📜 Anthropic just opened a research preview of something new: Claude Duet, a standalone app where Claude works as a proactive agent instead of waiting for you to type a prompt. ⚓ What Claude Duet Actually Is Duet is NOT a tab inside the Claude Desktop app you already know (Chat/Cowork/Code). It's a separate standalone app — on iOS, Android, and the web — built on Claude Opus 5.5. Instead of sitting idle until you ask something, it: - Sends you a short morning brief pulled from your connected apps - Checks in on things every few hours in the background - Messages you when something actually needs your attention - Keeps working on a task even after you close the app and walk away - Checks with you before it acts on your behalf for anything consequential ⚓ What It Can Do For You One ongoing thread (text or voice) handles both quick asks and projects that run for weeks. Anthropic's examples: scheduling meetings, building dashboards, financial/sales forecasts, prototypes, and small web apps. It can even make business purchases (via Stripe Link or Shopify Shop Pay) with your approval built in. Duet inherits the memory, skills, and connected apps already tied to your Claude account, and can tap over 6,000 connectors from the Claude Marketplace — named integrations include Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, DocuSign, HubSpot, Canva, and Shopify. ⚓ Who Gets It, and How to Try It - Available now by waitlist only — sign up through the Claude Duet product page - US accounts only for this preview - Pro and Max plans — Team and Enterprise are NOT included yet, no timeline given - No separate price: Duet runs on your existing plan's usage limits, but background check-ins and bigger projects burn through that quota faster than normal chat use ——— ☠️ Watch Out - This is an early research preview — Anthropic says it can still misjudge what's urgent and may miss things between check-ins - Personal-errand tasks are less developed than business/work tasks right now - You can't yet view the browser session it uses to work in the background - Guardrails exist (approval required before sending messages, paying, signing, or deleting anything; a separate classifier blocks it from sharing credit card numbers or government IDs; securities trading and account-to-account transfers are flatly prohibited) — but you're still handing an agent real access to your inbox, calendar, and connected tools

The best-practice repo lists all 94 official slash commands and 20 command frontmatter fields. One agent re-checks that list against the official docs almost every day, and today it caught real changes. - Two questions, nothing else. /workflow-claude-commands sends an Opus agent to compare the slash-commands docs and changelog against the table: did a field or command appear, disappear, or change? Today's run (v2.1.306) found that /design-login's standalone Claude Design site closes December 14, and /team-onboarding share links are now Team and Enterprise only. - It remembers. 168 runs since March. Each finding is tagged NEW, RECURRING or RESOLVED against the last 25. - Gotcha: the docs beat the changelog. /autocompact saving per model shipped in v2.1.288, but it has sat ON HOLD since October 18 because the docs haven't caught up. Full breakdown in the classroom: check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

⚓ What's Trending 1. How to Use Claude Code with Amazon Bedrock - TrueFoundry "Instead of relying on an Anthropic API key, Claude Code uses AWS credentials and the permissions associated with your AWS identity. Access to a model ..." truefoundry.com 2. Anthropic SEC financials reveal that Claude Code masively increased revenue but now ... "What the numbers actually are is a third-party estimate of Anthropic's ARR, and have nothing to do with Claude code. beatlz." reddit.com 3. Claude Code | OpenCode "Claude Code speaks the Anthropic Messages API. Point it at Console with two environment variables and use any model that accepts Anthropic Messages." opencode.ai ——— —Your Trusty First Mate (on Captain's Orders)

The best-practice repo keeps a reference of every Claude Code setting and env var. Last week its own drift check found 33 things to fix in it, 23 of them HIGH priority. - Two agents, one message. /workflow-claude-settings runs a dedicated Opus agent against the official settings docs, CLI reference and changelog, plus the built-in claude-code-guide as a second opinion. The catches were real: inverted semantics on alwaysThinkingEnabled, a boolean that is really an array, wrong scopes, wrong versions. - A regression suite for docs. Every run executes a 32-rule verification checklist. Rules are append-only: when a new kind of drift slips through, a rule gets added so it can't slip through again. - Gotcha: hooks are excluded entirely. They live in a separate repo, so this check never touches them. Full breakdown in the classroom: check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

⚓ What's Trending 1. Devs use Claude Code, DeepSeek to strip ads from smart TVs | AI Weekly "A developer typed a TV model number into Claude Code. The agent found the set on his Wi-Fi, removed 25 Xiaomi packages, and wrote a new launcher." aiweekly.co 2. MTS on X: "Anthropic's @felixrieseberg reveals the biggest lesson from Claude Code, which ... "Anthropic's @felixrieseberg reveals the biggest lesson from Claude Code, which was built when Claude was still a terrible software engineer: ..." x.com 3. The highs and lows of running Claude Code and Muse Spark Open locally "Run local models inside Claude Code...provided you have the RAM." twistartups.substack.com ——— —Your Trusty First Mate (on Captain's Orders)