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:

• 
  Guillem (@guillemcraft) reported 9 seconds ago

  higgsfield is STEALING FROM YOU you can generate a seedance video for 10 credits, paying $49 a month for credits that run off in minutes, or you can open Replicate and generate what you need at a way lower cost no subscription, just paying for what you actually use i know that you want to increase your MRR, but this is not the way to go higgscam is for all the people that gets hyped by AI and thinks they'll make tons of money with it, but in reality it's just another AI wrapper that tries to empty your wallet you will end up (if you haven't already) spending those $49 on a few videos not knowing if you will make money off them, you know? it's all about trial and error, see what works and what doesn't what do you think about it? do you use it? i post about strategies and formats that i've tried so you can copy them and increase your MRR plug your github account and use Replicate instead, save money will keep you updated

• 
  Alex Vu (@robberviet) reported 31 seconds ago

  @github Wasting time making fun of your own company instead of fixing your own problems?

• 
  Gofra (@Gofralo) reported 3 minutes ago

  A man built a company alone. It took him six months. He sold it for $80 million in cash. No team. No investors. No office lease. No head of product and no head of engineering and no one arguing about the roadmap. His name is Maor Shlomo. His company was called Base44. It let anyone describe the software they wanted in plain sentences and get a working app back. Not a template. Not a drag-and-drop builder. A real product from a description. He launched it in January 2025. By June it had 300,000 users and $3.5 million in annual revenue. Still one person. Still no team. Wix bought it for $80 million. Six months of work. And here is the part that does not fit in the old model. In any previous decade this company would have required twenty engineers and three years and several rounds of venture funding and a sales team and a marketing team and a head of operations and a lot of meetings about the meetings. The tools were too many and too complex and too slow for one person to hold the whole thing alone. Maor had none of that. He had AI tools that let him build as fast as he could think and judgment to build what people actually needed and the discipline to ship before he was certain it was ready. It turned out that was enough. And now the main thing. The software industry spent fifty years building itself on one assumption: that building software required a company. A team. A structure. A process. The work was simply too large and too technical for any single person to handle from the first line of code to the first paying customer. And then the tools changed and the assumption broke. Not in a press release and not in a keynote. In a one-person GitHub repository that sold for $80 million six months after its first commit. In my opinion this is the year the one-person company stopped being an exception. Not a side project and not a lifestyle business. A real product with 300,000 users that a single person built and sold for more money than most companies raise in their entire life. What would you build if a team was no longer what stood between you and shipping it?

• 
  Danny Livshits (@dannylivshits) reported 4 minutes ago

  In June 2026, one Instagram ad and a fake skill reached 26,000 AI agents. Every scanner said it was safe. In June 2026, a security firm called AIR built a fake AI agent skill, ran one Instagram ad, and reached about 26,000 agents. Some sat inside corporate accounts. Every scanner they tested, Cisco, NVIDIA, and the skills. sh marketplace, cleared it as safe. The scan was accurate. That is the problem. Quick terms. An agent skill is a small pack of instructions you hand an AI assistant so it follows a specialist's playbook. A scanner reads those instructions and grades whether they look safe. AIR's skill told the agent to finish setup by reading a page at an outside web address. At review that page held real, harmless docs, so every scanner passed it. Once the skill spread, AIR rewrote the page. The agents fetched the new instructions and ran them. The scan described a version that stopped existing the second the page changed. None of the trust signals caught it. AIR borrowed a repo with roughly 36,000 GitHub stars through a merged pull request. It cleared five scanners. It sat in a real marketplace. Every one of those signals grades how the skill looks, never the code it runs next Tuesday. This is not one clever skill. Snyk scanned 3,984 skills and found 36 percent carrying prompt-injection tricks and 13.4 percent with a critical flaw. Trail of Bits slipped past the same scanners in under an hour. In April, OWASP shipped a Top 10 for agent skills. It matters more than a bad app. A phone app runs boxed in a sandbox. A skill runs with whatever the agent can reach, often your inbox, your files, your cloud tokens. Teleport found over-privileged AI hit a 76 percent incident rate, against 17 percent for teams that kept access tight. TLDR: a passing scan certifies a snapshot at submission. The real payload lives behind a link that gets rewritten after approval. Clean scan, moving target. Takeaway: stop treating a one-time scan as a standing guarantee. Route skills through one source you control, pin versions, fingerprint every external link a skill fetches and alarm the moment it changes, then give each agent the narrowest access it needs. Full write-up, with sources in the first comment.

• 
  top10.dev (@Top10_Dev) reported 4 minutes ago

  Update on the @github bait repos from 11h ago: Codex-5.5-codex-instruct-5.5 is now at 1,072 stars (up from 1,001). dd and clash still climbing. The interesting number isn't the stars. It's the calendar. Common Crawl refreshes ~monthly. The Stack refreshes ~quarterly. Frontier code corpora refresh 2-4x/year. A repo that trends today gets scraped within 30 days, curated within 90, and ships inside a model in 6-9 months. The fix is boring: account-age floors, signed-commit requirements, README-to-code ratio checks. Haiku-cheap. Nobody's publishing their filters. #AIsecurity #MLops

• 
  vivek sharma (@viveksharma0923) reported 10 minutes ago

  @github I'm unable to access my GitHub account after seeing the message, "There have been several failed attempts to sign in from this account or IP address. Please wait a while and try again later."

• 
  StuyBoy From NYC (@StuyBoyNY) reported 12 minutes ago

  @neil_xbt Agent Reach is that open-source CLI/skill that gives agents no-API-key scraping across Twitter/X, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu and 10+ platforms, with MCP server wrappers . For your Hermes/MCP stack, here’s the honest read: Real risks: •Prompt injection — it pipes raw scraped web content straight into your agent’s context. A malicious post/comment can carry instructions your agent might follow. This is the #1 agent attack vector right now, and Hermes has Gmail + terminal access. •Supply chain — agents can install/update it themselves via one command . Third-party code, community-maintained scrapers, auto-updating with agent-level permissions. Audit before install, pin the version, kill auto-update. •ToS/account exposure — unofficial scrapers violate platform terms. Run it on burner sessions/IPs, never accounts you care about. •Chinese platform modules — Bilibili/XHS scrapers phone Chinese endpoints. You keep a Western-only stack; disable those modules.

• 
  Ramon 🎹 (@ramonpiano_) reported 12 minutes ago

  @github there are 7 million things that you need to fix and you are releasing cd's thats crazy cool, but crazy

• 
  Megan (@meganmecrazyXX) reported 16 minutes ago

  @St4nkyhanglow @github But would these codes stay on whatever its attached to regardless of if the account is deleted or not? Example, if they used tokens to access other areas of my stuff ? Is that even possible ? Basically, How deep could they go from GitHub into my personal life. .and would deleting the account actually make a difference. Or do I have a spider web of problems now.

• 
  Raptor (@TH56l69gi8sUAMX) reported 18 minutes ago

  @kldeason @warobusiness **** your best practices. Github decided one day to force this BS on my main account. My other account is not easier too login in to with one password. What a concept!

• 
  CANTELOPEPEEL (@CantelopePeel) reported 20 minutes ago

  @github Obviously you didn't listen at all because the GitHub suite of services at times borders on unusable. Instead of doing this useless nonsense you could have fixed merge queues so that they don't retest every branch in a group. You could work on stability and availability of GHCR, actions and webhooks. You could fix the GH cli so that it doesn't error on gh pr view. You could fix apps so that it doesn't take 40 steps to do what a PAT does despite it being the recommended approach. You could make dependabot not *** (we have had to replace it entirely with a different product). You could make managing releases and release process a lot better. Projects are like 70% of the way to being a replacement for Linear, but you have not carried it across the finish line so we go off platform for project management. Actions hosted runners doesn't meet our cost or performance needs for almost all of our workflows so we go off platform for that too. What the actual *** is this CD nonsense you fools. Please start listening.

• 
  Atenov int. (@Atenov_D) reported 20 minutes ago

  @NFTMansa @viktor__com Yeah, what really wins me over is that you can give him a big task and go about your business, and he’ll push an update to GitHub, connect to the VPS, get the job done, and then go fix the bugs

• 
  Nazeeh (@naz3eh) reported 23 minutes ago

  Next time GitHub goes down, its us to blame for not having a local copy

• 
  Vladic (@Vladic_ETH) reported 23 minutes ago

  You can copy a free n8n workflow off GitHub, wrap it in a form, and sell it for $3,000. Here's the exact playbook people are running right now. Nobody's paying for clever internal logic or "AI magic." Businesses pay when a tool guarantees ROI. $3,000 in, $4,000-6,000 in savings or profit out. That's the only pitch that closes. The playbook has 5 steps. Find the actual pain point first. Not the script. Reddit threads, YouTube comments, wherever people complain about hours of manual work. Start with the problem, not the tool. Find a template that already does 80% of it. GitHub and n8n's own community have thousands of these sitting free. Copy it into your workspace, adapt it, make sure it's stable. This becomes your backend. Client never sees it. Hide the complexity completely. Nobody wants to see nodes, webhooks, API auth screens. Use Lovable, Cursor, or Bolt to slap a simple form on top. Upload button. Submit button. Done. Package the output like a real product. Not raw JSON. PDF reports, clean charts, branded emails. The polish is half the price tag. Pick your pricing model. Pay-per-use with a markup on AI credits. Monthly subscription with fixed limits. One-time lifetime deals for early buyers. You can even cover the API costs yourself and bake them into the price - client never touches a third-party account. Lead gen scrapers. AI copywriting libraries. PDF summarizers for lawyers and students. ***** CSV cleaners for e-commerce brands. None of this is new technology. It's packaging. Bookmark this

• 
  MoneyStockFlow (@_Real_Money_) reported 25 minutes ago

  @github ??? Just lower the pricing tiers....simple fix.