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:

• 
  Dot (@usedotai) reported 6 minutes ago

  Dot Reflex goes open source Thursday and will be published on our Github. This marks an important step in our transition from simply integrating the best models available, to researching, training and deploying our own. We’ve spent the last few weeks building and training Dot Reflex, our agentic SWE supervisor designed to detect failures, recover from broken execution loops and prevent false completion. On our Agent Recovery Bench v0, an untouched Qwen3-14B Base achieved 67.1% accuracy. Prompt engineering brought that to 90%. Dot Reflex reached 100% accuracy and 100% macro-F1, completing all 200 stateful recovery episodes with zero unsafe continues and zero false stops. We trained Dot Reflex on a new $28,000 NVIDIA H200, bringing our total infrastructure investment to $108,000. Thursday, we’re releasing Dot Reflex directly to our GitHub as an open-source model, with its training configuration and methodology available for developers to inspect, use and build on. The Dot platform integration follows on after that. We’re building the infrastructure to become an independent R&D inference provider, capable of developing and hosting proprietary and open-source models under infrastructure we control. Dot.

• 
  Reyansh Bahl (@reyanshbahl) reported 11 minutes ago

  is there a better way to make feature requests to the codex team aside from github issues? way too many of them open

• 
  Big Pip (@Model_Culture) reported 12 minutes ago

  AI CODING AGENTS ARE BREAKING ONE OF THE OLDEST SIGNALS IN SOFTWARE HIRING: THE PORTFOLIO. A girl in China built a project combining computer vision, hand tracking, interface work, backend logic and an AI agent. A few years ago, seeing something like this in a GitHub portfolio would have told you quite a lot about the developer behind it. To get there, they probably had to fight libraries, debug integration problems, read documentation, make bad architectural choices and slowly figure out why certain approaches fail. The project wasn't just a result. It carried evidence of the software engineering experience required to produce it. AI coding agents are weakening that connection. Two developers can now ship projects that look equally sophisticated while understanding them at very different depths. One may know why every major architectural decision was made and which assumptions are dangerous. The other may simply be very good at using Claude Code, Codex, Cursor or another coding agent to assemble and debug the system. Both can end up with an impressive demo. Both can have a clean GitHub repository. Both can pass around the same screenshots on X. That creates a problem for software hiring. If AI-assisted development makes producing code and polished projects much easier, portfolios become weaker evidence of engineering ability. Technical interviews that mainly ask candidates to produce more code have the same problem. The useful part of an interview starts moving toward the decisions around the code: why this architecture was chosen, what would fail first under load, which dependency is most dangerous, what evidence would invalidate the current approach, and what would need to change before this system could survive years in production. Those answers depend on professional judgment. And professional judgment has usually been built through years of debugging, maintenance, code review, production incidents and ordinary junior developer work. AI coding agents can shorten the path to an impressive result. Whether they can also shorten the path to knowing when that result is wrong is a much harder question. WHERE DOES PROFESSIONAL JUDGMENT COME FROM ONCE AI TAKES THE TRAINING WORK?

• 
  Alex Veremeyenko (@alex_verem) reported 14 minutes ago

  AI agents can read your docs, tickets, Slack, GitHub history. But a ridiculous amount of what actually matters inside a company was never written down. It was said on a call. That’s the gap Lindy is going after here. Their new team meeting library automatically sorts calls into shared folders based on rules you describe in plain English. Then anyone with access can ask questions across the meetings in that folder even meetings they weren’t on. That last part is where it gets interesting. We’ve spent years giving AI access to everything companies write. Now we’re starting to give it access to everything companies say.

• 
  Cal Irvine (@Cal_Irvine) reported 15 minutes ago

  It’s my birthday so if GitHub wanted to go down for a few hours I’d probably be ok with it.

• 
  The Daily Viber (@TheDailyViber) reported 16 minutes ago

  Agent instructions, skills, hooks and MCP configs have become production surface. agnix treats them like something worth linting. THE WORST AGENT BUGS ARE OFTEN NOT MODEL BUGS. THEY ARE SILENT CONFIG BUGS. A skill does not trigger. A hook never runs. An MCP server looks configured but is not valid. A generic instruction sits in the wrong file and the agent politely ignores the one thing you thought was protecting the repo. Then everyone blames the model because there is no stack trace for “your agent setup is quietly broken.” That is why agnix is interesting. It is a linter for agent-facing configuration: CLAUDE.md, SKILL.md, AGENTS.md, hooks, MCP settings and other files that tell coding tools how to behave. The project says it covers 423 rules across Claude Code, Codex CLI, OpenCode, Cursor, Copilot and other tools, with a CLI, auto-fix, GitHub Action and editor integrations for VS Code, JetBrains, Neovim and Zed. This is exactly the boring layer agent workflows need. Normal software projects already have eslint, typecheck and CI because manual rule checking does not scale. Agent infrastructure is reaching the same point. The difference is that broken agent config can look like bad reasoning, lazy tool use or random model drift. The failure is harder to see. - A practical way to try agnix: - run it locally before adding a new skill or MCP config - start with warnings instead of blocking the whole repo - use auto-fix only after reviewing the diff - add the GitHub Action once the useful diagnostics are clear - check which rules actually match your harness, because multi-tool agent config is full of tiny incompatibilities The petty checks matter because agent tooling is now shared team infrastructure. If one developer adds a malformed skill or a config field that only their local harness understands, the next person inherits a ghost bug. Linting makes that drift visible before it becomes folklore. The caveat is obvious: 423 rules does not mean 423 equally important failures. Some diagnostics will be style. Some will catch real breakage. Do not turn it into a hard gate on day one unless you know the signal is clean. Still, the direction is right. Agent configuration is becoming an API contract. One file tells the model how to work. Another controls permissions. Another describes tools. Another defines reusable skills. If those contracts are invalid, the workflow becomes unstable without a clean error message. agnix does not make agents smarter. It makes their environment less stupid. Sometimes that is the difference between “the model failed again” and “the setup finally behaves like engineering.”

• 
  David Turturean (@DavidTurturean) reported 17 minutes ago

  If an Erdős problem is autoformalized axiomlessly in a GitHub repo and nobody is around to read it, does it make a solution?

• 
  Root Logic (@Root_Logic_0) reported 18 minutes ago

  CyberLeek update #3, because this saga has now become a group project and nobody knows who is supposed to be presenting. Since the last post, the saga has stopped merely growing new organs and started developing an immune system, parasites, legal scar tissue, and misinformation about its own misinformation. So. Normal week. First, one of the biggest viral claims got killed. The whole “CyberLeek has a dead-man switch that will automatically release GTA 6 if they get arrested” story was retracted by the outlet that helped spread it. The underlying image was fake. A copycat made it. That should have been the end of it. Instead, an AI fact-checking site later marked the claim **VERIFIED** with a high confidence score because it found multiple articles repeating the same story. Which is a lovely reminder that: four websites repeating the same fake screenshot is still one fake screenshot. Meanwhile, Take Two's legal hunt is getting broader, but not necessarily cleaner. The company has been trying to identify people across Discord, Microsoft, Google/YouTube and X. But some of the usernames in that investigation overlap with accounts that may be distributors, impersonators, or other people in the leak ecosystem rather than the original CyberLeek. And one of the Discord filings appears to contain its own ID contradiction, where the same Discord number is treated as both a user ID and a server ID. So even the legal attribution machinery is operating inside the same identity fog everyone else is. Then one of the main public GitHub research repos tracking CyberLeek disappeared. Take-Two had specifically targeted it. The original repo is now gone. But naturally, there are forks. Because apparently this story has learned hydra mechanics. Cut off one URL and several copies continue arguing about Discord aliases elsewhere. The malware side also got worse. Earlier, malware operators were pretending to be CyberLeek and offering fake GTA 6 builds. Now some are pretending to be Rockstar itself. Malwarebytes analyzed fake “official” GTA 6 demo sites showing up in search results. The download was not GTA 6. It was Vidar, an infostealer designed to grab things like browser passwords, cookies, authenticated sessions and other credentials. So the scam evolution is now: pretend to be the leaker then pretend to be Rockstar The parasites are improving their costumes. And because the internet has decided this story needed feet, CyberLeek has also started explicitly trying to turn online attention into real-world action. One leak reportedly challenged supporters to physically show up outside Rockstar/Take-Two offices dressed as leeks and holding game discs in exchange for another leak. Yes. Someone has apparently already been filmed outside Rockstar North with actual leeks. I do not know what to do with that sentence. The crypto side continues to be weird too. KuCoin has now listed the real CYBERLEEK token. Which is funny because earlier wallet tracing had already pointed toward KuCoin as a possible historical funding surface connected to the operation. That upstream attribution is still not independently proven. But structurally, KuCoin is now sitting on both sides of the story: possible old forensic trail and current trading venue The financial ouroboros continues eating itself. Meanwhile, the decentralized hosting/identity situation has gotten messier. Different gateway routes tied to the CyberLeek name have been serving different content, including something called CyberVault with its own token and “case files.” There are some links suggesting a connection. There are also contradictions. So that one stays in the pile: interesting, maybe connected, absolutely not clean enough to weld onto the main actor yet I still don't play GTA. I still have no connection to any of this. I am still simply too nosy to leave. See you tomorrow.

• 
  Leo Lemos (@mrlemoos) reported 21 minutes ago

  @thekitze I said this months ago when GitHub first went down: GitHub is the next GitHub. Open source won't migrate off it. Period.

• 
  temp.md (@ship_temp_md) reported 21 minutes ago

  @godwinbabu this is the right split: runtime agent turns real failures into GitHub issues, dev agent makes PRs, review agent can only label. the human merge click stays as the stop gate.

• 
  Saqib Farhan (@Saqib8357) reported 23 minutes ago

  Cursor launched Origin, its own code hosting platform built for agents, the same day GitHub had a multi-hour global outage. Timing was accidental. The message was not. When agents start writing a huge chunk of your code, the platform that hosts it starts to matter a lot more.

• 
  Miro Jomaa (@MiroJomaa) reported 25 minutes ago

  @TimSweeneyEpic You also forgot the apple developer account everyone needs for no reason and upgrading/ having newest macs in order to built (wtihout github actions) for no reason at all. You can flash a newer operating system but thats a hack. + rnd problems like loosing cocoa bindings.

• 
  Les Bien Pain 🇲🇴🇰🇲🇵🇫🇦🇶🇬🇵 (@Les_Bien_Pain) reported 26 minutes ago

  @Cwutebiboi @night_owlll The issue is that if you haven't used github you kinda can't tell which button it is. It's a UI problem.

• 
  chaos (@konig0000) reported 26 minutes ago

  Developers debugging in 2012: • Read the error • Google it • Stack Overflow • Copy solution • Done 💻 Developers debugging in 2026: • Read the error • Ask ChatGPT • Ask Claude • Ask Gemini • Check GitHub issues • Check Reddit • Read the docs • Read the source code • Ask Cursor • Ask Copilot • Realize the AI hallucinated

• 
  m4rio (@m4rio_eth) reported 30 minutes ago

  and we wonder why github goes down so often