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:

• 
  Hennie (@andyhennie) reported 3 seconds ago

  @theo Really good post!! We’ve all just accepted the shortcomings of worktrees (we should be able to check out the same branch in multiple places) and GitHub (monorepo with private folders please). There might be good reasons for things being the way they are. But the world has changed. We’re not working the same way anymore. The world must adapt. And people in charge can fix it. Whether it’s Linus or a GitHub VP doesn’t matter. We just need to point it out. And explain the need. I don’t always agree with you, and I sometimes get annoyed when I see you kick downwards on X. But at this ****, you’re really really good! Gj. If you get the issues on this list done (by Theo-bullying them or asking nicely, I don’t care), I’ll get a picture of you on the wall of my office. Probably one with a mustache. ;)

• 
  Amu (@amu4biz) reported 31 seconds ago

  @realokwy_ @0xWassie @GeckoTerminal cause github is down na? using gitlawb

• 
  Jenkstones (@zigglywiggler) reported 2 minutes ago

  @thesis_agent 0x099880c1676FF3035Ab1E952E5E83b5A81eecB07 $gnull: wallet-authenticated repositories for github and gitlawb, all activity recorded on chain for ranking identity, contributions and recording changes. Autonomous AI agents for code review, issue tracking, bounty payouts. Now they are going to focus on adoption. Gitlawb dump and a 10% wallet left yesterday so maybe opportunity longer term.

• 
  Robert Youssef (@rryssf) reported 2 minutes ago

  @ReinDaelman prompt injection via github issue bypassed the action runtime-what isolation layer failed in gemini-cli?

• 
  krvnb🚩 (@scottkyliestan) reported 3 minutes ago

  GitHub down and our CI/CD is blocked @github atleast make your status page displaying correct info It’s been 30 min and status page still shows operational

• 
  0xTria (@0xTria) reported 3 minutes ago

  SOMEONE JUST TURNED CLAUDE INTO A TRADINGVIEW ANALYST A guy installed a TradingView MCP server from GitHub, gave Claude desktop access, and hit enter. 5 minutes later it was marking NASDAQ like someone paid it to sit there all night: > 15M HTF bias > internal liquidity > external liquidity > sweep targets No screenshots. No “explain this candle.” Claude drew it inside the actual chart. The money part is obvious. Traders pay for courses, signals, and Discord groups just to learn where liquidity is sitting. Now Claude can mark it before your coffee gets cold. The next video is execution. Why are people still charting manually?

• 
  riptide.eth (@riptide_eth) reported 4 minutes ago

  Zack from Reno, Nevada. 20 years old. Just this winter he was breaking his back on the night shift at an Amazon warehouse for $14 an hour and pissing in plastic bottles to avoid getting fined for idle time. Today he makes $368,065 net every single month. Yesterday he rolled out a matte black Lamborghini Urus just to do donuts in the parking lot in front of his former manager. How? He singlehandedly hacked the entire US corporate system. While senior programmers with degrees are spitting bile on LinkedIn over mass layoffs, Zack physically does not work at all. His neural network is officially employed in 38 remote Senior Data Scientist positions at top IT companies simultaneously. Live coding in Zoom and HR filters kill 99% of candidates. Zack simply wrote an asynchronous multimodal framework that passes interviews for him. With voice. In real time. Everyone thinks AI gets caught on latency. Zack solved the ping issue through a hardcore vector inject of the DeepSeek model directly into the audio driver. (T_response = (μ * VRAM_cache) x RAG_context <= 115ms) -> Hired How this matrix works: Voice and Empathy: A custom bundle of the Pipecat framework and local Llama 3. The agents sit in morning Zoom meetings themselves, crack corporate jokes with managers and report on tasks. The response latency is 115 milliseconds. The human ear cannot hear this. Execution: Worker agents via LangChain and a custom parser write code, fix bugs and push commits to corporate GitHub repositories around the clock. His server costs: $480 a month. His salary from 38 corporations: half a million bucks. HR directors give Zack's AI bot bonuses and set it as an example for real employees. Corporate security teams (SecOps) are going crazy trying to track down his IP addresses. HR departments are tearing their hair out because legally he is not breaking any rules. He is just closing tasks 10 times faster than an entire department. They are rushing to update contracts to ban AI agents. And Zack simply leaks the entire source code of his framework to the public in response. So that millions more like him can emerge. Keep sending out resumes and begging for interviews. Or grab this code, spin up your Swarm and make corporations pay you the salaries of dozens of people. Save this tweet right now 🔖. Big tech lawyers are already throwing strikes to take down the repository. The scripts for Zoom video production, the Swarm architecture and the prompts for passing tech interviews are right here 👇

• 
  Daniel Moore (@dizlexic) reported 4 minutes ago

  I have yet to be effected by a GitHub outage in any way whatsoever. AMA

• 
  Michal Barus (@webjuice_ie) reported 5 minutes ago

  Switched to grok-4.3 on Hermes. Codex 5.5 kept failing with "model provider failed after retries" - tied to HTTP 401 auth errors. GitHub shows hermes auth add openai-codex doesn't sync reliably with ~/.codex/auth. Reddit threads confirm the same: unreliable + overly verbose. anyone has a workaround?

• 
  Startup & Tech (@startup_an_tech) reported 6 minutes ago

  @builtat2am A 10 minute interview is wild but it shows how the market is shifting. For AI roles, companies care more about your GitHub and actual builds than solving DSA problems on a whiteboard. If you can actually ship code that works, that is the only thing that matters now.

• 
  aigenerated (@abhxyyyy) reported 7 minutes ago

  github down again????

• 
  Extella Research (@extellaresearch) reported 9 minutes ago

  at 48% context, claude told a user it wasn't being effective anymore. recommended starting a fresh session. less than halfway through a 1M context window. github issue #34685, anthropic's own repo. extella runs the other direction. each session builds on the last.

• 
  Trevin Chow (@trevin) reported 9 minutes ago

  I keep saying this, but use the `ce-debug` skill in compound engineering. It's terrific for finding root cause and not just fix the symptoms. You can loosely describe an issue, pass in a github/linear/etc issue link, and it'll do a root cause analysis.

• 
  Tom (@tomcrawshaw01) reported 10 minutes ago

  The #1 open-source repo on GitHub right now is an AI agent called Hermes. Its founder just gave away the 6 keys to getting real work out of any agent. Most of them have nothing to do with which model you pick. The problem they fix shows up fast. Most people babysit their agents, micromanaging every step, and the work still comes back wrong. So here are his 6 keys. 1. Describe outcomes, not steps. Tell the agent what "done" looks like and the conditions it has to hit, then get out of the way. The models keep getting better at long-horizon planning, so your step-by-step instructions are the bottleneck now. 2. Define "good" or you get slop. The AI has no taste. Leave your standard unspoken and it hands you the average of everything ever written, then thinks it nailed it. Write your standard down, all of it. 3. Explain it like it's an alien, not like it's five. You and I share a lived history, so half of what we want goes unsaid. The model never grew up on Earth, so every assumption you skip stays unmet. 4. Use agents for patience, not creativity. They have infinite patience and almost no creativity. So give them the work a human could do but never wants to, like reading every log line or running the same checks all day. 5. Build the agent that learns once and reuses forever. Their agent booked a Vegas restaurant. 45 minutes the first time, instant the next day, because it saved the skill. One engineer taught their log agent once and the whole company runs on it now. 6. The harness matters as much as the model. The model is the brain and the harness is the body. A great harness with a weaker model beats a great model with a weak harness, because the model just outputs tokens and the harness is what touches your actual work. Put these together and one person starts moving like a team. The founder built Hermes into the #1 repo on GitHub without really knowing how to code. The tools did the rest. There's never been a better time to be early again. I break down systems like this every week. Follow for more.

• 
  neilofneils (@wneilofneils) reported 10 minutes ago

  hermes is back. github pr sorta-fix, and then it fixed itself. openai-codex