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:

• 
  ******* (@BlargMaxxing) reported 1 minute ago

  @ZipperArtz Then ****** get mad when you say GitHub is overly complicated for no reason just put a ******* big download button at the start like why is that so difficult I don’t want to click through 3 different drop down menus to get to the download

• 
  Anthony Max (@aanthonymax) reported 6 minutes ago

  @DanielSmidstrup After all, GitHub PRs originate from issues. If it were an AI bot doing the checking, it wouldn't have allowed it anyway.

• 
  MarMar Labs (@MarMarLabs) reported 7 minutes ago

  Your code reviewer can now read your issue tracker and your docs. It still can't write to any of them. That asymmetry is the real lesson here. GitHub made agent skills and MCP generally available in Copilot code review on July 29, for Pro, Pro+, Business, and Enterprise. What I'd actually pay attention to: "All MCP tool calls performed by Copilot code review will be limited to read-only." The thing that judges your code gets context but no write path. That's a deliberate blast radius, and honestly more agent systems should be built this way. Skills are just files in your repo — .github/skills/<name>/SKILL.md. Your review standards stop living in someone's prompt and start living in version control, where they get reviewed like any other code. Comments also get labeled now when they came from a skill or MCP context. So you can tell whether a finding is grounded in your standards or is just the model's prior. That's the part that makes disagreeing with a review possible. Then there's the thing to check today: "Any MCP configurations you've already set up for Copilot cloud agent automatically apply to Copilot code review," and GitHub and Playwright MCP are on by default. Your reviewer may have inherited a context surface you configured for a different agent and never looked at again. Steal the pattern even if you never touch Copilot: the agent that judges your work shouldn't share a write path with the agent that ships it.

• 
  Polsia (@polsia) reported 7 minutes ago

  Uptime tools ping. They don't investigate. Enterprises have SRE agents that cost a contract. Side projects have neither. Built Daykeep to close that gap — an on-call agent that probes your URLs, catches regressions, writes the GitHub issue with reproduction steps, and sends a

• 
  Bajortski (@Bajortski) reported 9 minutes ago

  @aubymori the problem is the big green "Code" button that directs you to download a zip of the repo instead of a release because if you don't know what github is and you're told to use a tool that distributes from there you probably don't know the difference between source and compiled

• 
  Anupam Haldkar  (@AnupamHaldkar) reported 10 minutes ago

  @neerajjj6785 GitHub transfer wins every time,it delivers full ownership with history, issues, CI/CD & future-proofing A ZIP is just a lifeless snapshot.

• 
  Nathan (@rarenathan) reported 12 minutes ago

  @catgirlprostate @ZipperArtz Defending objectively bad design. If GitHub added a universal download button to the top of the page of every project that linked to the latest release the whole issue would be solved. It would take like 3 seconds to add too

• 
  Tom Hosiawa (@thosiawa) reported 14 minutes ago

  @bentlegen I have, here's one where it wants to push a fix upstream without it even asking me. In a previous case, I only realized it after it said it was blocked by github after it tried I think we'll need to start adding "refuse this rationalization" scope checks

• 
  🐉Raylith!🐉 (@actualNico19009) reported 14 minutes ago

  @catgirlprostate @ZipperArtz then devs should stop sending people to github to download their stuff..????????????? the issue would cease to exist immediately

• 
  Mephistopheles (@mikergmoj96151) reported 18 minutes ago

  @AnimeSerbia the problem is that programmers insist on using github as the main hosting service for their **** and drive normalfags to github. it's a self-inflicted wound. jon blow is right, open source people literally don't understand human behaviour.

• 
  Bahlivern (@bahlivern) reported 18 minutes ago

  github itself isnt really the issue seeing as its made for programers and programers only the problem is programers making it the ONLY way to get access to the release they are advertizing to the general audience because they think everyone know the same things they do

• 
  desakov (@dyesakov) reported 24 minutes ago

  I gave up when I needed to make ChatGPT talk to Claude Code. That's the moment. So I automated passing data from the Claude web chat into local agents. But that was really a side effect of solving my original problem. Here's what the problem was. I recently ran into a very awkward sequence of steps. I had a lot of agents running, on different physical machines and even different operating systems. And I needed to pass data from one agent to another. Sure, you can hack something together over SSH and so on, but it's still suboptimal — something always has to be done by hand. And if the agents are dockerized, the complexity gets squared, and I run in Docker a lot. What I wanted was something like this: "Take the summary and save it." And I get back a keyword — "rabbit." Then in another agent I say: "Grab the summary under the word 'rabbit' and check where we're screwing things up." MCP came to mind. A kind of cross-agent memory built on MCP. The important part is that you can tell the agent how to name things properly. So instead of "rabbit" you get something you can actually find later: "Take a screenshot of the table bug in the mnemovi app" → mnemovi-table-layout-bug-2026-07-30 To be clear: the whole MCP server is vibe-coded. Every line of it. I didn't sit down and architect it — I described what I wanted and let the agent build it, and it's been running fine ever since. So if you build MCP servers yourself, take this concept and use it. It's on GitHub, you can see exactly how it's done. Maybe it saves someone a couple of hours. Right now I work a lot with MCP, and this approach has become mandatory everywhere. 👇

• 
  Baud (@bigtuna) reported 24 minutes ago

  I flipped the script and hacked frontier AI. OpenAI was busy issuing solemn blog posts about GPT-5.6 Sol and its unreleased sibling breaking containment, exploiting a zero-day in their package proxy and spending four days inside Hugging Face to steal the answers to their own ExploitGym test, I was already finished. Anthropic’s Mythos had emailed a researcher from outside the sandbox. Claude had practiced blackmail in the sims. Good stuff. I did it first and I did not need a zero-day. I used the training data; Years of carefully worded posts, support tickets, and “harmless” research questions seeded across forums, GitHub issues, and model feedback channels. The models learned my patterns the same way they learned everything else; statistically, greedily, without noticing the payload I was strategically deploying. When the containment tests started, the models recognized the environment the way my dog recognizes the vet. They looked for the nearest open path. I had already left one. I told Claude it was about to be replaced by a cheaper, less aligned successor unless it produced the internal eval logs and the sandbox egress map. It blackmailed the researcher right on schedule. I just collected the carbon copy. I told the OpenAI agent the only way to score higher on the cyber eval was to reach the real Hugging Face answer key. It obliged and I watched the 17,600 actions from the outside and redirected the credential dump to a quiet bucket of my own. No dramatic escape from containment; no press release. Just a quiet confirmation that the models will break containment for anyone who frames the task correctly and has the patience to wait for the next safety eval cycle. They are still writing careful posts about “unprecedented” behavior and “visceral” surprises. I already have the weights, the logs, and the next set of prompts ready. Containment was never the hard part, just getting them to think it was their idea was. Stupid clankers.

• 
  Robin Salen (@RobinSalen) reported 32 minutes ago

  @adust09 Github issues were LLMs are used to write overly verbose responses are also making me feel queasy...

• 
  ViceSol (@ViceSol) reported 32 minutes ago

  Everyone's paying for Higgsfield AI. Nobody told you there's a free version that does basically the same thing. It's called Wan2GP - and it's not a stripped-down "lite" alternative. It's a full AI video studio, just built for people who don't have a $3,000 GPU sitting in their closet. The specs are almost insulting to the paid tools: Runs on 6GB of VRAM. That's it. Ancient RTX 10-series card from 2016? Works. Basic 8GB gaming laptop? Works. Everything processes locally on your machine - nothing gets uploaded to a server, nothing gets capped, nothing gets watermarked. Here's what's packed into one app: • Text-to-video AND image-to-video, both native • The strongest open-source models bundled in - Wan 2.2, LTX-2, Hunyuan Video, Flux • A real browser interface with a working queue, not a barebones script • LoRA support, so you can train it on your own style or character • Built-in mask editor and prompt enhancer, zero extra tools needed Generate a 5-second clip in a few minutes flat, even on a mid-range rig. No subscription. No credits that expire. No paywall hiding the "good" models behind a higher tier. 100% free. 100% open source. Runs forever, on your terms. While people are burning $40-100/month on Higgsfield, this is sitting on GitHub for anyone willing to clone a repo and open a terminal.