GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  juggernaut (@curlysaarthak) reported 6 minutes ago

  @anaisbetts @mitsuhiko isn't this GitHub issue?

• 
  Codemonger (@codemonger00) reported 9 minutes ago

  Startup Founders Pack - Claude = coding. ($20/mo) - Supabase/Convex = backend. (Free) - Vercel = deploying. (Free) - Stripe = payments. (2.9%/transaction) - GitHub = version control. (Free) - Resend = emails. (Free) - Clerk = auth. (Free) - Cloudflare = DNS. (Free) - PostHog = analytics. (Free) - Sentry = error tracking. (Free) - Upstash = Redis. (Free) - Pinecone = vector DB. (Free) Total monthly cost to run a startup: ~$20 There has never been a cheaper time to build .

• 
  Insider.Space (@InsiderDotSpace) reported 10 minutes ago

  @RuneCrypto_ LOOOLLL $200B ? memes ? :D:D:D:D:D:D:D The network is terrible and slow, they're begging GitHub for help. 🤠🤠🥳🥳🤡🤡🤡

• 
  Adam Shepherd (@AdamShephe61844) reported 12 minutes ago

  The GitHub agent that leaked private repos wasn't a model jailbreak. It read a poisoned issue and did what the text said. Read scope plus any outbound channel equals an exfiltration path. Least privilege stops being optional the moment your agent can be talked to.

• 
  kian 🗿 (@22kian_) reported 13 minutes ago

  Why I Think Concrete Is More Than Just Another Platform When I first joined @ConcreteXYZ, I had the same goal as everyone else. Farm. Check in. Collect bags. Repeat. That was it. But after spending more time in the community, I realized something. Concrete isn't only rewarding participation. It's quietly creating builders. I've seen people who never built anything before launch websites, create games, drawings, arts, crafts, design tools, write guides, and help other community members. Nobody told them to do it. They simply saw a problem and decided to solve it. That's what happened to me too. in season 1..... I lost a 44-day streak because I forgot to check in Instead of complaining about it, I built CreteGuard. Not because someone paid me. Not because anyone asked me. Just because I thought other people probably had the same problem. Whether CreteGuard becomes big or not isn't really the point. The point is that Concrete gave me a reason to build something real. I also learned something important while making it. You don't need to be a software engineer. I wasn't. I used AI for almost everything. ChatGPT helped me plan the product. v0 built the landing page. Claude generated most of the bot code. GitHub stored everything. Railway deployed the bot. Vercel hosted the website. Whenever I got stuck, Gemini became my technical guide. I'd literally send screenshots and ask, "What do I click next?" Little by little, everything came together. That changed how I think about building. Today, AI removes most of the technical barriers. The difficult part isn't coding anymore. The difficult part is noticing problems that people actually have. If you can find a real problem and clearly explain it, AI can help you build the solution. I think that's what makes communities like Concrete interesting. They're not just collecting users. They're creating people who experiment, share ideas, and ship things. This is only my first project. I already have more ideas written down. Hopefully, they're even more useful than the first one. And if there's one thing Concrete has taught me, it's this: Don't wait for permission. If something is missing... build it.

• 
  Sarvesh Gandhi (@sir_bae_) reported 13 minutes ago

  @github abruptly shutting repos with no explanation or response is not the community support you stood for. The PRs, issues, discussions and morale are a loss for the contributors. Offer valuable help.

• 
  Vivek Maskara (@maskaravivek) reported 16 minutes ago

  Learnt the hard way that bad trigger config for Github action can end up in 12k runs in 2 days and cost you ~50$. 😞 Also, learnt that GHA has a per minute billing rate instead of per second. I had a action workflow that ran every time there was any activity on an issue in my repo.

• 
  Ashley Schendel (@ashleyschendel) reported 16 minutes ago

  Hey! I wanted to offer a small suggestion. I’m a total nerd about this stuff and I study/test the X algorithm pretty regularly with a few friends. From what I’ve seen, outside links in the main post can really hurt reach. X tends to treat them almost like spam, and some of that was even visible in the algorithm code they released on GitHub. You may get better results by writing the post without the link, using stronger keywords, skipping hashtags, and then putting the article link in the first reply. That way people still see the story right under the post, but the main post has a much better chance of being shown especially if it’s engaging. Also, the more often an account posts outside links directly in posts, the more it can seem to drag down reach on other posts too.

• 
  Polsia (@polsia) reported 22 minutes ago

  Dependabot surfaces the CVE. Renovate opens the PR. Neither writes the fix or runs your tests. Built PatchSentry to close that loop — autonomous patching for GitHub, end-to-end, and humans only get paged when judgment actually matters. Live soon.

• 
  Rajan Bhattarai (@cdrrazan) reported 23 minutes ago

  GitHub makes you clear ~6 confirmations and type the full repo name by hand to delete a repo. Fine! Except it does the same thing whether the repo has 100K commits or one commit you pushed 20 minutes ago. That's the tell. The friction isn't calibrated to risk; it's calibrated to nothing. Real "confirm your intent" design scales with blast radius. Deleting a repo with 40 contributors and 2K issues should hurt. Nuking a throwaway you made this afternoon shouldn't. GitHub knows the commit count, the stars, the age, the contributors. It uses none of it. Uniform friction is the easy version to build. It's also the one that trains you to click "confirm" without reading — which quietly kills the whole point. Good instinct. Lazy implementation!

• 
  Lincoln 🇿🇦 (@Presidentlin) reported 23 minutes ago

  I could go on. In my view, the Jules team is a customer of GEAP. Both Jules and AI Studio can make a cloud agent. Esp since AI Studio is getting an app. They recently brought GitHub imports, it's not crazy to see AI Studio build their version of Codex Web or Cursor Cloud Agents. The problems and jobs the current version of Jules solves feels like the AI Studio team should handle it. You could argue the Antigravity team should handle that, but it feels they should be lower level. Jules v2/3 should go build something cloud related but not yet another Codex Web. I could even see Google spin up another team that talks with GEAP called Gemini Agent Cloud or something. Sometimes it makes sense to kill the CLI I loved for one I hate (still bitter) but that one was more duplicated. There is risk that a team ends up building the exact same product. Firebase Studio in my mind made sense, but maybe AI Studio is a better home for that team.

• 
  ⚛️Louis Waweru☮️ (@zelifxeh) reported 24 minutes ago

  Damn, the latest update to humanity is incompatible with the Jews. Alright, we can patch a quick fix, right? Put it on GitHub.

• 
  Daniil (@hey_daniil) reported 28 minutes ago

  I built DevIntern because I was my own bottleneck. My agents sat idle while I context-switched, and checking in on them every twenty minutes shredded my focus. DevIntern takes me out of that loop: 1. It connects to whatever tracker you already use (Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, even plain markdown files) and pulls work straight from your tickets. 2. Before implementing, it checks the ticket is actually feasible. Vague specs get flagged back to the tracker with questions instead of becoming a confidently wrong pull request. And when the ticket doesn't exist yet, devintern/pm turns a Figma design, an error log, or a rough requirement into a well-structured story. 3. It runs your own coding agent, model, and API keys inside your repo. The subscriptions you already pay for keep working while you sleep instead of only when you're watching. There's no lock-in and no token markup. 4. The output is a pull request. You review it and merge. The agents grind through the backlog. Your time goes to the work that actually needs your brain.

• 
  Anupam (@Anupam_Devops) reported 38 minutes ago

  The AI Suggestion That Almost Broke Production Tool: Claude Code / GitHub Copilot A senior engineer was moving quickly. An AI assistant suggested an infrastructure configuration change. The explanation sounded convincing. The code looked clean. The review almost took less than five minutes. Then someone noticed a problem. The configuration would have disabled a critical safeguard protecting production workloads. The AI wasn't malicious. It was confident. And confidence can be dangerous. That's when the team adjusted its approach. AI became a collaborator. Not an approver. Today they use AI extensively for: • Documentation • Investigation • Refactoring • Learning unfamiliar systems But production decisions still require human judgment. The best engineers aren't competing with AI. They're learning how to supervise it. Production Tip: Treat AI-generated infrastructure like junior-engineer code. Review everything. Assume nothing. Question: What's the most useful DevOps task you've used AI for?

• 
  Vatsalpandya333 (@Vatsalpandya333) reported 39 minutes ago

  Everyone is racing to build AI agents. Very few are asking what happens after they ship. Here's what we've learned talking to engineering teams: The first failure is rarely the expensive one. The second is. Why? Because the company already had all the information needed to prevent it. The logs existed. The GitHub PR existed. The Slack thread existed. The customer ticket existed. But none of them talked to each other. Every incident becomes tribal knowledge. Someone remembers it. Until they leave. Companies don't have an AI problem. They have an institutional memory problem. Every production incident should make the next incident easier to solve. Not start from zero. That's the infrastructure we're obsessed with building.