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:

• 
  Jay.TL (@JayTL00) reported 4 minutes ago

  Three AI labs shipped the same feature within one hour today. That's not competition. That's a signal the unit of interaction just changed. For two years, the atomic unit of working with an AI agent was one prompt. You type. It responds. You type again. Every workflow was a chain of prompts, rebuilt from scratch each time. Today, OpenAI, Anthropic, and Cursor all shipped features that only make sense if the unit is no longer the prompt. The unit is now one workflow. 1. OpenAI Codex Record & Replay (3,807 likes): Do a task once on your Mac. Codex watches. It turns your demonstration into an inspectable, editable skill you can reuse. Not a prompt. A recorded procedure. 2. Cursor /automate (1,085 likes): Describe what you want in plain language. Cursor configures the triggers, instructions, and tools automatically. Plus five new GitHub triggers and Computer Use enabled by default for cloud agents. 3. Anthropic Claude Code Artifacts (6,829 likes): Your coding session becomes an interactive, shareable page. PR walkthroughs, project dashboards, living documentation. Shared at a private link, like a Figma file but for agent work. Each one alone is a feature release. Together they describe the same shift from three different angles: the agent session is becoming a reusable, shareable, composable artifact. Read them as one move: - Input side (Codex): teach by showing, not by writing - Configuration side (Cursor): describe in language, system assembles the wiring - Output side (Anthropic): the result of a session is a shareable object, not a chat log The Karpathy framing was right — we're moving from prompt iteration to plan, execute, verify, loop. What he didn't name is that this loop needs to be portable. A workflow locked inside one chat thread is useless the moment you close the tab. But here's what most coverage missed. Codex Record & Replay requires Computer Use enabled. That means OpenAI is watching your screen while you demonstrate an enterprise workflow. The EU version is blocked at launch. That's not a regulatory footnote — the entire feature is built on continuous screen access, and the EU looked at it and said no. Which raises the question nobody is asking: who owns the recorded workflow? You demonstrated an expense-filing procedure that touches your company's internal tools. Codex turned it into a skill. Where does that skill live? Can OpenAI see it? Is it training data? The product copy says you control when recording starts and stops — but says nothing about what happens to the recording after. There's also a fragmentation problem hiding in plain sight. Three companies, three proprietary formats for the same primitive. A workflow you record in Codex doesn't run in Cursor. An artifact you build in Claude Code doesn't render in OpenAI's product. We're watching the agent-workflow layer fragment into three walled gardens before it even solidifies. This is the SaaS integration mistake repeated, except worse. SaaS integrations are wrappers around APIs. These workflows encode institutional knowledge — how your team ships code, how your finance team files reports, how your ops team handles incidents. That's not data. That's operational IP. The economic implication: every recorded workflow is switching cost. The more skills you build inside Codex, the harder it becomes to leave. The more automations you configure in Cursor, the more your team's muscle memory is locked to one editor. Anthropic's artifacts are softer — they're shareable — but they only render inside Anthropic's ecosystem. The deeper question isn't which feature is best. It's whether the agent-workflow layer will be open or closed. Today, three companies bet on closed. Nobody shipped an export button.

• 
  Maurice Heumann (@momo5502) reported 5 minutes ago

  @disarray00 If you have concrete recommendations, I would love to hear them, either as GitHub issue, maybe even a PR. But also as a comment here, I'd appreciate it. So when speaking about redundancy, what precisely?

• 
  SOURAV PANDA (@i_d_skp) reported 5 minutes ago

  Scenario: You accidentally committed a plaintext database password to GitHub in a .tf file. Fix: Nuke the commit history immediately! Use environment variables (TF_VAR_db_pass) or fetch secrets dynamically at runtime from AWS Secrets Manager or HashiCorp Vault. 🔑 #Terraform

• 
  Andrew Kuncevich (@AndrewK404) reported 5 minutes ago

  Automatic generation of high-quality benchmarks Data for coding agents is scarce. Two pipelines invert the task: instead of searching for tasks, they produce them ① SWE-smith — bug factory Take a working repository -> intentionally break a function until a test fails -> then ask an LLM to write a “human” GitHub issue for that bug. The obvious risk is data leakage ② EnvScaler — environment factory Real tool APIs are closed, and LLM-simulated ones hallucinate. So the environment is synthesized as a real Python class: deterministic state, methods = tools. Plus a validator for each scenario that checks the agent’s path, not just the final answer --- What matters much more is not the implementations themselves, but the intuition behind them: ① You can generate benchmarks / eval cases directly from real data (for example, I constantly do this for RAG) ② It is important to look at the agent trace, not just the final answer (Specifically: did it call unnecessary tools? Did it call the required tools? Did it call the tools in the right order?)

• 
  AI Crave (@wecraveai) reported 8 minutes ago

  Open source NotebookLM alternative with no data limits and AI agents. Same idea as Google's NotebookLM. Same chat-with-your-docs. Same podcast generator. Same cited answers. Except this one has no source limit, no notebook limit, no 200MB file cap, and no Google login. It's called SurfSense. Google NotebookLM vs SurfSense: - Sources per notebook: 50 to 600 → Unlimited - File size cap: 200MB and 500K words → No limit - LLM choice: Gemini only → 100+ models via LiteLLM - Local LLMs: Not allowed → Full Ollama and vLLM support - Self-host: No → Yes, one Docker command - Price: $0, $19.99/mo Pro, or $249.99/mo Ultra → $0 forever Here's the wildest part: It connects to 27+ sources Google can't touch. Notion. Slack. Linear. Jira. GitHub. Discord. Dropbox. OneDrive. Gmail. Confluence. Obsidian. ClickUp. Microsoft Teams. Airtable. Your entire work life, indexed once, searchable from one chat box. 14.4K GitHub stars. 1.4K forks. 6,232 commits. Apache-2.0 license. One honest note: the README says it's not yet production-ready and still being actively developed. But it already does more than NotebookLM does, and the gap is widening every release. This is what NotebookLM should have been from the start. Repo in the first comment.

• 
  Roy (@__roycohen) reported 11 minutes ago

  The Github API is probably causing OpenAI millions of dollars in token spend a month because GPT web cannot actually browse Github properly. At this point hosting your own server for instant reads is likely the way

• 
  Plebian (@Plebian_2) reported 17 minutes ago

  @QuantumTumbler If GitHub wrote a program called an "agentic loop" that simply executed whatever code you find on their site, then they should be banned too. The problem is people are auto-running the LLM output without understanding it. Anthropic could simply release Fable as a chat bot.

• 
  Jared Tate ©️ (@jaredctate) reported 18 minutes ago

  I'm a blockchain developer. North Korean hackers tried to steal my crypto & hack my computer by pretending to recruit me on LinkedIn for a crypto consulting gig. Here's the entire attack: PHASE 1 — THE LURE A "recruiter" (a polished, good looking blonde "British woman in London") messages you on LinkedIn. Real-looking profile, perfect English. She books a Google Meet. On the call it's a different person — a bald Chinese looking guy, who quickly turns his camera off complaining of a bad connection, broken English — who asks two telling questions: "What crypto wallets do you use?" and "What code editor do you use?" That's not an interview. That's target profiling. PHASE 2 — THE DELIVERY He sends a GitHub repo (a "Web3 game" project) and says: clone it, run `npm install`, and open it in Cursor so we can see you know what your doing. It looks like a normal codebase. It's bait — with malware hidden in 3 separate files, each rigged to run automatically. PHASE 3 — PAYLOAD #1: package.json (the trip wire) They added one line: "prepare": "node server/server.js" npm runs the "prepare" script automatically right after `npm install`. So just installing the dependencies launches their code. No button. No warning. PHASE 4 — PAYLOAD #2: auth.js (the backdoor) In one file, the real code ends at line 18. On line 19 — shoved far off-screen behind ~600 blank spaces so you'd never scroll to it — sits obfuscated code that: • collects your hostname, OS, and MAC address • sends your entire environment (process.env) to a North Korean server: 147.124.212.180:1224 • runs whatever JavaScript that server sends back, via eval() = full remote control • repeats every 5 seconds That's a backdoor. The eval() is the doorway they use to push the actual wallet/password stealer. PHASE 5 — PAYLOAD #3: .vscode/tasks.json (the silent one) A second hidden file set to "runOn": "folderOpen" — it executes the instant you open the folder in your editor, with no window and no output: Translation: download a script from their server and run it immediately. THIS is why they demanded Cursor — they were betting it would auto-run the task. VS Code stops and asks "do you trust the authors of this folder?" first. I never said yes, so it never fired. DO NOT USE CURSOR (for many reasons) PHASE 6 — THE FALLBACK When the repo "wouldn't run," he sent a slick website with a "Connect Wallet" button — a straight-up wallet drainer — and said "just connect MetaMask." THE GOAL Drain your crypto and steal your credentials — money that funds the regime. It's a documented operation: the "Contagious Interview" campaign (aka Famous Chollima). HOW TO NOT GET HIT • A "job interview" that requires running their repo = malware until proven otherwise. • Never `npm install` a stranger's code on your real machine. Use a throwaway VM. • Keep serious crypto in cold storage — never a hot wallet on your daily driver. • If a recruiter's story falls apart the second you get on a video call, walk.

• 
  David O. Ehibor 🇦🇷 (@grayontop_) reported 22 minutes ago

  GitHub Copilot didn't make developers faster It made slow developers more confident about writing bad code quickly 😭

• 
  Alexander Cranga (@alexandercranga) reported 24 minutes ago

  Why should I find issues with GitHub faster than they update their status page?

• 
  Oluwatobi O (@ooluwatobig) reported 26 minutes ago

  More trouble for GitHub as Cursor has launched Origin, a product which is essentially GitHub for AI agents

• 
  Gabriel Denys (@gabedenys) reported 29 minutes ago

  @Marcos12345rico I posted a GitHub issue. Assuming you probably want bug reporting mostly there? It's a good tool. Locally I already patched and compiled the app to fix the bug.

• 
  Avi (@AvaneesaBee) reported 30 minutes ago

  @karankendre Good, GitHub is sooo slow these days

• 
  Ghostroot (@Gh0stroot) reported 30 minutes ago

  GitHub published a tool that forces AI agents to understand before they build. 95K stars in days. The problem with AI coding agents was never the model. It was this: You send an idea in text. The agent interprets whatever it wants. Builds the wrong thing. You start over. spec-kit fixes that with 6 commands. /speckit.constitution → sets the rules: quality, testing, architecture. /speckit.specify → you describe WHAT to build. Not the stack. /speckit.clarify → the agent asks what it doesn't understand before writing a single line of code. /speckit.plan → now you choose the technology. /speckit.tasks → ordered list of tasks by dependencies. /speckit.implement → the agent builds. The deliverable is no longer wildly generated code. It's a living specification that your AI reads, validates, and executes step by step. Works with Claude Code, Cursor, Copilot, Codex, Gemini CLI and 25+ agents. 95K stars. 8.3K forks. Published by GitHub itself. MIT license. Before spec-kit: "make me a task app" and you pray the agent doesn't get lost halfway. After spec-kit: specification first. code after. The agent knows exactly what to build. In what order. And why.

• 
  Eugene Ostroukhov (@eeuoss) reported 34 minutes ago

  @neogoose_btw Sure. *** has flaws. And let’s separate ****** GitHub UI from inherent *** issues. But: 1. *** was and is a huge improvement over what was before it (CVS, SVN, ClearCase). Effortless local version control is sick! 2. *** is designed for systems programming and works really great for that. Small self-contained changes. No sprawling rewrites.