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:

• 
  Patrick Gerard (@grumpypcodes) reported 9 minutes ago

  Github or Vercel has issues atm?

• 
  Pavan Kumar (@pavank38) reported 9 minutes ago

  If normies are using AI to write “technical code” what is the problem with using GitHub? AI can handle end to end version management with GitHub CLI. beyond shilling the “technical vs non-technical” distinction, what’s the actual use case?

• 
  Gareth Owen (@TheAnswers42Com) reported 10 minutes ago

  So here's just where I go past the 280 character limit and do a House of the Dead PC infodump as I love this game, and the code behind it's a pretty interesting snapshot of games development in the late 90s, pretty much the wild west of 3D graphics before things like renderers became a solved problem. Nowadays you can just download Unreal and get a world-class renderer and asset pipeline at your fingertips for free, but back then everything was pretty much handcrafted, and you can see that here. The renderer almost instantly gives away its 90s origin - most of the assembly of draw commands and meshes, etc. is done on the CPU, and then rasterization is carried out either by a software renderer onto a DirectDraw surface, or dispatched to a Direct3D 6 device for rasterization. It's from a world before programmable pipelines, and where software rendering was super common. I've been using TypeScript so far as it's just easier than the compile/crash cycle of C++, and the Canvas API of HTML 5 is a pretty good match for DirectDraw in many ways. Assets are stored in a format called RBH - all of them. The RBH file seems to be a form of RIFF/PIFF file, and the payload itself is either an uncompressed BODY, or can also be a PACK file that's been put together using a custom LZ compression algorithm that Tantalus themselves came up with. Game logic, sounds, the camera and cutscene opcodes for the events VM, pretty much everything is in there as an RBH. I knew it was some kind of compression, but chatting with Qwen and Cursor really helped pin down some stuff about it much quicker than I could have a few years ago. I came up with a set of TypeScript decompressors that run using Bun to get it all decoded and turned into formats that can be viewed correctly outside of the game engine, and I've also translated them into Python as well to stand independently of the asset pipeline tests I've been doing. Pretty much all House of the Dead code has been lost to time, but I'm hoping that eventually I can shed some light on it and figure out what made that game tick. It's a pretty special one to me, and I always found it sad that all of the artifacts from its development seem to have been lost. Still working a lot of stuff out, but I've been keeping notes about what I find as I go, and I figure I'll write it up in a wiki on Github as it's pretty interesting stuff. In its current state it's all pretty messy and there's a lot I'm not sure about right now, but in time I think I can crack it. Maybe there is some source code knocking about there on floppies or CDs in a drawer or attic, but even so I doubt if it's survived disc rot. This is probably the next best thing!

• 
  Kunal (@kunmath00) reported 13 minutes ago

  @pierceboggan I've had problems in making this work. The sessions are not starting and giving errors. Raised an issue on github

• 
  KISA aka Copenzafan.eth (@copenzafan) reported 14 minutes ago

  AI agents on a VPS are a terrible idea. And now I’m going to argue against an engineer from Google. His argument is that agents shouldn’t run locally. That includes Claude and Codex, meaning you shouldn’t have a harness on your own machine. I completely disagree, specifically within the bounds of “can I vibe code this?” For example, deploying a website takes a fair amount of resources before the site is even ready. Instead of building it on my PC and shipping it to a VPS, I’m supposedly meant to do the whole thing directly on the VPS. Which means I’m expected to pay several times more than the cheapest plan just to have enough RAM, so my agent doesn’t die along with a frozen server. The same problem comes up if I need to build a graphical app for managing lots of agents. So the agent writes code, has to push it to GitHub, and then I pull updates from there and test the app. Or I have to download the code or build every time. And where am I even supposed to compile that build without GitHub? Sure, you can host an agent on a VPS that counts your calories. But hosting an agent on a VPS to vibe code and build your projects? What kind of inconvenient nonsense is that? Is that why Google keeps shipping Gemini this ugly and dumb?

• 
  Iso Ledger (@JamesDula82) reported 15 minutes ago

  Should You Own A Tangem Wallet? Let's audit. @Tangem Founded 2017, Zug, Switzerland. Founders include CEO Andrey Kurennykh, Andrew Pantyukhin, and Anselm Schmucki. $23M raised across two rounds, led by SBI Crypto Investments ($15M) and Shima Capital ($8M). 250+ employees. 6 million+ wallets shipped across 230 countries. The chip. Samsung S3D232A/S3D350A secure element, Common Criteria EAL5+/EAL6+ certified, same class as passports. Private keys are generated on-chip via certified TRNG and never leave the card. Firmware is factory-flashed and permanently non-updatable, by design. The good. Three independent audits, Kudelski Security (2018), Riscure (2023), Cure53 (2026, mobile SDK), none found critical vulnerabilities or backdoors. App is open source on GitHub. No seed phrase to lose or phish by default. Genuinely simple UX, tap a card to your phone, no cables, no batteries. The bad. No screen. Every transaction is verified through your phone, a device Tangem doesn't control. If malware swaps a destination address, the card signs whatever the phone tells it to. Mobile-only, no desktop or air-gapped signing option. Lose all your backup cards and recovery is permanently impossible, no seed phrase fallback exists. The ugly, three disclosed incidents in under two years: December 2024. A bug logged private keys in the app during seed-phrase wallet creation, one path among Tangem's setup options, not the default seedless mode most users choose. If a user contacted Tangem support within 7 days of activation, that logged key was accessible to support staff via email and ticketing. Surfaced publicly via Reddit after Tangem's initial response drew criticism for being slow. Affected under 0.1% of users, per Tangem, and only those who'd opted into the seed-phrase setup. No funds were lost, but it proved the app can touch key material when a seed phrase is used, despite the "keys never leave the card" pitch. September 2025. Ledger's Donjon security team disclosed a "tearing" attack, cutting power to a card mid-authentication to dodge its built-in delay after failed password attempts, combined with reading electromagnetic emissions to detect a correct guess. Normally takes 148 years to brute-force an 8-digit code; the tearing technique cuts that dramatically. Unpatchable, cards have no firmware update path. Tangem disputed it posed real-world risk and declined to pay a bug bounty. July 2026. Donjon disclosed a laser fault-injection attack: a single nanosecond laser pulse aimed at the secure element's die bypasses a firmware check and resets the card's password outright, no old password or backup card needed. Affects every Tangem card in circulation. Cannot be patched, ever, same reason as above. Requires physical possession, ~$250,000 in lab equipment, and genuine hardware security expertise; it's destructive and leaves visible damage, so it can't be done covertly. Disclosed responsibly in February, published five months later. Tangem called the real-world risk "virtually non-existent" and noted Donjon is a direct competitor's research arm, both true, neither erases the underlying finding. Bottom line: the chip certification and independent app audits are real. But immutable firmware, marketed as a security feature, means every one of these three disclosed flaws is permanent across every card ever shipped. There's no patch, ever, for any of them. That's a fundamentally different trade-off than an updatable device with an NDA'd chip layer, not necessarily worse, but you're accepting a different kind of risk: nothing can be tampered with remotely, but nothing can be fixed either. We audit the plumbing. You decide what to do with it. 🛡

• 
  Jason 🇺🇸 (@privatename906) reported 17 minutes ago

  @FabMaximil @ProtonMail I think this was their first outage since last October. Go pick on GitHub.

• 
  Dhruv Patel (@dppatel_) reported 17 minutes ago

  An AI agent has been fixing bugs at @aftersell for 6 months. It costs us $400/mo. Meet Watson👇 We audited all of our engineering tickets last year and 35% weren't code bugs. They were config questions, duplicates of bugs we already knew about, confusing UI, etc. And then the real bugs? The fixes are always easy but investigation was always the inefficient part. Watson now finds and squashes bugs and has started to implement features fully autonomously. Here's how it works: 1) Triage - is this worth looking into? Watson classifies every ticket before it digs in - is it a config issue, known issue, does it need more info? Output is structured data so downstream stages branch off of a real contract instead of parsing what a model is saying. Helps a lot with latency, cost, and performance. 2) Investigate - what happened? This is the core of why we built Watson - it'll reconstruct the entire event timeline, read the codebase, pull cloud logging, etc. And we built this all directly on top of the Claude Platform Client SDK so we can have full control over prompt caching, per-tool call caps, cost accounting, etc. It's a super powerful way to control cost - on a "warm run" - 98% of inputs are from the cache. 3) Implement - write to the codebase Watson directly writes to the repo, makes a PR, etc. It also monitors and reviews PR comments from reviewers. This is the simplest part - writing the code is so simple when you've gotten all of the context from triaging and investigation. 4) Memory - self improving loop (MUST HAVE) Watson has a persistent memory mounted via GCS Fuse. Every time Watson is invoked it brings back findings and artifacts. We're layering it into episodic (past investigations), semantic (distilled patterns and playbooks), and reflective (what investigation strategies actually work on this codebase). Watson integrates into everything we have - all internal chats, support systems, Linear, Github, etc. It's constantly improving and every single Linear ticket we have gets looked at by Watson. Our first run was ~$50 and now it costs us $3.87 per ticket all in. Comment below if you want me to write up a guide for how to build your own Watson.

• 
  ₿themoon | ∞ vs 21m | Capital ₿ear Stackers member (@Bthemoon2030) reported 18 minutes ago

  @github some of my website are giving 404 There isn't a GitHub Pages site here. Is there any issues on the servers?

• 
  James Liberty (@JamesLiber54019) reported 18 minutes ago

  The fix was simple: I had to generate a digital VIP pass (a private security token) so GitHub recognizes my app every time it checks in. Works smoothly now without getting locked out. 🦾 (2/2)

• 
  Millsy.dev (@Millsydev) reported 19 minutes ago

  @TimSweeneyEpic @fortniteonlinux Well your Linux support has been terrible and very lacking If you had the API I can access to create my own for everyone to use and put it on GitHub I would. I would make it RUST /REACT Tauri2

• 
  The DevOps Guy (@justdevopsguy) reported 22 minutes ago

  honest question: why are people paying @linear when they could just use @github issues and projects that already map nicely to their code and PRs?

• 
  André (@andr3barroso) reported 25 minutes ago

  Everyone is saying ChatGPT Plus isn’t worth it because you burn through the Codex limits too fast. I think they’re looking at Plus the wrong way. Most of what I do happens in Chat mode, not Codex or Work. From my iPhone, in a normal Chat, I can: • Connect to GitHub, PostHog and other tools • Research anything on the web • Review code and PRs • Work through problems in my projects • Take a photo and instantly translate or understand it • Talk naturally using probably the best dictation I’ve used And this doesn’t consume my Codex/Work allowance. Then, when I actually need an agent to work on a repo, I switch to Codex or Work and use that allowance. I can even connect my iPhone directly to my Linux VPS and keep working without opening my MacBook. For me, ChatGPT Plus isn’t just a Codex subscription. It’s an AI assistant

• 
  oscar (@oscargaske) reported 26 minutes ago

  @itsnotryan @github did it catch the outage today?

• 
  Jose Pollman (@josepollman82) reported 28 minutes ago

  A malicious GitHub issue hijacking a coding agent: GitHub's risk page says only users with write access can trigger it, and comments from anyone else never reach the agent. The documented injection defence is filtering hidden characters. Plain prose is not hidden.