Harun R. (@harundotdev) profile picture

Solutions Architect, Building https://nitter.gq/t.co/DWHMZ7lDj4 https://nitter.gq/t.co/QA2pi22k9U https://nitter.gq/t.co/RbKGj42cvv https://nitter.gq/t.co/7qqfjtlvhm https://nitter.gq/t.co/eIgI6Pq1CH https://nitter.gq/t.co/0UtHG2ZfTg +2 more. Helped 160+ clients scale to $1B+

ap-harun-1
Joined June 2013
Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)Photo by Harun R. (@harundotdev)
day 24/100 of shipping cloudploy - cloudflare cdn per domain, with a clear cache button - sign-in links that only work in the browser that asked - a team activity page socialreins - broadcasts to every contact with a tag - newsletters: compose, test, send, unsubscribe - a template gallery for automations opsagent - change your name in settings - a switch for human review - paid top-up credits are not stopped by the daily limit toolblip - an ai image provenance checker that runs in the browser cloudploy waitlist: cloudploy.com/?utm_source=x
1
102
13 years after gta 5, gta 6 is out. and the thing that fell over wasn't the game, it was the login playstation network went down for many players right around launch. sony hasn't said why launch day is the one traffic spike you know the date of years ahead. if sign-in can't take the rush, nobody gets to play what's the one service your whole stack waits on?
208
patching a few years ago: a fix ships, you have weeks before anyone weaponizes it patching now: google says llms read new patches and build the exploit before teams install them the fbi says an ai tool called artex goes from first look to privileged access in under three hours if a human has to remember to patch, the human is the gap images that rebuild on their own when a fix lands. that's the job now
2
2
191
anthropic's guidance last month: one agent with a clear prompt and a few tools often beats a sprawl of specialized subagents this is microservices all over again teams split a simple app into 20 services before they had 20 problems. now they split one task across 8 agents before one agent stopped working what i check before adding a second agent, same as before adding a second service: - does it need different permissions? real reason - does it need to scale or fail on its own? real reason - "it feels cleaner"? not a reason one agent, one loop, logs you can read. split when it hurts, not before
7
2
384
connecting mcp servers to ai tools is everywhere now is mcp just a normal api, or something different?
1
2
214
php's critical security release is out, for every supported version per php.net, the critical bugs need non-default config or uncommon setups. so check before you panic, but patch anyway how i'd roll it out: - find every php you run: web servers, queue workers, cron boxes, the forgotten wordpress - containers: rebuild from the patched base image. restarting the old one changes nothing - if you set session.upload_progress.enabled=0 in october, keep it - one server first, watch errors, then the rest how fast does a php patch reach your prod?
161
day 23/100 of shipping cloudploy - stop and start an app from the console - a system, light and dark theme switch - a failed deploy shows the app's last output socialreins - approve a post or request changes - a repeat rule on scheduled posts - schedule posts from a csv file crontinel - trial emails go out on time - apps and monitors over the plan limit pause after a downgrade skaleagents - went live today - the early-user promo code works at checkout and renewal opsagent - members choose whether an agent may make changes - agents can triage findings and connect a cloud account cloudploy waitlist: cloudploy.com/?utm_source=x
1
92
rag, fine-tuning, workflow, agent: four fixes for four different problems. the agent is the last one to try, not the first. wrong facts: rag right facts, wrong tone or format: fine-tuning steps you already know: workflow steps you can't predict: agent which one did you reach for too early?
2
2
157
attackers got into optimizely, an a/b testing platform, and used it to put payment-stealing code on the sites of 178 organizations. hsbc, zoom, ikea, salesforce among them, per report uri most of those companies weren't breached. their visitors were, through a script they chose to load every third-party script on your page runs with the same power as your own code. it can read the card field how i'd set up any page that takes money: - the checkout page loads no marketing, testing or analytics scripts - card fields come from the payment provider in an iframe, so my page never sees the number - a content security policy that lists exactly which domains can run code there - an alert when any script on that page changes your a/b test doesn't need to run on checkout
3
2
442
japan's fugaku supercomputer got hacked through its public status page not the cluster. not a login node. the page that tells you everything is fine from there attackers reached the servers and may have taken ssh private keys and access keys for 4,222 users, per riken your least important page is still a door
1
1
190
seniors and juniors, both answer what's the biggest misconception junior devs have about ai tools?
3
216
day 22/100 of shipping cloudploy - connect aws in four steps on the providers page - cpu, memory and disk on the server page, read over ssh - a logs tab with your app's live container output socialreins - posts list with search, status and network filters - calendar weeks with equal rows and a +n more per day - newsletter sign-up on public pages, double opt-in crontinel - the python package is on pypi from 0.2.1 - analytics show a job's name, not its raw command - tapping the page closes the open phone menu opsagent - mails for trial started, organisation deleted and account deleted - agent answers show open findings by default - a disconnected connection is labelled in deliveries skaleagents - owners can delete their saved prompts - the audit email links straight to the audit cloudploy waitlist: cloudploy.com/?utm_source=x
72
the hole in a vibe coded app usually isn't the missing login. it's the missing ownership check. you sign in, change /orders/41 to /orders/42, and someone else's order loads. the server never asked whose record it is. what's the first thing you test before sharing a vibe coded app?
3
181
day 21/100 of shipping cloudploy - servers in your own aws account use no credits - verify in place instead of signing in again - the pricing page says what happens when a paid plan ends crontinel - free by default, with a card-backed 14-day pro trial from billing - alerts and run lists show a job's name, not the raw command - ruby gem 0.1.1 fixes task_started opsagent - security alert emails are one readable line, not raw event json - tag values and names are redacted before inventory is stored - billing works before you connect an account socialreins - long posts for accounts on a paid x plan - an x plan setting, asked right after you connect x - the same tiktok @username under another app counts as the same account skaleagents - owners can delete their own bots, and deleted bots stop being served - an empty lockfile scan is no longer reported as clean cloudploy waitlist: cloudploy.com/?utm_source=x
109
openai wants a real account email and password to review a chatgpt plugin. i don't store user passwords. plugin review is still in the stone age. how are you handling it?
1
1
204
if you issue api keys, store only a hash and show the key once. if you hold your users' keys for them, you can't hash them, you need the original back, so you encrypt when did you last check that no key sits in plain text in your db or your logs?
3
2
357
ai is really bad at garbage collection. every task gets a new worktree, every worktree gets its own node_modules, and nobody deletes them after. 256gb used to feel huge. now 1tb doesn't feel like enough. what's eating your disk?
8
5
331
day 20/100 of shipping cloudploy - zero-downtime redeploys with a health-gated swap - sign-in codes are 8 digits and tied to your browser - certificate and dns notices show in the deploy log socialreins - post versions, with their own media per version - a public page per post, with comments - a team switcher in the sidebar opsagent - an email to the owners when another organisation tries their cloud account - a root account without mfa is now a high finding - sign out everywhere crontinel - the trial end date and what follows, shown before sign-up - a welcome email after google and github sign-up skaleagents - organization addresses you choose yourself, plus your google photo cloudploy waitlist: cloudploy.com/?utm_source=x
3
151
cursor published how its october hack happened. it didn't start at cursor - a key cursor gave its ai vendor, fireworks, got exposed - it opened a bucket of old coding agent traces - a few traces still had unredacted api keys - those keys led into internal tools, and a stolen credential put a fake claude code extension in cursor's marketplace for ~11 hours the lesson for me: agent transcripts are a secret store, and almost nobody treats them like one how i set it up: - agents read keys from env or secrets manager, never from the chat - a key i give a vendor only works from that vendor's network - logs and traces get secret-scanned before they leave my account when did you last check what's sitting in your agent logs?
10
6
509
first time ever: my $200 claude ultra plan did $10k+ of api usage in one month. mythos, fable and opus all month, plus one of the five limit resets claude gave me. that $10k is the api estimate, not a bill. how much does your plan actually give you?
3
5
371