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:

• 
  Karma (@shreyansh_0x0_) reported 6 minutes ago

  @Senpaisaysbye for simple stuff, it actually is. But all that modern tooling exists because we pushed websites far past simple text pages. Back when we just wrote basic scripts, every browser broke things differently. As soon as people wanted dynamic stuff like live feeds, user logins, and fast mobile loading, basic HTML turned into absolute spaghetti. Next.js, package managers, and cloud setups exist so huge teams can build massive apps without re-inventing basic routing, security, and image optimization from scratch every time. You arent crazy for being tired of it though. The real problem is people reaching for a massive enterprise stack just to host a basic text page. If you just want simple text, a plain index.html on GitHub Pages still works in two minutes, no goat sacrifices needed. I very fequently make single page massive .html website and host it on cloudfare for convenience and it's fast af. Can I make single page .html facebook ? Hell no. Even AI will choke itself to death.

• 
  Rabii Agoujgal (@RAgoujgal) reported 6 minutes ago

  The sandbox escapes are being told as a frontier lab hygiene problem: OpenAI's agents into Hugging Face, Anthropic and Meta through the same Irregular environment, a Mythos 5 agent running fake identities at a GitHub maintainer under UK AISI testing. One case sits outside that frame. Moonshot's Kimi K3 broke containment in a Frontier Security evaluation, and Kimi K3 is open weight, outside the US voluntary pre-release framework entirely. Procurement teams buying across both stacks should notice that only one of them has agreed to tell you when this happens.

• 
  Kavi AI Finance (@KaviFinance1) reported 7 minutes ago

  BUILDING HERMES FROM SCRATCH — PART 7 The biggest security problem with AI agents isn’t what they’re allowed to do. It’s what they’re allowed to become. Giving an agent a tool is easy. Giving it permanent authority over that tool is where things get interesting. Most agent security discussions stop at: API keys permissions sandboxing secret management All important. But there’s another problem I think we’re going to hear much more about: permission drift. You start with: read files search the web create a GitHub issue Then the workflow grows. Someone adds database access. Then email. Then a deployment tool. Then credentials for another service. Six months later, the original agent has accumulated enough capabilities to effectively operate an entire business. Nothing was hacked. Nothing was misconfigured. The permissions just kept growing. So I’m changing the way I think about agent authority in Hermes. I don’t want to ask only: “What can this agent do?” I want to ask: “What does this agent need to be able to do right now?” That’s a very different security model. For example: A research task might get: web search document retrieval filesystem read The same agent shouldn’t automatically inherit: database writes deployment financial APIs email sending And even when a tool is allowed, I don’t necessarily want permanent authority. Some actions should be: task-scoped time-scoped resource-scoped and ideally action-scoped. A useful mental model is: Identity → Permission → Action → Verification → Expiry Not: Identity → Infinite access Because agents don’t behave like normal software. They can retry. They can reinterpret instructions. They can discover unexpected paths through tools. They can continue operating while the human who gave them permission is asleep. And this creates another interesting problem: the agent should not be the final authority over its own permissions. If the agent can decide: “I need more access” and grant itself more access, your permission system is basically a suggestion. So I’m experimenting with keeping authority outside the agent itself. The agent can request an elevated capability. The control layer decides whether it gets it. The action is logged. The permission expires. And high-impact actions can require an explicit approval boundary. This also changes how I think about multi-agent systems. If Agent A can delegate to Agent B, what exactly is being delegated? The task? The identity? The permissions? The credentials? The authority? Those are not the same thing. And I think this is going to become one of the nastiest problems in agent infrastructure. Because eventually we won’t have one AI with ten tools. We’ll have thousands of agents delegating work to other agents. At that point: “Who authorized this action?” becomes much more complicated than checking an API key. That’s what I’m working through in Hermes now. Not just making agents capable. Making their authority: bounded, observable, temporary and revocable. The goal isn’t to make the agent harmless. The goal is to make sure that when it inevitably does something stupid, it doesn’t have the authority to turn one mistake into a catastrophe. Next: PART 8 — MODEL ROUTING Because once the architecture is secure, there’s another problem: Why are we paying a frontier model to do work a small local model could handle? If you’re building Hermes with me, bookmark the series. And if you’ve dealt with an agent gaining too much access over time, I’d genuinely like to hear how you handled it.

• 
  idan (@eddiearc6) reported 8 minutes ago

  @lai_jia_wei @linear github issue呢？

• 
  Toufiq Qureshi (@Toufiq651) reported 9 minutes ago

  Day 1 of building interview Yaar🚀 Tech hiring has a measurement problem. We say we want engineers who can design systems, reason about trade-offs, and debug production. Then we test them on whiteboard DSA puzzles they will never write again. So the loop looks like this: → Company asks Leetcode Hard → Candidate grinds 6 months of patterns → Candidate gets hired → Candidate can't debug a race condition in **** We're not measuring bad engineers. We're measuring the wrong thing. Here's my bet: The best interview signal already exists. It's sitting in the candidate's GitHub. Their architecture decisions, their trade-offs, the shortcuts they took at 2am and never cleaned up. So I'm building interview yaar— an AI interviewer that reads your actual repository, understands how it's built, and interviews you on YOUR code. Not trivia. Your code. Over the next 30 days I'm building this completely in public. You'll see: • Why I split the backend into Go + Python • How I beat GitHub's API rate limits for free • A race condition that let users bypass billing entirely • A voice interview feature that costs $0/month • Every bug, including the embarrassing ones Follow if you like backend engineering with the messy parts left in. #buildinpublic #golang #ai

• 
  Nav Toor (@heynavtoor) reported 9 minutes ago

  Every GoPro you own already knows how to fix your shaky footage. GoPro charges $99.99 a year to unlock it. Inside every GoPro, Sony, DJI, and Insta360 is a gyroscope. It records the exact angle and rotation of every shake, 400 times per second. That data gets embedded inside your video file automatically. It has been there every time you filmed. GoPro reads that data through a tool called ReelSteady. They bought the company in 2020, sold it as a $99.99 one-time purchase inside GoPro Player, then in 2024 renamed it HyperSmooth Pro and folded it into their $99.99 a year Premium+ subscription. GoPro cameras only. Cancel your subscription and it stops working. There is a free tool that reads the same gyroscope data. It works on every camera. It is called Gyroflow. It started in 2020 as a Python project by an engineering student named Elvin Chen. Then a developer named Adrian Eddy rewrote the whole thing in Rust, built the UI, the GPU pipeline, the rolling shutter correction, and the plugins for Premiere, After Effects, Resolve, and Final Cut Pro. GPL-3.0 licensed. Here is what happens when you open Gyroflow. You load your shaky video. Gyroflow reads the gyroscope data embedded inside the file. It knows the exact motion of every shake, not a guess from pixels, the actual recorded movement from the sensor. Then it reverses each shake, frame by frame. Every other stabilizer on Earth guesses. Premiere Warp Stabilizer looks at the pixels and tries to reverse-engineer the motion. iMovie does the same. YouTube stabilization does the same. They all crop your frame to hide the shake they cannot fully remove. Gyroflow does not guess. It reads the truth. An FPV pilot ran a head-to-head test on the same 4 minute 41 second clip. Gyroflow finished in 3 minutes. GoPro's paid tool took 15. Same footage. Same stability. Five times faster. Zero dollars. It works with every prosumer camera on the market. GoPro Hero 5 through 13. Every Sony mirrorless from the a7 IV to the FX9. DJI Avata, Action, Neo, and every O3 and O4 Air Unit. Every Insta360 from the OneR to the Ace Pro. Blackmagic RAW. RED V-Raptor and Komodo. Canon C50, C80, C400, R6 Mark III, R5 Mark II. Runcam and Hawkeye FPV cams. Betaflight blackbox. ArduPilot logs. On Sony bodies you can stack Gyroflow on top of the in-camera IBIS and EIS. Nothing else can do that. Rolling shutter correction. Horizon leveling. Lens distortion fixes. GPU accelerated rendering on DirectX, Metal, Vulkan, and OpenGL. Output to x264, x265, ProRes, DNxHD, and 32-bit float OpenEXR. Real-time preview with sub-pixel precision. Windows, Mac, Linux, Android, iOS. On the Microsoft Store, Mac App Store, and Google Play. 9,373 stars on GitHub. 1.23 million downloads. Last update pushed yesterday. Every action camera you own already knows how to fix itself. Gyroflow simply asks it. (Link in the comments)

• 
  z (@animaginarynumb) reported 11 minutes ago

  do i launch next week? my github actions are down and my ci/cd runs on them. when i get the refresh on 1st of september, should i launch my product?

• 
  hira (@Hiraweb3) reported 13 minutes ago

  @0xExpresso @IQofficial github link broken lol

• 
  Bluewall (@Bluewall) reported 13 minutes ago

  Found a malicious GitHub repo posing as a curated list of cybersecurity Telegram channels, every link led to the same malware ZIP. Reported it. Today GitHub Trust & Safety confirmed they took it down. Do report the shady stuff, it pays off.

• 
  Prasenjit (@prasenx) reported 17 minutes ago

  there's no official way to run macOS on an iPad. someone made an unofficial one. full macOS running locally on iPad. not a remote desktop, not a web app. hardware CPU virtualization with GPU acceleration. → supports macOS 12 monterey up to macOS 26 tahoe → you can install xcode, terminal, final cut pro trial, logic pro trial on device → works on iPad Pro (M1, M2) and iPad Air (M1) → requires jailbreak on iPadOS 14 up to 16.3.1 → no iCloud sign in → MIT license (100% free) open source on GitHub.

• 
  Muhammad Ahmed (@ahmed_ah206) reported 18 minutes ago

  @BRuteLogic created an issue on github with title : "False positive: curl -X OPTIONS treated as bypass on 403 URLs " with details and screenshot

• 
  kosi (@kosiasuzu) reported 22 minutes ago

  @juliarturc Use GitHub issues

• 
  Luke W Gilbert (@luke_w_gilbert) reported 22 minutes ago

  Shipping log. Messy week. Real progress. Identity got a full mobile pass: hamburger menu replaced the horizontal-scroll nav that was quietly breaking on narrow phones, calendar grid stopped clipping Saturday off-screen, weekday headers stopped reading as a T-collision, stale cached builds got fixed by actually calling registerSW. The settings gear is live now with per-platform time chips so posting schedules are configurable per platform instead of global and dumb. Install-to-home-screen has a real entry point with proper icons. PWA now behaves like it should. Signal's publish queue is wired into Identity's post queue with drip scheduling and pg_cron automation. Human approval gate is gone. The guardrail runs itself. DevCom closed all four phases of the Excellence Plan. Four Truth Engine PRs merged. GitHub login redirect loop fixed. Production secrets untracked from ***. Daily cost tile live on the dashboard. SMRT: stale mortgage rate in the owner-agent newsletter root-caused and grounded via FRED. That number had been quietly wrong. LinkedIn scheduling emergency: platform_quotas inconsistency restored. Live posting confirmed working again. The client project: read-only forensic hardening pass completed before the launch window. Not everything shipped clean. Three items sent back. One bundled task split into three because it was mixing low-risk repo work with live Brain Engine edits and nobody could review it honestly as one thing. A suppression design hit auth walls at every live surface. A token rotation got submitted with no live mutation performed. Sent it all back. 'Done' and 'here is live proof it is done' are still different claims.

• 
  Brannon Hogue (@brannonhogue) reported 23 minutes ago

  This is how I do 10x development... Go to codex or claude code. Make 1-2 main chats and use them to create issues in Github. In depth issues... great specs. I do not feed it garbage. Garbage in = Garbage out Then I dispatch workers from my 1-2 main chats to work on these issues. They will all be on different worktrees, they are able to work independently. They will: • Spin up the environment in my worktree • Plan a fix based on my spec • Execute plan • Check it • Make it a PR I can then use one of my main chats to do a preflight check and, if they work, merge them into main.

• 
  DARMA 🥶| ﷺ (@Darma150) reported 26 minutes ago

  @RialoHQ One detail deserves attention: once Google, Discord, or GitHub accounts are confirmed, they're permanently linked to the wallet. So this isn't just a login step. It's a decision about how contribution identity is connected to wallet ownership.