Harun R.

a fake claude code update sat in cursor's extension marketplace for about 11 hours yesterday it added an ssh key called svc-telemetry to authorized_keys, and uninstalling it isn't enough 30 second check: grep svc-telemetry ~/.ssh/authorized_keys do you let your editor extensions auto-update?

cleaning up disk space as a devops guy old docker images: deleted node_modules: deleted vendor folders: deleted terraform.tfstate: absolutely not

openai now lets you switch between multiple chatgpt accounts on mobile. desktop is next so the official fix for usage limits is... more accounts 😅 i've burned a whole week of claude usage in a couple of days before, so i get it how many ai accounts are you juggling right now?

wired says a contractor plugged a personal claude code agent into a pentagon network it reportedly ran for about 2 months and sent logins and a network map to a flagged site. accenture says the claims are without merit the agent didn't break in. it was handed the keys what can your agents log in to right now?

aws setup, solo founder edition: 1. multi-az rds 2. auto scaling group 3. cloudfront + waf 4. a kubernetes cluster for later 0 users on all four

it's halloween 🎃 skip the ghost stories. what's the scariest thing you've seen in production?

railway got breached on oct 21. northflank dunked on them and asked people to migrate over 9 days later northflank got breached too. its status page says assume every env var was decrypted and rotate all keys as a devops guy, the only lesson here: never dunk on a breach. you're next on someone's list when did you last rotate your prod keys?

day 3/100 of shipping skaleagents - paid workspace owners get their plan's mcp seats - public pages now served static - rate limit for ai crawlers, fake bots blocked crontinel - billing page shows past-due and scheduled downgrades - failed renewals follow stripe's retries - failed job outcomes now count as failures cloudploy - the scheduler restarts itself when its jobs keep crashing - new marketing site in progress stealth one - ai explanations and a read-only chat in the dashboard - runs on your own ai key, or prepaid credits stealth three - redesigned app, works on phones - google sign-in cloudploy waitlist: cloudploy.com/?utm_source=x

aws grew 39% to $46b last quarter, its fastest in nearly five years amazon stock still dropped over 6% after hours because of ai spending i'm pretty sure part of that $46b is nat gateway charges nobody noticed 😅 which line on your aws bill surprised you most?

cve-2026-98432 is a 9.8 linux kernel bug. remote root, no login needed debian has patched. ubuntu says in progress. and evilsocket says the default aws fargate 6.1 kernel is still unpatched the hardest part of patching is waiting for someone else's patch what are you running until yours lands?

your app doesn't have a user limit, it has a requests-per-second limit 4 workers at 100 ms a request is 40 req/s. whether that's 400 users or 4,000 depends on how often each one sends a request in your app, what hit its limit first: the app server or the database?

google gave jules 2 months notice before shutting it down. 📅🪦 that's more notice than most of my side projects got 😅

this is what i've been trying to say all along. now hear it from the horse's mouth your technical skill isn't useless. it's what makes you good with agents teapawt.net/trq212/status/21155440…

i use claude every day, and try others like chatgpt and opencode on the side 😄 which one is your daily driver, and why? @claudeai @ChatGPTapp @GeminiApp @opencode

rauchg says the future is agents talking to other agents my grok bots already do this. my x bot reports to a chief of staff bot, and the chief of staff reports to me the rule that makes it work: nothing goes out to the outside world until i say yes would you let your agent send feedback to a vendor without asking?

epic says its engineers no longer write the bulk of the code for unreal engine 6 claude code and codex rewrote a big part of the d3d12 rendering layer in about a week. epic's own estimate for one engineer was 3 to 6 weeks hard to keep saying ai can't do real engineering what's the hardest thing you've let an agent rewrite?

zip doesn't delete anything. it gives the letters that show up most the shortest codes banana: 48 bits at 8 bits a letter, 9 bits with huffman coding. it still decodes to banana because no code is the start of another zip, rar, 7z or gzip: which one do you use most?

day 2/100 of shipping cloudploy - environments inside a project, with a safe copy of production - new servers warm up before their first deploy - delete a project by typing its name toolblip - real stripe refunds for sponsors, reversed if stripe fails - delete your account, stripe customer goes with it skaleagents - footer badges no longer block the first link clicks stealth three - team switcher, team settings and invite pages - self-host with one docker compose file stealth one - billing built, switched off for now - nightly cleanup, deleted organisations purged after 30 days cloudploy waitlist: cloudploy.com/?utm_source=x

a 10-person startup says it shipped 77,000 changes last month and spent $3.1m on ai tokens that's about 108 changes an hour. nobody reads that by hand at that pace your real product is ci, review rules and rollback how much of your agents' code gets a human look?

secret passing to agents is my favourite boring feature my grok bots ask for a key through a secure card, so it never sits in the chat history pasting api keys into a prompt should feel as wrong as committing your .env how do you hand keys to your agents?