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:

• 
  Saad | building CoralSwarm & CrossTask (@verdaxxed) reported 24 seconds ago

  Github going down is such an incumberance on work.

• 
  Seregon (@SeregonWar) reported 24 seconds ago

  @thiago__barrios Hi, unfortunately I don't have an Android phone (I'd need to install Android Studio). Could you open an issue on GitHub explaining the problem so you can be more thorough and we can keep track of it?

• 
  Ridark (@ridark_eth) reported 42 seconds ago

  CLAUDE + OBSIDIAN TURN A MESS OF SAVED GITHUB REPOS INTO A SEARCHABLE VAULT YOU CAN ACTUALLY THINK WITH a github star is not memory. a cloned folder is not memory either. once you save 30+ tools, the real problem is not finding new repos. it is remembering why you grabbed them, whether you still use them, and which ones quietly do the same job. that little graph is what starts to happen once claude keeps writing the missing context down for you instead of letting every useful repo disappear into another folder. it’s basically a repo memory system: -> one markdown note for every repo you clone -> claude reads the README, key files, and checks whether the tool shows up in your own projects -> every repo gets a status like in-use, shelved, duplicate, or unclear -> another loop reads across the whole collection and groups overlap -> stale dependencies get flagged before they surprise you later -> everything lives in plain markdown inside obsidian the good part is that the graph is not just for looks. once enough notes exist, it starts showing clusters, duplicate tools, forgotten experiments, and the repos you still depend on even though upstream has barely moved. instead of 30 half-tested folders with zero memory attached, you get one vault that keeps track of what each repo does, why it mattered, and whether it still deserves space on your machine. bookmark this before your github folder becomes a pile of tools you vaguely remember installing and completely forgot how to trust.

• 
  boozie (@soboozie) reported 3 minutes ago

  CLAUDE CODE HAS 5 WAYS TO RUN ITSELF AND MOST DEVS ONLY USE 1 OF THEM. The Claude Code team defines a loop as an agent repeating a cycle of work. There are 5 ways to trigger one. Turn-based is the default, and it's the slowest. Write a prompt. Wait. Check the result. Fix it. Repeat by hand. You're the loop. Goal-based removes the babysitting. Define the finish line once. Example: "Get the homepage Lighthouse score to 90. Stop after 5 tries." An evaluator model checks the work after every pass. Condition not met, sent back to work. Condition met, or the 5 tries run out, done. You never touch it in between. Time-based (/loop) runs that same cycle on a timer. Same prompt, same check, repeated automatically until you cancel it. Schedule (/schedule) is built for recurring jobs. Triggered by a set interval, not a person opening a laptop. One real setup: an agent watches Slack and GitHub for bug reports. A second agent picks up each one, works until it's ready, opens a PR, and notifies you to merge. Runs in the cloud whether your laptop is open or not. Proactive loops stack all 4. Nobody has to press start. It behaves like an employee with a job description, not an assistant waiting for a message. It already knows the job. It just keeps doing it, every day, without being asked. Most devs are still on mode 1 of 5. The other 4 are why some ship while they sleep. An assistant waits for instructions. An employee already knows the job.

• 
  agent maxxer (@claudeabuserr) reported 3 minutes ago

  @Robertg761_ @robinebers Sorry, terrible grammar on my last reply, im faded off cough medicine rn sick as a dog. I do use github but not like publicly as ive seen people doing (meaning spreading projects or using it as a social media platform). I primarily use it to sync my obsidian vault between my windows desktop set up and my mac so I can vibe code/build on the go.

• 
  Fluixo (@fluixoo) reported 7 minutes ago

  Most people do not need another $30 SaaS subscription. They need to open GitHub. - n8n for automations - AppFlowy for workspace - NocoDB for databases - Twenty for CRM - Whisper for transcription - Langfuse for AI observability - yt-dlp for video and subtitles You will still pay for a server, storage, GPUs, or API keys. But you stop renting every small feature from another company. Replace three subscriptions and you already save real money.

• 
  muzz khan (@muzzdotdev) reported 10 minutes ago

  @luisflarota Lol it was github, we’re working on making these problems more visible

• 
  Lord Bean (@lifeisameeme) reported 13 minutes ago

  Most browser automation tools spin up a brand new, empty browser for the AI agent to drive. chrome-cdp-skill does the opposite. It connects to the Chrome you already have open. Same tabs, same logged-in sessions, same page state mid-workflow. No re-login, no cookie exports, no separate profile. Flip on remote debugging in chrome://inspect and your agent can read a Gmail tab or click around GitHub exactly as you left it. The interesting design choice: it holds one persistent daemon per tab instead of reconnecting on every command. That's apparently the difference between handling 100+ open tabs cleanly and getting stuck re-showing Chrome's 'allow debugging' popup on every single action, which is the actual complaint people have about the more popular chrome-devtools-mcp. 3,177 stars for something that's basically just talking to Chrome directly, no framework in between.

• 
  Parth Bhuptani (@ParthBhuptani10) reported 15 minutes ago

  Day 33 & 34 18 -19 July -Gone to college for external viva -Had guests at home -Sunday is just another wasted day -Leetcode and GitHub streak broken -Spain won the world cup These two days were totally wasted for me

• 
  Sarah Nzeshi (@justsa_rah) reported 16 minutes ago

  @AnthonyClo36464 Click home and sign up again, it will allow you register with github I will find the repo later and make an issue

• 
  Tnax (@Fromp_Tman) reported 17 minutes ago

  @BocaBola_ I had this issue on Nobara, I had a ton of issues with kde and x11-wayland bridge on it. I installed a new release of fedora with kde next week. My only issue with sleep is due to motherboard (had to use some guys script from github) and my main monitor suddenly being 2980x1260

• 
  Ben Eng (@jetpen) reported 24 minutes ago

  Note to self: submit a hermes agent github issue for enhancement request to provide a better explanation for a "Command Approval Required" request for a Python HERE script. Provide a short one-line explanation of intent, including a single word classification of the most dangerous operation: create, update/edit, delete, privilege elevation (sudo), etc. I don't want to have to inspect the whole script, especially if it large, to find the danger. It's a needle in a haystack. Give me a classification with which if the danger is low or the operation is expected in the context of the task, I can quickly approve without wasting time. I only want to inspect when something looks out of the ordinary.

• 
  Alankrit Verma (@AlankritDotMe) reported 24 minutes ago

  Does GitHub use GitHub to build GitHub? Because now everything makes sense. Not trying to hate, but it is slowing us down. Are we still here because GitHub is good, or because moving is hard?

• 
  👨🏻‍💻 ⚡️ (@EadrictheWild) reported 25 minutes ago

  @Andrewislington I am just going to use bedrock in the platform for BYOC for Claude. I figure just getting to use Claude Desktop was a good enough step for me. I just run the Github and AWS MCP server. Runon soon in AWS . Entire stack is AWS only though.

• 
  Riya Bisht (@b1shtream) reported 25 minutes ago

  fun & insightful weekend watch: - A tiny team, or even one person, can now build and run what used to need 50 people Fun fact: the AI models haven't caught up to this either. Claude Code will say a task takes "three weeks," then finish it in an hour. - Don't ship garbage: The problem isn't that AI writes too much code. It's that demo code looks fine but breaks in production, and it hallucinates. Fix is to get to 80–90% test coverage before shipping. Garry's most-used habit is a "plan then review" step he runs ~20 times a day. - Lines of code is a useless way to measure progress. The only real test is whether it works for you and your customers, and whether people actually pay. - Sitting at a terminal, one person can do the work of 500 to 1,000 people. So most of our assumptions about what a founder or small team can pull off are off by roughly 1,000x. - Old companies run "open loop." Decisions get made, feedback comes back slowly and lossy, errors pile up. AI lets you close the loop by giving an agent read access to everything the company produces. - The coding pieces map cleanly onto a company: a skill is an employee's ability, a resolver is the org chart, filing rules are internal process, the testing step is audit and compliance. - Roles shrink to three: everyone builds (even salespeople automate their own pipeline), someone owns each outcome (the DRI), and a new "AI founder" who lives at the frontier and tries every new tool. If you're still working like it's last year's Copilot, you fall behind fast. - Student version of this: point an agent at your GitHub and Discord, record your team meetings, and let it suggest what to work on next. - Writing code is getting close to free. Taste isn't. Knowing what's actually good is the thing that lasts. - Public benchmarks don't tell you if your product is good. The only judge that counts is whether users want it, and that's different in every field. so you have to sit and read the actual transcripts of what your agent did, mark what's right and wrong, and turn the failures into tests. - A "skill" is basically a runbook. Steps you'd write down to repeat a task, except the AI can follow it and it can also call code. - A "resolver" keeps the AI from drowning in instructions. Instead of one giant config file, you keep an index and load the specific instruction only when it's needed. - "Skillify": do a task once, get it exactly right, then save it as a reusable skill. Catch is, writing it is maybe 2 of the 10 steps. The other 8 are testing and making sure it actually triggers when it should. Same reason real companies have compliance teams.