this is f*cking insane... motion studios are cooked a 30-second launch video used to cost $5,000 and two weeks of back-and-forth. now it's one prompt, one night and a $20 Claude plan Anthropic tried to keep it for Team plans. GitHub shipped 347 open versions before most teams even got access. #20 is almost a 1:1 copy, and it's free 20 GitHub projects worth opening first: 01 OpenDesign → 100K+ stars. a full design agent workspace: start from a brief and your coding agent builds prototypes, slides and motion in your own design system github.com/nexu-io/open-desi… 02 HyperFrames → HeyGen's open-source engine with 72K stars. your agent writes HTML and it renders a frame-perfect MP4. one plugin install inside Claude Code github.com/heygen-com/hyperf… 03 OpenMontage → 71K stars. a whole video studio: your agent researches, writes the script, pulls free stock footage, edits the timeline and renders the film github.com/calesthio/OpenMon… 04 /brag → one command turns the project you just built into a 20-second launch video with music, motion and the copy to post it github.com/latent-spaces/bra… 05 video-shotcraft → 150 tuned shot recipes. the agent captures your real pages, storyboards the film, cuts it to the beat and checks every frame github.com/Vincentwei1021/vi… 06 html-video → paste an article link or a GitHub repo and get a fully animated MP4 rendered on your laptop, with no per-render fees github.com/nexu-io/html-vide… 07 anything2explainer → topic in, narrated explainer out: voiceover, subtitles and a chapter bar, every frame drawn in code and no stock footage github.com/Vincentwei1021/an… 08 Herk Edit Kit → Nate Herk's own editing kit. it cuts dead air, plans the story and adds B-roll and motion graphics to your raw recording github.com/nateherkai/herk-e… 09 15 Motion Design Styles → 15 films, 10 seconds each, all made by Opus 5.5 writing code. every film ships with its full prompt and source github.com/Vincentwei1021/mg… 10 Oil Motion → animations that follow the scroll, the mouse or a phone tilt, like a product that breaks into parts as you scroll down github.com/oil-oil/oil-motio… 11 video-talkcraft → give it a script and a voiceover and it locks 108 animated cards, captions and sound effects to every word you say github.com/Vincentwei1021/vi… 12 Motion Video Kit → built from studying 28 professional SaaS launch films. the agent that builds a scene never judges it, fresh critics review every render github.com/echris6/motion-vi… 13 motion-anything → describe the feeling and get the animation. 403 motion recipes, and you edit the motion right on the running page github.com/nexu-io/motion-an… 14 motionmaxxing → looksmaxxing for motion graphics. it turns flat AI videos into films that look like a motion designer made them github.com/Tejashmakwana/mot… 15 Null Motion → see how a finished motion ad was built, every rough draft playing under the final film, frame by frame. no install, no keys github.com/blixvip/NullMotio… 16 Awesome Claude Video Skills → 347 video repos for coding agents, each one read and security-graded. the team ran 16 end to end and ranked them github.com/zhuyansen/awesome… 17 motion-graphics → free skills that each do one job: B-roll timed to your words, 15-second hype promos and demos built from your real UI github.com/Barty-Bart/motion… 18 claude-motion-design → pure code with HTML, Playwright and ffmpeg. no After Effects, no license, and every sound effect lands on its measured peak github.com/howseen-ai/claude… 19 Claude Remotion Skill → teaches Claude springs, staggered choreography, color grades and word-synced captions, so you stop getting flat fades github.com/haidrrrry/claude-… 20 open-motion → a report goes in, a code-built animation comes out as an MP4. the same input and output as Claude Motion, minus the Team plan github.com/Archive228/open-m… pick one and make a 10-second clip from a page you already have. once it looks right, give it the whole launch render one frame before you render the film. in code, most fixes take one line instead of a new export Team plans got Claude Motion first. you can ship your first motion video tonight full setup, step by step ↓

i still can't believe a gaming PC did this much in one night Jensen Huang (CEO of NVIDIA) posted an idea that hit 14 million views: "the most powerful AI computer most people own is the one they game on" one builder took it literally: an RTX 4090 left running overnight with 4 years of bank statements, receipts and emails. none of it ever left the PC by 7am Claude Code had a list of subscriptions nobody used, charges that hit twice and refunds promised by email and never paid. every claim letter was already drafted and waiting for a yes a week later the builder had $3,260 back in the account 10 GitHub repos behind that night: 01 gpu-unlock-ada → an experimental Linux module that doubles an RTX 4090's tensor speed, from 165 to 331 TFLOPS, so the night shift reads twice as much before morning github.com/JustAResearcher/g… 02 Docling → turn years of PDF bank statements into clean tables the agent can actually read github.com/docling-project/d… 03 Tesseract → read photos of paper receipts, so the ones in a shoebox count too github.com/tesseract-ocr/tes… 04 Ollama → run the model on your own GPU, so not one transaction leaves the room github.com/ollama/ollama 05 llama.cpp → fit a big model onto a gaming card, so it reads like a pro without a server github.com/ggml-org/llama.cp… 06 vLLM → let several agents read statements, receipts and emails at the same time instead of one by one github.com/vllm-project/vllm 07 DuckDB → query four years of transactions in seconds to catch the same charge hitting twice github.com/duckdb/duckdb 08 Qdrant → match every charge to the email or receipt behind it, so a missing refund stands out github.com/qdrant/qdrant 09 Firefly III → keep the cleaned-up ledger in a personal finance app you host yourself github.com/firefly-iii/firef… 10 Open WebUI → check what the night shift found from your phone over coffee github.com/open-webui/open-w… start with one year of statements from one card. once the list matches what you'd catch by hand, give it the rest most people use a $1,600 card to render grass. this one paid for itself twice how to set up the same night shift on your PC ↓

this will cut your costs 900x a 23-year-old builder from China just ran 300 Claude Code agents on Opus 5.5 and brought back an online game whose servers died 10 years ago the agents got nothing but the game itself. no source code, no docs, nobody left to ask. he hooked them to a verification layer, so no agent could call the server working until it actually was by Sunday night 8 players were online again, on a server the game had never seen a studio asked him for $180,000 and six months to do this. he did it in one weekend on a $200 Claude plan here's how the agents split the work: → readers point REA at the game's .exe and list every message it sends or expects back: login, lobby, match and results. each one gets its fields and the function that handles the reply → coders write the server backwards from the client. for every message the game expects, they write the answer it accepts, then check it against the code that reads it → testers run 400 automated matches and compare every result with the game's own logic, because a server can accept your login and still corrupt your save → anything the code only half shows, like tournaments and friend lists, stays marked as a guess and switched off until a human tests it → the game files never change. one line in the hosts file sends the game to the new server instead of the dead one → the server runs on a $5 machine and logs every message, so when a match breaks you see which message failed and which agent wrote it you can run the same setup on your own Claude plan tonight. all it takes is a game you own and this prompt: "Bring this game's multiplayer back with a replacement server. Split the work between agents 1. Readers: map the network code with REA. List every message the game sends or expects back, its fields and the function that handles the answer. Never modify the game files 2. Coders: write the server backwards from the client. Answer each message the way the game's own code expects, check it against that code and mark anything unclear as a guess 3. Testers: run automated matches and compare every result with the game's own logic. Only testers can mark a message as working, and every guess stays switched off until I test it 4. Point the game at the new server with one line in the hosts file. Log every message in both directions so any failure shows which message broke and which agent wrote it" start with the login and one match before anything else whole run below: from a dead server to 8 players online

Jev + Opus 5.5 is f*cking unfair... one prompt and your own Claude Duet runs 40x cheaper than on Opus alone the shift: Jev makes the thousand tiny calls in milliseconds (urgent or not, reply or wait, who is this) → Opus 5.5 only wakes up for work that needs a brain same morning brief. 40x cheaper. 12x faster. and it still asks before it sends anything paste this prompt into Claude Code and let it do the rest ↓

THIS IS F*CKING INSANE Claude Duet does the job of an $8,000/month chief of staff. Anthropic put it behind a waitlist of 300,000+ people GitHub didn't wait. 140+ open versions dropped in 72 hours. the code is FREE and they run on a $20 Claude plan. 400x cheaper every night these agents read 200 emails, draft every reply in your voice and prep tomorrow's meetings. you wake up and press yes 20 GitHub projects to open first: 01 Open Duet → the biggest one, 6,000+ stars. build a roster of named AI coworkers with their own memory and approvals. a lead agent hands work to specialists github.com/Anil-matcha/open-… 02 OpenDuet → bots brief you every morning, watch your apps and wait for your OK before they send, post or pay. try the whole thing with zero keys github.com/Studio1-OSS/OpenD… 03 Tandem → the Duet experience on a $5 server. morning brief, check-ins that stay silent unless something matters, and every byte of your data in one file github.com/arjunkshah12345-h… 04 Covo → runs jobs end to end on your server, your PC or your phone. plug in your own Claude key or a local model and it does the rest github.com/hmody1m/covo 05 openduet → an always-on work partner that learns the routines you teach it once and keeps running them long after you close the browser github.com/ychampion/opendue… 06 hermes-duet → the same idea on the Hermes agent: one thread, a morning brief at the hour you pick and check-ins 3 times a day github.com/DevvGwardo/hermes… 07 claude-chief-of-staff → the exact setup the CEO of Ada runs on Claude Code. it took his morning inbox from 90 minutes to about 5 github.com/mimurchison/claud… 08 Claudia → remembers people, not just tasks. say "I'll send that by Friday" once and on Friday morning she's already reminding you github.com/kbanc85/claudia 09 Chief of Staff plugin → one install, one /chief-of-staff:setup, and Claude Code interviews you and builds your memory and daily brief by itself github.com/loganhc-09/claude… 10 claude-code-cos → the blueprint for people who never wrote code: what connects to what, what each agent owns and where you keep the final word github.com/jimprosser/claude… 11 chief-of-staff-2 → a full chief of staff living in plain markdown inside Obsidian. paste one file into your AI and it walks you through everything github.com/jdpolasky/chief-o… 12 Morning Brief Kit → wake up to the 3 things that need you, the next step already prepared and what can wait. it never sends a thing on its own github.com/onittindwani-ai/c… 13 Tacitus → reads your mail at 4 AM while you sleep. by breakfast every routine reply is written and waiting on a card with Send, Edit and Cancel github.com/arpg/tacitus 14 Zyggy → lives on a tiny Linux server around the clock. every night it "dreams", cleaning and merging its memory so tomorrow it knows you better github.com/zyggy-org/zyggy 15 Scout → cross-checks Slack, Gmail, Calendar, Linear and GitHub on a schedule, so nothing slips between your tools before your morning brief github.com/Raven-Scout/Scout 16 NachtSchicht → queue tasks in the day, agents work them all night in isolated worktrees, and you merge what's ready over coffee github.com/FlorianRiquelme/n… 17 agent-brief-me → your agents drop every question into one inbox instead of stopping. you answer them all in one sitting and they finish the job github.com/RyanLeeYi/agent-b… 18 owl → approve or block Claude Code actions straight from Telegram while you're away, and let safe commands pass on their own github.com/pkronstrom/owl-af… 19 claude-interact → lets Claude Duet itself start and steer Claude Code sessions on your own machines. no open ports, nothing exposed github.com/ebarr24/claude-in… 20 Inbox → pulls only the invoices, letters and confirmations you care about, files them and asks before anything goes on your calendar github.com/googlarz/personal… start with one. give it a job with an obvious finish line: tomorrow's brief, 10 drafted replies or one prepped meeting then check what you actually accepted. 3 agents you trust beat 20 you have to babysit 300,000 people will spend the next month on a waitlist for something GitHub already built for free the full build, every file explained ↓

my MacBook was talking to a server every night at 3:12am. Claude Code caught it 18 minutes before someone tried to take $6,800 from my exchange it started with the fan. spinning up at night, lid closed, nothing open I asked Claude Code one thing: "what is this Mac doing when I'm not using it?" 6 minutes later it came back with a line I didn't expect: "a browser extension is sending your open tabs and session cookies to a domain registered 11 days ago" the extension → a "free PDF converter" with 4.8 stars the domain → registered 11 days before I installed it the server → a rented box on a cheap hosting plan what it had → every URL I opened, plus the session cookie for my exchange when → every night at 3:12, in one batch it was still sending while I was reading the report they never needed my password. the cookie was enough to walk straight in, and 2FA would never get asked Claude Code told me to log out of every session from my phone first. I did 18 minutes later the exchange sent a login alert from a city I've never been to. the session they tried was already dead $6,800 in USDC was one transfer away. it's still there the real PDF tool has millions of users. mine was a copy with a different publisher name and 9 reviews. I installed it from an ad how not to be me: → install extensions from the developer's own site, never from an ad → check the publisher name, not just the stars → log out of exchanges and banks when you're done. a closed session can't be stolen → keep crypto on a hardware wallet, not one browser tab away → run a read-only audit once a week that last one is now a scheduled job. every Sunday night Claude Code checks what runs, what it sends and who signed it, and asks me before touching anything watch the agents that saved my wallet at work. the full prompts for Mac and Windows are below ↓ free, steal them

CLAUDE DUET IS A RESEARCH PREVIEW. THESE 10 REPOS WORK TONIGHT everyone is refreshing for Duet access while the pieces to build your own sit on GitHub with over 600,000 stars between them 10 GitHub repositories to build the same thing on Claude Code: 01 MCP Servers → give Claude Code hands through official connectors for files, git and the web, so your agent can read the places your work actually lives github.com/modelcontextproto… 02 Basic Memory → keep everything Claude learns in plain Markdown files it can search and update, so you never re-explain a project again github.com/basicmachines-co/… 03 Mem0 → add a memory layer that keeps your preferences, people and past decisions across sessions, so every brief starts from what it already knows github.com/mem0ai/mem0 04 Graphiti → track how facts change over time, so your agent knows the deadline moved instead of quoting the old one github.com/getzep/graphiti 05 n8n → schedule the morning brief, trigger your agent when an email lands and route its drafts back to you for approval github.com/n8n-io/n8n 06 Huginn → run agents that watch sites, feeds and inboxes on your behalf and only ping you when something actually changed github.com/huginn/huginn 07 whisper.cpp → turn voice notes into text locally, so "remind me to call the bank" from your walk ends up in tomorrow's brief github.com/ggml-org/whisper.… 08 Paperless-ngx → scan bills, contracts and letters into a searchable archive your agent checks before it reminds you of a deadline github.com/paperless-ngx/pap… 09 Karakeep → drop links, notes and screenshots into one inbox with auto-tagging, so your agent sorts what you saved instead of you github.com/karakeep-app/kara… 10 Langfuse → trace every run your agent makes while you're away, so you can see exactly what it read, what it drafted and why github.com/langfuse/langfuse start with memory and one connector. a morning brief that knows your people beats ten automations that don't then add the schedule. once it runs at 7am without you, put "ask before acting" in place before you connect anything that can send memory → connectors → schedule → approvals. that's the whole Duet idea, built from parts you can inspect how to wire all 10 together ↓

holy sh*t. I built my own Claude Duet and left it with 14 years of family photos overnight Anthropic's Duet announcement hit 8 million views in under a day. it's still a research preview, so I built mine from 9 Claude Code agents and gave them the hardest job I could think of backups from 6 old phones, 3 hard drives and one rule: "never delete anything. ask me first" give each group of agents a specific job: → sorters date every file from its metadata, its filename and what's in the picture → matchers find the same photo across drives and keep the best copy → fixers repair corrupted videos and turn sideways photos the right way up → the builder puts it all on one private timeline you can scroll year by year a photo can carry the wrong date and still look right. checkers compare every date against the folder, the device and the photos around it, and mark the ones that are only a guess by 7am: 41,286 files became 18,904 unique photos and 612 videos on one timeline my mom can scroll by year. 2,140 dates stayed marked as guesses. 0 files deleted. 1 question waiting for me give Claude Code your old backups, then ask: "Turn these backups into one family timeline 1. Scan every folder, phone backup and drive. Date each photo and video from its metadata, filename, folder and the shots around it. Attach the source of every date and mark the ones you had to guess 2. Group exact and near duplicates across all drives, including resized and edited copies. Keep the sharpest full-resolution version in front, move the rest to a review folder and never delete 3. Repair what can be saved: rotate sideways photos, rebuild videos that won't play, recover broken thumbnails. List every file you couldn't fix and why 4. Build a private timeline I can scroll by year and month and search by date. Show me one page of open questions and ask before anything is removed" start with one phone's backup and check a month of dates by hand before adding the rest keep delete on the "ask" list. agents sort everything, you decide what goes the whole build below

Claude Code + Opus 5.5 is godmode for building your own Claude Duet it reads your inbox and calendar before you wake up. drafts every reply. sends nothing without your yes it keeps working while you sleep full course, step by step:

Anthropic just shipped Claude Security, but only for Enterprise. these 10 free repos get you most of the way there tonight the idea is simple and overdue: one deep scan of each repository, then a fresh check on every change, so a bad commit gets caught the day it lands, not the day it gets exploited the open-source crowd didn't wait. these shipped in the last few weeks and run on the Claude Code you already pay for. 5 guard your code, 5 guard you from your own agent: 01 airsec → run a multi-pass security audit with the agent already on your machine. no new API key, no cloud upload, a report you can act on github.com/bas3line/Airsec 02 paranoid → type /hack-me and Claude attacks your own running app, proves each bug with a real request, patches it and checks the exploit stops working github.com/kulchankas/parano… 03 security-harness → map a codebase, hunt vulnerabilities class by class, chain findings into escalations and export the report as SARIF, JSON or PDF github.com/dmdhrumilmistry/s… 04 claude-security-scan → ask for a security check and get a read-only sweep of your Mac, every local repo, GitHub, Vercel, cloud accounts and Gmail github.com/chaenmasahiro0425… 05 prod-readiness → audit the whole repo through 8 lenses, from security to DevOps, and get one verdict: SHIP, FIX THEN SHIP or HOLD github.com/Taimoorkhan1122/p… 06 jev-guard → score every tool call your agent makes as deny, ask or allow, and flag prompt injection hiding in what it reads github.com/leepokai/jev-guar… 07 CC-Monitor → log every file Claude Code touches, every command it runs and every network call, and block the risky ones before they happen github.com/cn0xroot/CC-Monit… 08 agents-sandbox → run each project's agent inside its own microVM, so it only sees the folders and secrets you hand it github.com/inoio/agents-sand… 09 shim → see exactly what Claude Code sends to the model and what each turn costs, and mask secrets and personal data before they leave github.com/GetSHIM/shim-cli 10 AuditSentry → audit Solidity and Vyper contracts with Claude Code, get a working exploit for each finding and test it on a mainnet fork github.com/iktok90-design/ai… start with the risk you've been putting off: > you ship fast and never pentest → paranoid > old commits might still hold your keys → claude-security-scan > your agent runs commands you never read → CC-Monitor Claude Security is live for Enterprise. these work for everyone tonight

this is f*cking insane. a retry bug was quietly charging customers twice and costing a payments startup $40,000 a month. Opus 5.5 found it overnight while I slept. no engineers, $0 Boris Cherny (creator of Claude Code) posted one line last month that hit 3.1 million views: "the hardest part of a codebase isn't writing code. it's knowing what already happens" that's the job I gave 6 Claude Code agents at 10pm: the startup's 720,000-line payments repo and one question, "what really happens when a user clicks pay?" who did what: → entries found every route, background job and CLI command that can start a payment → stores mapped where the money data lives: postgres, redis and s3 → the tracer walked the click from the button into the payment service → the verifier re-checked every link against the real file and line → tests looked for a test and a log line behind each step → the writer turned it all into one ARCHITECTURE.md what was waiting in the morning: > a path from button to receipt across 41 files, every link tied to its line > 5 of 6 links confirmed, 1 marked "can't prove this" > the one they couldn't prove: a retry in the queue worker with no test and no log > that retry was charging 1 in 300 payments twice. two weeks of the team's time, $40,000 a month in refunds a contractor quotes $6,000 for an audit like this. mine cost one night and $11 in tokens the "can't prove this" flag is the part I'd pay for. an agent that fills gaps with guesses hands you a map that's confidently wrong somewhere in your codebase a step like that is already costing you money here's the full workflow, step by step: