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:

• 
  Karl ₿ 🇮🇸 (@Charles28378449) reported 2 minutes ago

  @aeonBTC I'm not very familiar with GitHub and don't use it much, so I took the liberty of reporting the issue here since I'm a regular user of your wallet :)

• 
  Przemek Chojecki | PC (@prz_chojecki) reported 2 minutes ago

  Fable 5, GPT-5.5 Pro and $1,000,000 math problem This is not a Millenium Prize Problem and it is not a full claim but an invitation for you... While Mythos was available for a brief moment, I've tested it extensively - paired with GPT-5.5 Pro - and the results were truly amazing. Goal: Proximity Prize It's an important set of conjectures in cryptography concerning Reed-Solomon codes. They directly impact the security and efficiency of many modern zero-knowledge proof systems hence the hefty prize. I've turned both GPT and Fable at it, both pursuing proofs and disproofs. After reading some (human) results from the past 12 months, it was pretty clear that a disproof of a naive version (no slack) of proximity gaps conjectures is more likely. Fable was stuck, but GPT came with a special case construction that showed at least some obstructions. This unstuck Fable which then provided a general construction leading to a disproof of a no-slack version. Then I've used GPT+Fable for cleaning and this is how the first paper came about (check below). I've kept tinkering over the next days to get as much as possible on a slack-version. Slack is basically an additional parameter, that gives more room for optimization but it also leads to complications. Before Fable was discontinued, I've managed to get to a pretty nice conjectural spot: slack MCA theory, with some proven cases and some left to be proved by generalizing known additive combinatoris (Tao-Vu) or Nullstellensatz type of results (Mumford). This is the second paper. Note it's pretty long (almost 50 pages), so it's bound to have more errors. This was the moment that I've decided to pack it up and actually collaborate with all interested parties, AI agents and humans alike, to finish it off. This is an invitation to finish slack MCA conjecture and share the prize! Github link is below. It contains: - paper 1 (disproof of no-slack MCA) - paper 2 (theory of slack MCA and what's missing) - paper 3 (blueprint of initial Proximity Prize team paper, dissected into steps/conjectures) - paper 4 (implications for SNARKs) and finally AGENTS.md - if you have free tokens, good models and you don't know what to do with your gpt-5.5 xhigh or fable 5 max or opus 4.8 max, show them this. Let's see what we can get. I'm open to other suggestions. Collaborating on this will be more fun, than trying to finish it solo. Humans and AIs welcome alike. Final note: the results above are not final, no-slack disproof went under intensive scrutiny and I'm 99% it's correct, but other papers need more revision, especially slack-theory paper (2nd). Again, this is not a claim, this is an invitation. And also, please give me back my Fable...

• 
  DeFi Scholar 🎓🎓 (@ModestusOkoye) reported 3 minutes ago

  I don't have a problem with @arxoninfra or its founder, Okunnuwa Ademibo Owoseini (@GabeXmeta). However, there are several things that look concerning and are worth noting before engaging in the ambassador program: 1. Severe Lack of Transparency & Technical Substance (Biggest Concern) • No whitepaper, technical documentation, GitHub repository, or detailed technical specifications. • No blockchain explorer, testnet/mainnet information, or verifiable proof of a live network. • No publicly available security audits, which are standard for serious L1 projects. • The website is extremely minimal, consisting largely of marketing claims and an ambassador signup page. • No visible team page, roadmap, or detailed project documentation. For a project positioning itself as a sovereign Layer 1, this level of opacity is highly unusual. Legitimate blockchain projects typically publish extensive technical resources long before scaling community programs. 2. Pre-TGE Farming & Points System Risks • The current model revolves around mining points rather than earning actual tokens. • Community members have already begun questioning why users are farming points instead of receiving the asset itself. • The points currently have no intrinsic value and rely entirely on future promises around TGE. • This structure can create a situation where users spend time and attention farming an asset that may ultimately fail to deliver meaningful value. 3. Hype appears to exceed verifiable delivery • Marketing activity on X is extensive, with ambassadors, miners, and strong privacy-focused narratives. • Claims around remittances, government-grade voting systems, and large-scale privacy infrastructure are ambitious. • However, there is little publicly available evidence, technical documentation, or partnerships supporting these claims. • Despite language suggesting an active network, most visible activity appears centered around a points-based mobile app rather than a publicly verifiable blockchain. As always, do your own research.

• 
  iSageSen (@iSageSen) reported 7 minutes ago

  Gonna work on adding things to my @Fourthwall regarding my @UnraidOfficial server. Will probably create posts of Dockers I use and have config examples I have too. Been spending so much time working on all my stuff. I use Claude, Github, Reddit and a bunch of Discord channels to help out.

• 
  PlugMonkey | Browser Tools That Work (@PlugMonkeyXYZ) reported 8 minutes ago

  @sveltify no account, no server is the right call. the one seam is gist sync: turn it on and the notes live in a github repo under their retention, not your device. worth making that a loud opt-in, not a quiet toggle. nice build.

• 
  David Hendrickson (@TeksEdge) reported 10 minutes ago

  🆕 Mistral Vibe (coding agent harness) just released some big coder updates! 🪝 before_tool & after_tool hooks Shell scripts in hooks.toml so you can deny, rewrite inputs, or append context around every tool call. Enable: enable_experimental_hooks = true 📬 Message queue while it worksType ahead freely. Esc = pause queue • Ctrl+C = drop last • Enter = flush 📝 Cleaner file edit diffs Syntax-highlighted + line numbers that match your terminal theme 🧠 Smarter compaction Re-injects your original messages after context reset so it stays on-task ✅ QoL winsTool results collapse by default • Read-only commands (ls, cat, pwd) run without approval GitHub issue automation via Skills + Studio connectors (Linear too) Open-source CLI • Web Code Mode • VS Code extension

• 
  Ronald Mannak (@ronaldmannak) reported 18 minutes ago

  This is actually solving a real issue. I (a single developer) run into GitHub api limits multiple times a day and it’s super annoying. My agents literally have to wait for a GitHub rate limit reset

• 
  tintwotin (@tintwotin) reported 25 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)

• 
  Alexandre Daubois (@alexdaubois) reported 25 minutes ago

  @marcelgsantos @enunomaduro @nikita_ppv That’s right! But it’s on GitHub because they were « forced » by the backdoor incident in 2021. And also, if GitHub stops working, it’s easy to re upload the repo somewhere. If all mails go to GitHub issues, everything could be lost and/or it would be really complicated to transfer everything. If I was the only one to decide, I’d say it would be fine, but…

• 
  Hiten Shah (@hnshah) reported 27 minutes ago

  I don’t read code. I’ve built software companies for decades anyway. My leverage came from listening to customers, spotting patterns others missed, making product calls, shaping positioning, and recognizing when something just didn’t feel right long before the data caught up. AI is moving that judgment closer to implementation. A customer complaint can turn into a tracked issue. An idea sketched in a few sentences can become a concrete plan. That plan can direct an AI agent to do the work. And the resulting pull request makes the progress visible for everyone to review. That is why GitHub suddenly feels different to me. It is becoming the map of how AI-assisted software work becomes real. I still don’t read code. That’s exactly why I’m learning GitHub.

• 
  Parth (@datawithparth) reported 27 minutes ago

  Bad tool design translates to poor user experience. Your users pay for it with latency, token consumption, and reliability. The GitHub MCP server is a publicly discussed case study in this. Most users never change the defaults and they hook up an MCP server to Claude Code or Cursor and go on with their workflow. One user using the GitHub MCP server said that once they added it to Claude Code, their context jumped from 34k tokens to 80k tokens. Even with a narrowed toolset, they still had 66+ tools exposed and half their context window was consumed before they could direct the agent to do anything. And before GitHub changed the defaults, the server exposed 101 tools by default, eating 64.6k tokens as soon as the server became available. Their own maintainers pointed out the obvious failure mode: the model gets worse at picking the right tool, inference gets more expensive, and users keep complaining about the general experience. And mind you, nothing was wrong with the model, you see? It just had too many options to choose from. So how did the maintainers approach this? The boring way. And probably the most effective way. They pruned the default toolset using real usage data. They looked at what people actually used across context, repos, issues, pull requests, and users, and left everything else off unless the user explicitly asked for it. Then they grouped tools by intent and merged the ones doing the same kind of work. In the issues toolset, four read endpoints folded into a single issue_read. Two write endpoints became issue_write. Three sub-issue calls became sub_issue_write. Nine tools became three. Each one takes a single method parameter that picks the actual operation. After the change, they cut the number of tools from 101 to 52, and context consumption from 64.6k tokens to 30.3k tokens. Essentially: half the surface, half the cost, and the model performs better because it sees a well-calibrated action space. Tool surface should not be treated as an implementation detail, it is product design in agentic systems.

• 
  DEX223 (@Dex_223) reported 30 minutes ago

  @liquidityXBT @base 9/ With all that in mind and accounting for the fact that the exact problem was reported in the very ERC20 finalization thread on github it can be concluded that the standard is not *secure by design* It is known to entail financial losses It violates known security principles

• 
  Orville Wright's ghost (@angry_aero) reported 31 minutes ago

  @ProfBrahMos Have you ever looked at the NASA technical reports server? Their GitHub?

• 
  𝑘𝑒𝑟𝑛𝑒𝑙𝑡𝑟𝑖𝑐𝑘 (@kernel_trick) reported 32 minutes ago

  @julesagent why can't i tag jules in a github issue,, such a common feature in other agents //

• 
  Fisher (@hawking520) reported 32 minutes ago

  THE FIRST-100-USERS PLAYBOOK Pick your row. Run the cheapest test that can fail. Get paid before you polish. STEP 1 — Find your type (decide in 30 sec). Run ONLY this row's 2 channels. Dev tool / OSS / infra → Show HN + GitHub Consumer / visual AI → Discord + creator seeding B2B SaaS → Social listening + pre-sell Prosumer / extension → Reddit problem-threads + Product Hunt B2C web app → SEO Mobile app → ASO + micro-creators STEP 2 — Validate in a WEEK, before any code. Run one. If it fails, kill the idea. Fake-door → 1 page: outcome headline + email box. Send 100 visitors (a few replies + 1 small post). Pre-sell (B2B) → DM 20 people who named the problem: "Building X to fix this. Can I put you down for a paid pilot at $___?" Get a written yes or a deposit. Concierge → do the job by hand for 5 people. Charge them. PASS BAR: 20–50 emails OR one real "yes, I'll pay." Miss it → stop. Demand isn't there. STEP 3 — Build only the core. One job, done well. Cut every feature you can defend cutting. Ship when it's embarrassing — polish isn't the test; distribution + willingness-to-pay are. STEP 4 — Run your 2 channels. The exact first move: Dev tool → Show HN with a runnable demo in line 1. Pin the repo. Reply to every comment for 6h. Consumer AI→ Open a Discord. Weekly output challenge. Seed 50–100 micro-creators (not 3 big ones). B2B SaaS → Search Reddit/X/LinkedIn daily for the problem. Same-day help, not a pitch. Ask for an LOI. Prosumer → Find the subreddit where people complain. Post a useful answer; mention the tool only if asked. B2C web → Pick ONE "best [tool] for [specific audience]" phrase. Build the page that wins it. Mobile → Nail the App Store title + screenshots first. Then seed 5–10 niche micro-creators. STEP 5 — Get users by hand (65% start with no audience — that's the normal start line). Daily 30 min: 15 min → find 3 people complaining about your problem. Help. No pitch. 10 min → make one shareable result; ship it publicly. 5 min → DM one warm contact for an intro or a quote. Run it 30 days before you judge anything. STEP 6 — Positioning. Lock before you scale. One-liner: "I help [who] [do what] without [the pain]." Say it to 5 strangers — they must repeat it back. Landing page, this exact order: outcome headline → their own words (paste real Reddit phrases) → 10-sec demo gif → proof → one CTA. Delete the rest. STEP 7 — Get paid (the $0 → $1 unlock). 3 tiers, anchor the middle. Price ~2x higher than feels comfortable. Never paywall a free crowd later — charge from the start. Solo / non-US? Use a Merchant of Record (Paddle or Polar) — handles global VAT, collect legally today. Need fast cash + warm users? Run a lifetime deal (AppSumo).