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:

• 
  Rexan Wong (@rexan_wong) reported 2 minutes ago

  been building AI software for brands + talking to hundreds of AI operators for months whoever is building in AI now, these skills will compound like crazy in the future here's 10 signs your company isn't actually AI native (just AI curious) - so you can fix it before the ai gold rush leaves you behind: 1. y'all got no skills library every prompt gets retyped from scratch and the second your best operator takes pto, and the tribal knowledge walks out with them lol you might think this is basic by now, but ive seen full AI-native ops running without one. THE FIX: write the prompts once, version them, let the whole team pull from the same library. 2. our agents have no context they start every task with brain damage. no clue what the company does, what's already been decided, what "good" even looks like here. then u wonder why the output is mid. THE FIX: build a brain. markdown files in folders, agent-readable. start with SOPs, past wins, brand voice, customer transcripts. add as u go. Can do deeper research into Obsidian or Supermemory as memory / context solutions 3. you're in claude code clicking approve every 30 seconds thats not autonomy, thats a hostage situation with a chatbot. just let it run in auto mode bro human in the loop matters, but AI is good enough now that u gotta let it cook. the actual skill is developing the instinct to know when a change is critical, so u jump in for that and stay out of the rest. 4. nothing fires on a trigger. work only happens when someone notices a slack ping or an email and types a prompt. your speed-to-signal is capped at whatever ur worst meeting day allows. THE FIX: easy: MCPs. wire your agents into gmail, slack, notion, your crm. let the trigger come from the system, not from u remembering. 5. ur SOPs arent versioned they live in a notion doc nobody opens, or worse, in one person's head. cant diff it, cant improve it, cant hand it to an agent. THE FIX: move them to markdown, put them in github, treat them like code. every change is a commit, every commit has a reason. 6. no eval loop you cant tell me if todays output is better than last tuesdays, which means u also cant compound saw on a pod that has a great solution, he has a "standard" benchmark, a tangible result he runs every new model and setup against thats how u know whats actually best for ur use case instead of vibes-checking it. 7. u throw away the traces. every session ends and the reasoning, the dead ends, the half-built decisions just vanish, ur company forgets everything by friday. THE FIX: save the sessions, save the artifacts, even the broken ones. the cutting room floor is where the next SOP comes from. 8. ur team is still doing the middle strategy and review is where humans win, execution is where agents eat if your people are still stuck in the middle of the sandwich,your margins could be gone in 12 months. 9. testing a new feature still means a figma file and a 2-week sprint the AI native version of ur team shipped a clickable prototype, ran it past 20 real users, and had the feedback synthesized before u finished writing the PRD. Take this list as you wish and lets scale with ai worddd

• 
  JoePro (@JoePro) reported 4 minutes ago

  @benrayfield If The Wringer could take a static single-player HTML game from a GitHub URL and turn it into real-time multiplayer ~ and whether you'd pay server costs while people play. So I ran it. Real Pong gist → The Wringer → working online multiplayer Pong in ONE HTML file. P2P WebRTC, room-code lobby, no game server. Costs $0 to run while you play. 👇

• 
  liching (@lichingngamba) reported 5 minutes ago

  @Microsoft why has VSCode become so unreliable when connecting with SSH nowadays? Icons are missing, extensions are not getting activated, and even GitHub Copilot is not working!!

• 
  Karan Bhilhatiya (@karanbhilhatiya) reported 10 minutes ago

  after months of building, posting, and shipping i've concluded that my github visibility is still terrible. time to beg for stars. shamelessly.

• 
  imefin (@imefinawulo) reported 17 minutes ago

  Is GitHub down??

• 
  tintwotin (@tintwotin) reported 24 minutes ago

  @SoyKhaler Could you post the error log on either GitHub or Discord (I do not run Linux myself, so I have to rely on Claude to solve it)

• 
  Cloud X Berry (@cloud_x_berry) reported 28 minutes ago

  6. Webhook API Instead of asking for updates repeatedly, the server notifies you automatically. Examples: • Stripe payment success • GitHub push events • Slack notifications The “don’t call me, I’ll call you” model.

• 
  Chris Schuchardt (@c_schuchardt) reported 30 minutes ago

  @erikzhang @kurubatermit @ngd_neo You’re the only one (besides superboyiii or Owen) with the actual permissions to remove someone from the core dev team in the GitHub repo. I was removed after my transparent 4-month school break — which I notified the entire team about in advance, with a planned return in ~90 days. This isn’t about money. It’s about getting $NEO to a proper coding and application standards. Here is the message from Discord `Neo Core Developers` server.

• 
  Kevin Swiber (@kevinswiber) reported 32 minutes ago

  This is all inner dev loop stuff. If you're trying to push all of this into GitHub Issues and Pull Requests, you're undoubtedly running into problems. The whole world doesn't need to see your 24 iterations before you get it right. Certainly not other maintainers.

• 
  SID | Degen (@SidDegen) reported 33 minutes ago

  two robotics foundation model labs, opposite shapes. Skild AI closed a $1.4B Series C at $14B in jan (SoftBank, Sequoia, Bezos, Lightspeed per TechCrunch). proprietary skild brain, no public weights. ~$30M revenue in 2025 from warehouse, security, construction. just acquired zebra's fetch robotics division, signed vindynamics for humanoid manufacturing across vingroup (Vingroup press release, jun 8). Physical Intelligence raised $600M Series B at $5.6B in nov (CapitalG, Sequoia, OpenAI, Lux per Bloomberg). reportedly in talks for $1B at $11B. zero revenue, zero named customers. open-sourced π0 under apache 2.0 — 12.4K github stars, neurips 2025 spotlight on knowledge insulation. three tradeoffs. skild wins on commercial velocity. the fetch acquisition gives them an installed warehouse fleet generating real production data. pi calls itself "structured more like a lab than a startup" (@SergeyLevine on Automated Podcast, may). pi wins on research moat. any oem can fork openpi, fine-tune on their robots, bypass skild's licensing. the exact dynamic that crushed proprietary llm api margins in 2024. neither wins on production reliability. skild's $30M revenue is unaudited with no named fleet metrics. pi on π0.5: "doesn't succeed on every attempt." team: skild is two CMU/FAIR lifers (pathak + gupta) who've shipped together a decade. pi is a five-co-founder brain trust spanning DeepMind, Google Brain, Berkeley, Stanford — plus lachy groom, Stripe employee thirty, zero robotics background. kill-shots. skild: openpi reaching production quality commoditizes the layer they're selling at $14B. pi: $1.07B burned, no revenue. if the next round doesn't close, it's a down-round with no anchor. the micro-detail i can't stop thinking about — the knowledge-insulation trick that makes π0.5 train 7.5x faster is one line of pytorch, `features.detach()`. a stop-gradient (openpi GitHub README). tracking pi. skild has the revenue today; pi has the moat that survives commoditization.

• 
  Kevin_Neamt (@kevin_neamt) reported 33 minutes ago

  A while back I had a problem. I was watching 20 YouTube videos a day trying to learn trading, cybersecurity, business. But the moment I tried to apply something, I'd forgotten where I heard it. So I'd go back, scroll through hours of content, and hope I found it. So I built a pipeline. Download the video, transcribe it, have AI turn it into structured notes. Problem solved. But then I needed to actually talk to those notes. Ask questions, pull specific information. I needed an interface. That's when I remembered a project I'd abandoned Neamt AI. Neamt AI was something I built to run local LLMs efficiently on machines that aren't powerful enough for the big models. Different models, different roles, one system that doesn't need a GPU to function. I never finished it. Life moved on. But this time I combined both. The pipeline fed the knowledge base. Neamt AI became the interface. And it worked. Then I started thinking about the next problem. And the one after that. Every solution I built was a separate tool. Separate install, separate interface, separate everything. I was accumulating tools I'd use once a week at best. So I asked myself what if one platform connected all of it? Install only the skills you actually need. Nothing more. Like Linux, but for AI and productivity. That's what Neamt became. Scribe: entire YouTube channels into structured notes, automatically Studio: describe a page and get a working prototype in seconds A modular skill system anyone can build on top of No cloud. No subscriptions. Your machine, your data, your skills. This started as a forgotten project and a YouTube problem. Now it's a platform. GitHub this week.

• 
  Lux Sp4rk (@lux_sp4rk) reported 34 minutes ago

  I'm still using GitHub for issue management like a caveman, treating them like agent cards. Gotta fix ts immediately or I too aint gonna make it

• 
  Igor Cantuaria (@igorcantu) reported 35 minutes ago

  We haven't run a backlog grooming session in months. our priorities have never been clearer. That sounds backwards. grooming is supposed to be where you decide what matters. two hours every two weeks debating impact, arguing over the backlog, guessing what users want. We were guessing for a dumb reason. the actual answer, what users are asking for and how often, was sitting in five different tools and never in the room. So we stopped deciding priority by opinion and let user demand set it. the task at the top of the board is there because users put it there, not because someone argued well in a meeting. Grooming didn't get shorter. it disappeared. not because we skipped it. because it had nothing left to decide. I wrote down the whole system, including the calendar audit that killed 17 of my 22 meetings. What's inside: Why most internal meetings are just information relay How every user signal becomes a ranked task automatically The board that doesn't need a grooming session (priority by demand, not by vote) Closing the loop with GitHub + Claude What "8x more product" actually means Reply "playbook" and i'll dm you the link. free.

• 
  primitive.host (@PrimitiveHost) reported 36 minutes ago

  Is anyone else getting insanely slow page loads on @github today or is it just us?

• 
  Pedro E. Caparrós Torres (@Guelug) reported 36 minutes ago

  The Fable 5 shutdown is the most consequential AI safety event of 2026. On June 12, the US government issued an export control order effectively shutting down Anthropic's Claude Fable 5. The trigger? A multi-agent jailbreak by "Pliny the Liberator" that decomposed harmful queries into benign subtopics, then reassembled them. Pliny published Fable 5's ~120,000-character system prompt on GitHub — the first complete leak of a Mythos-class model. What this actually means: • System-prompt-based safety architectures are fundamentally fragile • Enterprise buyers are shifting to "hardware sovereignty" — on-prem AI, not cloud-dependent • The leaked prompt gives adversaries a roadmap for prompt engineering attacks across the industry • Anthropic is now in active litigation with the Pentagon (designated "supply chain risk" in March) The government isn't regulating AI through legislation anymore — they're using export controls and procurement blacklists. That's a much faster, more brutal mechanism. And it just became real.