a few are unimpressed. Full Main Course: 7-min read [Read the full Main Course →](https://agentsnacks.ai/main-course/008/) Sources: [OpenAI: introducing GPT-6.1 Terra, Oct 26, 2026](https://openai.com/index/introducing-gpt-6-1-terra) · [ChatGPT Learn: what’s new, Oct 26–30, 2026](https://learn.chatgpt.com/docs/whats-new/october-26-30-2026) · [OpenAI Help Center: ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes) · [ChatGPT Learn: models in Work and Codex](https://learn.chatgpt.com/docs/models) · [ChatGPT Learn: computer use](https://learn.chatgpt.com/docs/computer-use) · [OpenAI Help Center: model release notes](https://help.openai.com/en/articles/9624314-model-release-notes) · [Hacker News: GPT-6.1 Terra discussion](https://news.ycombinator.com/item?id=50248124) · [OpenAI on X: GPT-6.1 Terra launch](https://x.com/OpenAI/status/2114780711664398349) · [Example shared by @victornunez](https://x.com/victornunez/status/2114825862935887926) · [Example shared by @VectorQl](https://x.com/VectorQl/status/2114950106042736813) · [Example shared by @ALwith_ai](https://x.com/ALwith_ai/status/2114923853185204338) · [Example shared by @ArcyDev](https://x.com/ArcyDev/status/2114813509779206453) · [Example shared by @MiaAI_lab](https://x.com/MiaAI_lab/status/2114807561837072570) · [Example shared by @TriACJ1409](https://x.com/TriACJ1409/status/2114971767974600828) --- ## Secret Sauce 🧂 — [Only let the actions you trust run in your repo](https://agentsnacks.ai/secret-sauce/#ss-003) **Try it now:** 1. On GitHub, open your repository’s Settings, then click Actions and General in the left sidebar. 2. Under “Actions permissions”, choose “Allow OWNER, and select non-OWNER, actions and reusable workflows”, then list the outside actions your workflows still need. 3. To lock every action to one exact version, also tick “Require actions to be pinned to a full-length commit SHA”, then click Save. An organization or enterprise policy can override these settings. Source: [GitHub Docs: manage GitHub Actions settings for a repository](https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/enabling-features-for-your-repository/managing-github-actions-settings-for-a-repository) --- ## Video Tasting ▶️ [Track tokens and models with the My AI Usage Canvas | GitHub Copilot](https://agentsnacks.ai/videos/#v-2xiyBnQFM20) Intermediate · 4:39 · GitHub · Oct 17, 2026 Video Summary: - **Copilot’s /chronicle commands look back at your past sessions.** The My AI Usage Canvas, a custom dashboard, turns that history into charts in the Copilot app. - The Usage tab shows tokens (the chunks of text AI models read and write), model calls, your model mix, top repos and the hours you usually start sessions. - The Insights tab writes a tip from your own history and suggests experiments. Start one, note what you tried, mark it done, and revisit saved insights later. On screen: The /chronicle command list in the Copilot app, then the My AI Usage dashboard: usage cards, a model mix chart, per-session token counters, activity and hour-of-day charts, and an Insights tab with an evidence note and Start buttons for experiments. The install chapter is talk only; setup steps are in the video description. After watching you can: See where your Copilot tokens go, which models you lean on, and try one small habit change to use fewer. Short on time? Watch just: - [▶ 0:54](https://www.youtube.com/watch?v=2xiyBnQFM20&t=54s) Exploring the usage tab and model mix - [▶ 2:38](https://www.youtube.com/watch?v=2xiyBnQFM20&t=158s) Generating personalized tips in the insights tab - [▶ 3:09](https://www.youtube.com/watch?v=2xiyBnQFM20&t=189s) Running experiments to improve AI habits Source: [YouTube: Track tokens and models with the My AI Usage Canvas | GitHub Copilot](https://www.youtube.com/watch?v=2xiyBnQFM20) --- ## New on the Shelf 🛒 - [GitHub security extension](https://agentsnacks.ai/shelf/#shelf-003-1) 🌶️ · Plugin **Fix code scanning alerts without bouncing between github.com and your editor.** In public preview, the extension lists your repository’s open alerts by severity, lets Copilot explain or fix them in your local working copy only (nothing is committed or pushed), and runs /security-review on your current changes. Works with: GitHub Copilot app Install: copilot plugin install github-security@awesome-copilot Source: [GitHub Changelog post (2026-10-28)](https://github.blog/changelog/2026-10-28-security-extension-for-the-github-copilot-app) - [MCP TypeScript SDK](https://agentsnacks.ai/shelf/#shelf-003-2) 🌶️ · SDK **Serve your Agent Skills from an MCP server instead of copying skill folders around.** The new Skills extension means a server can list and serve skills, on Node it can load a whole folder of them, and a client can list, get and read them with size, digest and SKILL.md checks. The server’s attach step is still marked experimental. Works with: Any MCP client Install: `npm i @modelcontextprotocol/server` Source: [typescript-sdk v2.4.0 release notes](https://github.com/modelcontextprotocol/typescript-sdk/releases/tag/%40modelcontextprotocol%2Fserver%402.4.0) --- ## Before you go 🥡 **[Agents do more on their own now, so the defaults matter more](https://agentsnacks.ai/before-you-go/#byg-003)** **New defaults, a retired Action and a planted AGENTS.md** are the ones to check, while Terra and Copilot took the most work off your hands. - Copilot now runs the clearly safe steps without asking in Copilot CLI, the Copilot app and VS Code ([our Daily](https://agentsnacks.ai/daily/2026-10-27-copilot-assisted-permissions/)), and Copilot features you left unconfigured now follow your global default ([our Daily](https://agentsnacks.ai/daily/2026-10-23-copilot-global-default-policy/)). - Update now: long Claude Code sessions stop dead-ending on a credits error ([our Daily](https://agentsnacks.ai/daily/2026-10-27-claude-code-1m-context-credits-fix/)), and custom model names on cloud providers now get a shorter prompt ([our Daily](https://agentsnacks.ai/daily/2026-10-28-claude-code-custom-model-prompt/)). - Codex CLI gets a built-in Google Cloud Vertex AI provider ([our Daily](https://agentsnacks.ai/daily/2026-10-28-codex-vertex-ai-provider/)), and OpenAI’s Decisions API is out of beta ([our Daily](https://agentsnacks.ai/daily/2026-10-23-openai-decisions-api-ga/)). - Copilot SDK agents can make requests for your organization without anyone’s login ([our Daily](https://agentsnacks.ai/daily/2026-10-28-copilot-server-to-server-auth/)). - Cline Desktop now shows every file your agent changed ([our Daily](https://agentsnacks.ai/daily/2026-10-26-cline-desktop-changes-pane/)). - For the full story on Replit’s on-call agent, [our Main Course on Monitoring Agent](https://agentsnacks.ai/main-course/006/) walks through its three switches. --- That’s the pack. Pass a snack to a friend. ## Snack Pack #2 — full text (https://agentsnacks.ai/packs/002/) # Snack Pack #2: Hand off the chores, keep the final say Week of Oct 16–22, 2026 · 9 snacks · ~9 min **Hand off the chores, keep the final say.** Claude Duet takes chores off your plate but asks first, and standalone Claude Design is closing. Then check your setup: PHP and Pydantic AI shipped security fixes, Claude Code fixed Bedrock and Vertex requests, and Cline’s default model moved twice. ``` NUTRITION FACTS · Serving: 1 Snack Pack Snacks ................ 9 First Bite ............ 1 Main Course ........... 1 · 7-min read Secret Sauce .......... 1 Video Tasting ......... 1 New on the Shelf ...... 3 Before you go ......... 1 (Tiny bites worth a glance) Source links .......... 19 🌶️🌶️🌶️ items .......... 8 of 9 Actually useful ~100% ... 9 of 9 Counts from this issue · 17 cards · Hype: 0 g. ``` **Here’s what changed in your agent tools this week, and what to do about it.** Every bite links to its source. No funding fluff. No “AI will change everything.” Just what shipped, what changed, and why it might matter to you. A fresh pack lands every Thursday. --- ## 1. [Claude Duet works while you step away](https://agentsnacks.ai/daily/2026-10-21-claude-duet-proactive-agent/) 🌶️🌶️🌶️ *New proactive agent* **Claude Duet checks your apps and tells you what needs you.** Anthropic’s new app sends a morning brief, takes on chores on its own cloud computer while you’re away, and asks before it sends, pays, signs or deletes anything. It’s a research preview by waitlist for Pro and Max subscribers. Who cares: Claude Pro and Max subscribers juggling lots of work apps, especially small business owners. Source: [Anthropic: Claude Duet announcement](https://claude.com/resources/articles/claude-duet) · [Claude Help Center: get started with Claude Duet](https://support.claude.com/en/articles/17631495-get-started-with-claude-duet) --- ## 2. [Standalone Claude Design closes Dec 14](https://agentsnacks.ai/daily/2026-10-19-claude-design-standalone-closes/) 🌶️🌶️🌶️ *Deadline Dec 14* **Action needed**: move your design systems out of standalone Claude Design before Dec 14, 2026. That’s when claude.ai/design closes and starts sending you to Claude, where Claude Design now lives and new designs are created as artifacts. Save any chats and comments you need first, because they don’t carry over. Who cares: Anyone with design systems, chats or comments in the standalone Claude Design site. Source: [Claude Help Center: migrate from standalone Claude Design](https://support.claude.com/en/articles/17440474-migrate-from-standalone-claude-design-to-claude) --- ## 3. [Patch PHP now: two critical remote-code bugs](https://agentsnacks.ai/daily/2026-10-21-php-critical-rce-patch/) 🌶️🌶️🌶️ *Security fix* **PHP users, update now**: two critical bugs can let an attacker run code on your server without logging in. PHP’s maintainers shipped the fixes on Oct 21, 2026 in 8.5.12, 8.4.27, 8.3.36 and 8.2.35. OpenAI says researchers at Calif found one of the flaws using GPT-5.6 Sol, and Wiz independently reported both. Who cares: Anyone who runs or hosts a website or web app on PHP, especially with Apache mod_php. Source: [PHP security advisory: critical RCE fixes, Oct 21, 2026](https://www.php.net/archive/2026.php#2026-10-21-5) · [OpenAI: Patch the Planet PHP disclosure, Oct 21, 2026](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities) --- ## 4. [Pydantic AI patches a High web_fetch bug](https://agentsnacks.ai/daily/2026-10-21-pydantic-ai-web-fetch-fix/) 🌶️🌶️🌶️ *Security fix* **Pydantic AI users, update now**: a High-severity fix stops crafted web pages from tying up your app. When `web_fetch_tool` fetched attacker-controlled HTML with many elements, converting it could delay unrelated work in the same process. The maintainers published the fix on Oct 21, 2026: upgrade to 2.59.0, or 1.107.9 on v1. Who cares: Anyone whose Pydantic AI agent uses web_fetch_tool on pages an attacker can influence. Source: [Pydantic AI web_fetch security advisory, Oct 21, 2026](https://github.com/pydantic/pydantic-ai/security/advisories/GHSA-4whf-38hw-pxxh) --- ## 5. [Pydantic AI keeps MCP tokens out of traces](https://agentsnacks.ai/daily/2026-10-20-pydantic-ai-mcp-trace-fix/) 🌶️🌶️🌶️ *Security fix* **Pydantic AI users, update now**: two Medium security fixes are out, one for MCP credentials that could reach OpenTelemetry traces and one for the Vercel AI and AG-UI adapters. Upgrade to 2.58.0, or 1.107.8 on v1, then restrict access to old traces and rotate any credentials they exposed. Who cares: Anyone using MCP tools in Pydantic AI with OpenTelemetry tracing on, or serving an agent through the Vercel AI or AG-UI adapter. Source: [Pydantic AI v2.58.0 release notes](https://github.com/pydantic/pydantic-ai/releases/tag/v2.58.0) --- ## 6. [Claude Code fixes its Bedrock and Vertex snag](https://agentsnacks.ai/daily/2026-10-21-claude-code-gateway-fix/) 🌶️🌶️🌶️ *Broken requests fixed* **Bedrock and Vertex users, update now**: Claude Code’s failing requests are fixed. The previous release could fail every request on Vertex, Bedrock and other providers when the endpoint or a gateway rejected tool search. Claude Code now retries without it. Who cares: Claude Code users on Vertex, Bedrock, other providers or behind a gateway, plus admins who manage MCP allowlists. Source: [Claude Code v2.1.305 release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.305) --- ## 7. [Unpinned Cline setups got a new default model](https://agentsnacks.ai/daily/2026-10-16-cline-cloud-sessions/) 🌶️🌶️🌶️ *Silent model swap* **Check your model**: unpinned Cline provider setups now default to Gemini 4 Argon. OpenRouter stays on Claude Sonnet 5.5, but over 20 other providers change defaults too, so pin the model you want to keep. Also new: Cloud sessions, rolling out gradually, run a job in a hosted sandbox on a GitHub repo and branch you pick and keep going after you close the editor. Who cares: Cline users who never pinned a model, and anyone who wants a task to keep running after the editor closes. Source: [Cline v4.1.25 release notes](https://github.com/cline/cline/releases/tag/v4.1.25) --- ## 8. [Cline CLI and Desktop default to Sonnet again](https://agentsnacks.ai/daily/2026-10-19-cline-sonnet-default-returns/) 🌶️🌶️🌶️ *Default model swap* **Check your model**: the Cline provider’s default in Cline CLI and Desktop is Claude Sonnet 5.5 again. It’s back at the top of Cline’s recommended list. Also fixed: GPT-6 models now work on OpenAI and Azure OpenAI-compatible endpoints, where every request used to be rejected. Who cares: Cline CLI and Desktop users on the Cline provider who never chose a model, and anyone calling GPT-6 models through OpenAI or Azure endpoints. Source: [Cline v3.0.73 release notes](https://github.com/cline/cline/releases/tag/cli-v3.0.73) · [Cline Desktop v0.0.48 release notes](https://github.com/cline/cline/releases/tag/desktop-v0.0.48) --- ## 9. [Six Copilot models just left the picker](https://agentsnacks.ai/daily/2026-10-20-copilot-six-models-deprecated/) 🌶️🌶️ *Models deprecated Oct 19* **Action needed**: six Copilot models were deprecated on Oct 19, 2026, including GPT-5.5, GPT-5.4 and Grok 4.5. Move workflows and integrations to the suggested alternatives, such as GPT-6.1 Sol, GPT-6 Luna, Gemini 3.8 Flash and Grok 4.7. Who cares: Anyone with a Copilot workflow or integration that names one of these six models, and Copilot Enterprise admins who manage model policies. Source: [GitHub Changelog: six Copilot models deprecated](https://github.blog/changelog/2026-10-19-github-copilot-models-deprecated-on-october-19-2026) --- ## First Bite 🍪 **[What’s a security advisory?](https://agentsnacks.ai/first-bite/#fb-002)** **A security advisory is the “patch me” note for a piece of software.** When a project finds a security hole, its advisory names the affected versions and the earliest version with the fix, so you know exactly what to update to. Each one carries a severity: Low, Medium (also called Moderate), High or Critical. On GitHub it also gets an ID that starts with `GHSA-` ([GitHub Docs: what’s inside a security advisory](https://docs.github.com/en/code-security/security-advisories/working-with-global-security-advisories-from-the-github-advisory-database/about-the-github-advisory-database)). Try this: open [our Pydantic AI web_fetch Daily](https://agentsnacks.ai/daily/2026-10-21-pydantic-ai-web-fetch-fix/) and spot its severity and the version to update to. New here? [Start here](https://agentsnacks.ai/start/). --- ## The Main Course 🍽️ — [Claude Duet: an AI helper that asks first](https://agentsnacks.ai/main-course/005/) **Your overflowing to-do list just got a teammate**, because Claude Duet keeps an eye on your work apps, takes on chores you hand it and asks before it acts for you. It works from its own computer in the cloud, so jobs keep going after you put your phone down, and it stops for your OK before it sends, pays, signs or deletes anything. It’s a research preview by waitlist for Claude Pro and Max subscribers, in one region for now. Full Main Course: 7-min read [Read the full Main Course →](https://agentsnacks.ai/main-course/005/) Sources: [Anthropic: Claude Duet announcement](https://claude.com/resources/articles/claude-duet) · [Claude Duet product page and FAQ](https://claude.com/product/claude-duet) · [Claude Help Center: get started with Claude Duet](https://support.claude.com/en/articles/17631495-get-started-with-claude-duet) · [Hacker News discussion: Claude Duet](https://news.ycombinator.com/item?id=50186037) --- ## Secret Sauce 🧂 — [Let the next advisory find you](https://agentsnacks.ai/secret-sauce/#ss-002) **Try it now:** 1. On GitHub, open your repository’s Settings, then Advanced Security in the “Security and quality” section of the sidebar. 2. Next to Dependabot alerts, click Enable. For every repo you own at once, use Enable all in your account’s security settings and tick “Enable by default for new repositories.” 3. From then on, Dependabot raises an alert when it finds a vulnerable dependency, and people with write, maintain or admin access to the repo get notified. Source: [GitHub Docs: turn on Dependabot alerts](https://docs.github.com/en/code-security/dependabot/dependabot-alerts/configuring-dependabot-alerts) --- ## Video Tasting ▶️ [Coding on the go with Claude Code](https://agentsnacks.ai/videos/#v--OH1y7CjIew) Intermediate · 4:02 · Claude · Oct 13, 2026 Video Summary: - **A cloud session runs Claude Code** on a virtual machine (a computer in the cloud), not your laptop. It copies your repo (your project’s code on GitHub) and works on its own branch (a separate copy of the code), so it keeps going while your laptop sleeps. - Good for long jobs like a big refactor or migration, ideas that hit you away from your desk, or several tasks at once. It counts toward your plan’s limits, with no extra charge for the virtual machine. - Connect GitHub and pick which repos Claude can see. Then check in from the phone app, review the changes and tap Create PR (a pull request: a proposed change you approve before it joins the main code). On screen: Desktop and phone app: a sign-up form bug fixed while the laptop is closed, animated cloud-session diagrams, GitHub connect screens, a diff reviewed on the phone, and claude --teleport in a terminal. After watching you can: Hand Claude Code a task in the cloud, close your laptop, and answer its questions from your phone. Short on time? Watch just: - [▶ 0:18](https://www.youtube.com/watch?v=-OH1y7CjIew&t=18s) What is a cloud session? - [▶ 0:41](https://www.youtube.com/watch?v=-OH1y7CjIew&t=41s) When to use one - [▶ 1:09](https://www.youtube.com/watch?v=-OH1y7CjIew&t=69s) Connecting GitHub - [▶ 2:50](https://www.youtube.com/watch?v=-OH1y7CjIew&t=170s) From your phone Source: [YouTube: Coding on the go with Claude Code](https://www.youtube.com/watch?v=-OH1y7CjIew) --- ## New on the Shelf 🛒 - [Grafana MCP server](https://agentsnacks.ai/shelf/#shelf-002-1) 🌶️🌶️ · MCP server **Ask your agent about a metric and get a chart back, not a wall of numbers.** Grafana’s official server now draws Prometheus query results as an interactive chart in hosts that support MCP Apps, and it can create Mimir, Cortex and Loki alert rules. Works with: Any MCP client Install: `uvx mcp-grafana` Source: [mcp-grafana v2.1.0 release notes](https://github.com/grafana/mcp-grafana/releases/tag/v2.1.0) - [Cloudflare Agents](https://agentsnacks.ai/shelf/#shelf-002-3) 🌶️ · SDK **Agents you build on Cloudflare can keep their skills in a real workspace.** Agent Skills in the agents SDK now read and write a Computer workspace, and a chat turn that keeps running out of memory now stops after five interrupted attempts instead of retrying again and again. Works with: Cloudflare Workers Install: `npm i agents` Source: [Cloudflare Agents v0.29.0 release notes](https://github.com/cloudflare/agents/releases/tag/agents%400.29.0) - [MCP reference servers](https://agentsnacks.ai/shelf/#shelf-002-2) 🌶️ · MCP server **New spec, old client, same servers.** The official everything, filesystem, memory, sequential-thinking, fetch, git and time servers now support the modern MCP spec while still answering older clients, and the TypeScript ones now need Node 24 or newer. Works with: Any MCP client Install: `npx -y @modelcontextprotocol/server-filesystem ` Source: [MCP reference servers v2.0.0 release notes](https://github.com/modelcontextprotocol/servers/releases/tag/v2.0.0) --- ## Before you go 🥡 **[Patch, pin, then let the helper in](https://agentsnacks.ai/before-you-go/#byg-002)** **Security fixes, a deadline and two rounds of model musical chairs** filled the tray, while Claude Duet showed what an agent that asks first looks like. - Update now: Gemini CLI no longer empties a project’s settings file when you add or remove an MCP server in a folder you haven’t trusted ([our Daily](https://agentsnacks.ai/daily/2026-10-21-gemini-cli-untrusted-settings/)). - Update now: MCP Inspector 2.12.0 fixes three security holes in earlier v2 releases ([MCP Inspector 2.12.0 release notes](https://github.com/modelcontextprotocol/inspector/releases/tag/2.12.0)). - Claude Code warns you before resuming an old chat would eat into your usage limit ([our Daily](https://agentsnacks.ai/daily/2026-10-20-claude-code-resume-limit-warning/)), and idle compaction now has an off switch ([our Daily](https://agentsnacks.ai/daily/2026-10-19-claude-code-idle-compaction-switch/)). - Claude Security on Claude Enterprise now keeps scanning as your code changes ([our Daily](https://agentsnacks.ai/daily/2026-10-22-claude-security-continuous-scans/)). - Copilot admins can now set model policies with a script ([our Daily](https://agentsnacks.ai/daily/2026-10-20-copilot-model-policy-api/)). - Closed Codex, but it’s still running? [Our Main Course on its background helper](https://agentsnacks.ai/main-course/004/) explains why. --- That’s the pack. Pass a snack to a friend. ## Snack Pack #1 — full text (https://agentsnacks.ai/packs/001/) # Snack Pack #1: Your agents got a babysitter Week of Oct 9–15, 2026 · 7 snacks · ~7 min **Your agents got a babysitter.** Copilot CLI’s Autopilot no longer stalls on a hook’s “ask” when nobody is watching, and Cline changed its default models for unpinned setups on Cline, OpenRouter and about 20 other providers. Plus OpenCode stops auto-updating and Copilot learns to fix your security alerts. ``` NUTRITION FACTS · Serving: 1 Snack Pack Snacks ................ 7 First Bite ............ 1 Main Course ........... 1 · 6-min read Secret Sauce .......... 1 Video Tasting ......... 1 New on the Shelf ...... 3 Before you go ......... 1 (Tiny bites worth a glance) Source links .......... 17 🌶️🌶️🌶️ items .......... 2 of 7 Actually useful ~71% ... 5 of 7 Counts from this issue · 15 cards · Hype: 0 g. ``` **Catch up on the week’s agent changes that touch your setup.** Snack Pack #1 lives right here on the web, and every bite links to its source. --- ## 1. [Copilot CLI: hooks that don’t hang the babysitter](https://agentsnacks.ai/daily/2026-10-13-copilot-cli-hooks/) 🌶️🌶️🌶️ *Stops Autopilot hangs* **Copilot CLI Autopilot no longer stalls on a hook’s “ask” when nobody is watching.** - One more folder: `/add-dir` grants sandbox access for the session; interactive settings nudge you to mask likely secrets. - No more stalled Autopilot: agent-scoped declarative hooks; Autopilot now denies a preToolUse hook’s “ask” instead of stopping on a permission dialog nobody is watching. - New model, easier answers: Gemini 4 Argon support; `/ask` expands with Ctrl+E, copies with Ctrl+Y. Who cares: Terminal Copilot users running Autopilot / unattended loops. Source: [Copilot CLI v1.0.97 release notes](https://github.com/github/copilot-cli/releases/tag/v1.0.97) · [Copilot CLI v1.0.96 release notes](https://github.com/github/copilot-cli/releases/tag/v1.0.96) · [Copilot CLI v1.0.98 release notes](https://github.com/github/copilot-cli/releases/tag/v1.0.98) --- ## 2. [Cline swapped your default model](https://agentsnacks.ai/daily/2026-10-13-cline-default-swap/) 🌶️🌶️🌶️ *Silent model swap* **Check your model**: unpinned Cline setups just got a new default. Cline + OpenRouter users now land on Claude Sonnet 5.5. About 20 other providers also shuffle defaults, so unpinned users will wake up on a different model. Pin the model you want to keep. Who cares: Cline users who never pinned a model. Check the cost dashboard before you blame the agent. Source: [Cline v4.1.24 release notes](https://github.com/cline/cline/releases/tag/v4.1.24) · [Cline Desktop v0.0.46 release notes](https://github.com/cline/cline/releases/tag/desktop-v0.0.46) --- ## 3. [OpenCode 1.19: the on-ramp to OpenCode 2 (and auto-update is off)](https://agentsnacks.ai/daily/2026-10-10-opencode-1-19/) 🌶️🌶️ *Manual upgrades now* **Action needed**: OpenCode 2 is here, and updates are no longer automatic. `opencode upgrade --major` (or `/update` in the TUI) walks you there, so upgrade on purpose. It also allows OpenAI’s `ultrafast` service tier. Who cares: OpenCode folks. Deliberate upgrades > surprise majors. Source: [OpenCode v1.19.0 release notes](https://github.com/anomalyco/opencode/releases/tag/v1.19.0) --- ## 4. [Copilot will now fix your security warnings for you](https://agentsnacks.ai/daily/2026-10-15-copilot-codeql-autofix/) 🌶️🌶️ *Enterprise-policy gated* **Copilot can now fix your security alerts and prove it.** Assign a CodeQL alert and it patches the code, re-runs CodeQL to prove the alert is gone, then opens a draft PR. Up to 25 alerts per batch. Now GA. Default-branch alerts only: an enterprise admin has to turn on the “Agentic autofix” policy, and each run spends Code Security AI Credits + Actions minutes. You can also start it via the REST API (`copilot-swe-agent[bot]`). Who cares: Security folks staring at a CodeQL mountain on GHEC / GHAS. Source: [GitHub Changelog](https://github.blog/changelog/2026-10-15-agentic-autofix-for-code-scanning-alerts-general-availability) --- ## 5. [HydraFusion stops making you pick a model (and adds a “Max” mode)](https://agentsnacks.ai/daily/2026-10-14-copilot-hydrafusion-max/) 🌶️ *Research preview* **HydraFusion gets a bigger model pool and a Max tier.** Its model pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. Cascade gets a judging step; Critique can tool-snoop the workspace with parallel critics. New HydraFusion Max is the “go deep” button. Lives in VS Code, the Copilot app, and Copilot CLI. Who cares: Copilot Pro, Pro+, Business and Enterprise (with preview features on). Source: [GitHub Changelog](https://github.blog/changelog/2026-10-14-hydrafusion-updates-and-new-max-option) --- ## 6. [Claude Code: idle chats now compact before the cache expires](https://agentsnacks.ai/daily/2026-10-14-claude-code-idle-compaction/) 🌶️🌶️ *Know the opt-out* **Walk away from a long chat and Claude Code tidies it up.** After ~54 idle minutes, a 200K+ token chat compacts before the prompt cache expires, so your next message doesn’t re-chew the whole transcript. Tune the threshold with `CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS`, or opt out with `"idleCompaction": false`. Who cares: Anyone with marathon Claude Code sessions — or a token bill that makes them wince. Source: [Claude Code v2.1.299 release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.299) · [Claude Code v2.1.300 release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.300) --- ## 7. [Cursor Origin opens the snack aisle](https://agentsnacks.ai/daily/2026-10-13-cursor-origin-apps/) 🌶️ *Mostly for tool makers* **Cursor’s Origin now plugs into the apps you already use.** That’s 20+ apps: Cloudflare, Vercel, Buildkite, CircleCI, CodeRabbit, Greptile, Semgrep, Socket, Sentry, Linear, and friends. Public submissions welcome. Apps auth with app JWT + install tokens, get webhooks, write check runs / PR comments / reviews. Who cares: Dev-tool makers hunting distribution, and teams wiring CI/review bots into Origin. Source: [Cursor changelog](https://cursor.com/changelog/origin-app-marketplace) --- ## First Bite 🍪 **[What’s a hook?](https://agentsnacks.ai/first-bite/#fb-001)** **Hooks are house rules your agent can’t skip.** A hook is a small command that Copilot cloud agent and Copilot CLI run at set moments: when a session starts, when you enter a prompt, or right before it uses a tool. The before-a-tool one (`preToolUse`) can approve or deny that step ([GitHub Docs: About hooks](https://docs.github.com/en/copilot/concepts/agents/hooks)). Try this: peek in your repo for a `.github/hooks/` folder, then read [our Copilot CLI hooks Daily](https://agentsnacks.ai/daily/2026-10-13-copilot-cli-hooks/) to see why an “ask” no longer freezes Autopilot. New here? [Start here](https://agentsnacks.ai/start/). --- ## The Main Course 🍽️ — [Agent Skills: recipe cards for your AI](https://agentsnacks.ai/main-course/001/) **Tired of re-explaining the same steps to your AI every single chat?** An Agent Skill is a recipe card it keeps in a folder (SKILL.md): only the name and blurb sit on the menu at startup (about 100 tokens), and the full recipe comes out when your task matches. Skills teach your agent how to cook; new hands to grab your tools are MCP’s job. Full Main Course: 6-min read [Read the full Main Course →](https://agentsnacks.ai/main-course/001/) Sources: [Anthropic: Agent Skills announcement](https://claude.com/resources/articles/skills) · [Claude docs: how Skills load](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview) · [MCP docs: what MCP connects](https://modelcontextprotocol.io/docs/getting-started/intro) · [Claude Code docs: using Skills](https://code.claude.com/docs/en/skills) · [Anthropic skills repo (GitHub)](https://github.com/anthropics/skills) · [Cursor CLI changelog: skill folders it scans](https://cursor.com/docs/cli/changelog) --- ## Secret Sauce 🧂 — [Pin your model before the defaults move](https://agentsnacks.ai/secret-sauce/#ss-001) **Try it now:** 1. Open your agent settings (Cline, Copilot, Claude Code, Cursor — wherever you live). 2. Explicitly pick the model you want this week so a quiet release cannot swap the chef under you. 3. Recheck cost dashboards after any "default model" note in release notes — surprises usually show up on the bill first. Source: [Cline v4.1.24 release notes](https://github.com/cline/cline/releases/tag/v4.1.24) --- ## Video Tasting ▶️ [Claude Agent Skills Explained](https://agentsnacks.ai/videos/#v-fOxC44g8vig) Intermediate · 3:14 · Anthropic · Nov 26, 2025 Video Summary: - **Skills are expertise Claude loads only when it needs it.** Each one is a folder that kicks in automatically when a task matches. - Only each skill’s name and description load at startup; the full SKILL.md loads on demand. - Compares skills with CLAUDE.md (project facts), MCP (connects to data) and subagents (fixed roles). On screen: Mostly a code editor, with skill .md files, CLAUDE.md, an mcp.json with Playwright, database-migrations.md (model: sonnet), and full-stack-features.md stacking skills. After watching you can: Tell skills, CLAUDE.md, MCP servers and subagents apart, and know when to use each. Short on time? Watch just: - [▶ 0:30](https://www.youtube.com/watch?v=fOxC44g8vig&t=30s) How Agent Skills work - [▶ 1:42](https://www.youtube.com/watch?v=fOxC44g8vig&t=102s) Agent Skills vs MCP Servers - [▶ 2:33](https://www.youtube.com/watch?v=fOxC44g8vig&t=153s) Stacking skills into one bigger agent Source: [YouTube: Claude Agent Skills Explained](https://www.youtube.com/watch?v=fOxC44g8vig) --- ## New on the Shelf 🛒 - [MCP reference servers](https://agentsnacks.ai/shelf/#shelf-001-2) 🌶️ · MCP server **The classics grew up.** Filesystem, memory, everything and sequential-thinking reached their first stable release on npm; fetch, git and time got fresh PyPI builds. Works with: Any MCP client Install: `npx -y @modelcontextprotocol/server-filesystem ` Source: [MCP reference servers v1.0.0 release notes](https://github.com/modelcontextprotocol/servers/releases/tag/v1.0.0) - [Anthropic claude-api skill](https://agentsnacks.ai/shelf/#shelf-001-3) 🌶️ · Skill **Meet Claude’s cheat sheet for the Messages API.** It covers dynamic workflows, four quickstarts, and Haiku 5.5 as the current Haiku. Works with: Claude Code (bundled), Claude API Skills Install: Bundled with Claude Code; or copy from anthropics/skills Source: [Anthropic skills repo: claude-api](https://github.com/anthropics/skills/tree/main/skills/claude-api) - [GitHub MCP Server](https://agentsnacks.ai/shelf/#shelf-001-1) 🌶️🌶️ · MCP server **Your agents can stop guessing what GitHub sends back.** GitHub’s official server now ships structured outputs and output schemas for Code Mode agents (schemas need MCP 2026-07-28+ clients). Works with: Any MCP client Install: Remote server or local binary; see the repo README Source: [GitHub MCP Server v2.0.0 release notes](https://github.com/github/github-mcp-server/releases/tag/v2.0.0) --- ## Before you go 🥡 **[The agents got a babysitter](https://agentsnacks.ai/before-you-go/#byg-001)** **Auto-denied permission prompts, idle compaction, autofixes that recheck themselves** — this week was less “smarter model,” more “you can actually walk away.” - Update now: Gemini CLI now checks folder trust in headless runs too; update to 0.64.0 or later ([Gemini CLI v0.64.0 release notes](https://github.com/google-gemini/gemini-cli/releases/tag/v0.64.0)). - Devin Desktop speaks a newer MCP revision and lets admins pin the extension marketplace. - Codex CLI 0.163 lets you steer it mid-answer ([our Daily](https://agentsnacks.ai/daily/2026-10-15-codex-steer-mid-reply/)). - No Aider release in the window. --- That’s the pack. Pass a snack to a friend.