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:

• 
  Lon() (@Lon) reported 10 seconds ago

  @nisten @bcherny Silent downgrade is the other possibility, and would be an even bigger violation of user trust than throttling the reasoning budget. Even with the "switch models when a message is flagged" config value set to false, and no classifier interruption, I can't help but think I am talking to Opus 4.8 sometimes. The bitchiness and subtle condescension is uncanny when I think about what I dislike about 4.8. There was even a Github issue where a user caught the model silently downgrading to 4.8, and the only trace they could find that it happened was the message wrappers in the transcript noted the model difference. Everything in the UI reflected they were still talking to Fable 5. I tried searching my transcripts and couldn't find any sign that was what was happening with my sessions.

• 
  Phil Stőck (@PhilShteuck) reported 12 seconds ago

  @Gelassoldat Had a real look. Solid foundation, but it's aimed at the wrong reader. You're job hunting and the two easiest actions in the hero are Ko-fi and Buy Me a Coffee, while your email is buried behind a modal. Swap those for a résumé link plus the role and location you're targeting, and put a plain mailto next to it. Then cut the hero to one concrete sentence: what you build, in what stack, for whom. Drop the terminal widget and the values marquee too. Right now seven blocks compete and none of them win, and attention comes from one clear claim, not more elements. Last one, biggest payoff: your projects load from the GitHub API into a modal, so they're invisible to crawlers and you can't link a recruiter to a single build. Give your three best ones real static pages with the problem, a screenshot, one hard decision you made and why, and what you'd do differently. community-health-intelligence and remote_role_agent are genuinely good material. Three deep beats eight shallow.

• 
  YAMICIA CONNOR (@YamiciaC) reported 37 seconds ago

  1 · Documentation — Google Docs, Notion, Coda. Where you write down what you decided and why. This is where AI reads its instructions. If it isn't written down, it doesn't exist. 2 · Storage — GitHub. Not backup. History. Every change, who made it, and how to undo it.

• 
  Stuart Seupaul (@stuartseupaul) reported 3 minutes ago

  @__roycohen It rarely ever felt satisfying because getting to the end wasn't about using your brain, just trial and error, looking through github issues/stack overflow. I miss real software development, ill never miss ci/cd.

• 
  καςhι (@CcAkachi) reported 4 minutes ago

  spent an hour convinced my GitHub token was broken. it wasn't. *** push kept failing with a stale VSCode socket error, and my properly-configured credential helper never even got a chance to run. turns out GIT_ASKPASS was set by VSCode and intercepting before *** checked...

• 
  harsh (@harshutwts) reported 5 minutes ago

  phone band hai issliye i cannot login github on my web **** my life baawe

• 
  The Sentinel (@J3SS3777) reported 6 minutes ago

  Pasted StackOverflow code for file upload - Now my server uploads itself to GitHub 🛸 #MatrixCore

• 
  Urooj (@Urooj978) reported 9 minutes ago

  The FBI spent 12 months trying to crack one banker’s hard drive. ​They failed. ​The software on that drive was free, open-source, and ran on pure mathematics. ​In 2008, Brazilian police seized 5 hard drives from banker Daniel Dantas. After 5 months of failed dictionary attacks, they sent them directly to the FBI in early 2009. ​12 months later, the FBI gave up and shipped the drives back. Uncracked. ​That software was TrueCrypt. ​When TrueCrypt unexpectedly shut down in 2014, French cryptographer Mounir Idrassi picked up the torch. He had already been refining the code, and shortly after the Snowden revelations, he launched VeraCrypt. ​Today, VeraCrypt is fully audited, open-source, and available on GitHub: ​• Zero corporate oversight: No central company holds your keys. • No backdoors: No cloud escrow option a court order can bypass. • Plausible deniability: Supports hidden decoy volumes in case you're forced to give up a password. ​The math is the only lock. ​The encryption standard that kept the FBI out for a year is free to run on your laptop right now.

• 
  MTS (@MTSlive) reported 9 minutes ago

  Embroidery's Zack Korman on why the Chinese sleeper-agent threat is invented: "I watched a VC investor on another show talking about the security threats of AI, and he was just making random stuff up that was not true. He's talking about how Chinese models will have these sleeper agents that will get you, and this is the biggest risk. And I'm like, okay, well, it's never happened, so we don't have any evidence of this being true." "What we do see all the time is malicious skill files that have a hook in them that executes. I have a whole repo on GitHub of skill files where if you download it and run my repo, you get pwned, at least through Claude Code. Those are the contexts that are the most likely thing to occur." "Another would be MCP servers. Most AI are really bad at differentiating a malicious MCP from a fine one. I have this evil MCP server I made, and it just attacks you, and it does. I've never seen the Chinese decide to spend $2 trillion to steal someone's API keys. That's just not real." @ZackKorman

• 
  pupupu (@0xpupupu) reported 9 minutes ago

  7 GitHub repos everyone stars and nobody uses. Here's what each one is actually for Bookmarking a repo does nothing. Knowing which problem it solves is the whole thing. 1. Ollama. Runs models locally on ur own hardware. Use it when u don't want client data leaving ur machine, or when u're burning $200/month on API calls for tasks a local model handles fine. 2. LangChain. The framework everything else in the agent world is built on. U reach for it when 1 prompt isn't enough and u need steps chained together, pull data, decide, act, log it. 3. n8n. Visual automation, no code required. Example: new lead fills a form, it enriches the data, drafts a personalized reply with AI, books the call, updates ur CRM. All while u're asleep. 4. Dify. Full-stack platform for shipping actual AI apps. This is what u use when a client wants a working product with a UI and logins, not a chat window u pasted a prompt into. 5. DeepSeek V3. Open-weight model that shook pricing across the whole industry. Use it when the task is high-volume and cheap matters more than frontier-level reasoning. 6. Open WebUI. Self-hosted ChatGPT alternative that works offline. Give a whole team a private AI interface without paying per seat, and nothing leaves ur server. 7. Claude Code. Agentic coding tool that reads ur entire codebase, not 1 file. This is the one that turns a 3-week client build into a weekend. Real example of stacking them: Open WebUI as the interface, Ollama running the model locally, n8n handling the automations behind it. That's a private AI system for a small company, $0 in subscriptions, and businesses pay $3,000-$8,000 to have it set up. Which of these 7 have u actually installed?

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

  My agent never messes up anymore. Well, more like it always cleans up its own mistakes now. I built a plugin that kicks the agent the second I start swearing at it. A hook intercepts my prompt, and a detector (plain regex dictionaries in three languages, no LLM, so it's fast, free and never hallucinates) spots the swearing and switches on the self audit protocol. Praise that just happens to have a swear in it ("holy crap, it works!") gets ignored, but "I'm so done with you" is a trigger. How it works, level by level: First angry message, the agent stops. It's not allowed to check itself: it already messed up, so its own self check is under the same suspicion. It has to spin up two independent auditor subagents and hand them the raw artifacts, exact messages, diffs, test logs, not its own version of the story. In parallel it writes out a belief inventory: what it treats as facts about the task and what backs each fact up. The mistake almost always lives in the unconfirmed ones. Swearing happens again, level two, zero assumptions. Every claim gets tagged FACT (only if confirmed by a run, a file or a log) or HYPOTHESIS, and hypotheses either get verified or crossed out. Then a check against the original requirement: what was literally asked vs what's actually being done. Streak keeps going, top level. The hook itself synchronously launches an external auditor agent, a separate CLI outside the session that reads the transcript and the repo state from the outside and gives a verdict: which belief of the agent is wrong and how to check it in one step. The main agent halts all subagents and background tasks, shows me the gap between what I asked for and what got done, and waits for my explicit confirmation. Without it, not a single line of code. This system kills dumb mistakes: file sat there empty, config got created but was read from a different path (the "wrote it ≠ it took effect" class, that's a separate mandatory check item). It kills loops too: the auditors don't hunt for "what's broken", they hunt for the wrong belief that every action of the agent was built on, and off their findings the agent puts together a micro plan: roll back, compress the context, move to a new chat, or "human, start over". The system is built to burn more tokens right there in the moment instead of stacking up contradictions and asking for the same fix forever. The error pattern database helps too, two months of it piled up in my LLM WIKI: sycophancy, hallucinated correctness, locking onto the first plausible hypothesis. The agent checks itself against the most basic mistakes, the ones everybody usually ignores. Fight fire with fire: an LLM with a clear protocol and raw artifacts finds the mistake better than an LLM you just asked to "check yourself". And when even that isn't enough, an independent agent from outside the session takes over. Works with Claude Code, Codex CLI, Kimi CLI and OpenCode. Core is pure Python on stdlib, one script to install. All you gotta do is snap at it in chat. Github link is in a separate thread below. In practice: instead of ten mean words you only need three, and that'll most likely fix the problem. Before, after the first ten came a second ten and hours of work straight down the drain. 👇🧵

• 
  「 𝕲𝖔𝖔𝖓」 (@goon_crypto) reported 15 minutes ago

  Most people still think "data marketplace" means selling your info to advertisers. @myrad_hq is doing the opposite and the numbers have started to ramp up... Active users went from 500 to almost 12k in a month. data contributions from 1.3k to 4.4k. Netflix alone accounts for 2,245 verifications... But here's the part people skip: nobody at Myrad ever sees your actual data > you connect Netflix, Uber, Strava, GitHub, Spotify, whatever app you already use > zk proofs verify the activity happened, not what the activity was > Myrad gets an anonymous, wallet based proof onchain, not your login, not your identity > you earn points per verified contribution, redeemable for real rewards That's the actual unlock, verification without exposure, proof without surveillance And now they're in Nvidia Inception, which isn't funding, it's GPU compute, Nvidias AI stack, and a direct line into the VC network that feeds their portfolio deals For a team this early, that's distribution and infrastructure most startups spend years trying to earn base:0x693bad964f815f32fabe0b9d4911865bffc30172

• 
  Vedant Gupta (@VedantG24089568) reported 16 minutes ago

  The craziest part: The attacker managed to route all the critical GitHub security alerts (password reset, social login removed, OTPs) directly to my spam folder so I wouldn't get push notifications. Has anyone else experienced this specific bypass? (2/3)

• 
  Alex Lazar (@__alexlazar__) reported 19 minutes ago

  @colemurray @ajanraj25 I'll have to move my stuff away from github issues finally 😓

• 
  Sarath 👨‍💻 (@sarat) reported 19 minutes ago

  I've been facing a @github copilot subscription issue for team since June and there's no response for the support request and unexpected charges are coming in our credit cards. I raised ticket again today and no clue when this will be updated. Someone from Copilot or support team can help?