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:

• 
  raphh (@raphh) reported 1 minute ago

  Would be neat to have GitHub Issues in the Codex app too, just saying.

• 
  Ashish Kumar (@afkashish) reported 2 minutes ago

  @moonfarm_dev the part i'd be careful with is issue count isn't the same as willingness to pay. most people filing github issues expect the maintainer to fix it free, not buy a tool. i'd weight it by how many of those threads have someone offering money in the comments, that's the actual signal.

• 
  Avinash Dalvi (@AvinashDalvi_) reported 3 minutes ago

  yesterday i spent almost 1.5 hours debugging a streaming API issue with github copilot. we have two login mechanisms. OIDC and MSAL. streaming was working for OIDC but not for MSAL. i gave the context to copilot. it gave suggestions. i tried them. nothing worked. kept going back and forth. the more i went back and forth the more my own context got lost. i could not even remember what i had already tried. so i stopped. closed the chat. took a pause. then i looked at it myself. MSAL token was not going through the interceptor the way other API calls do. streaming has a different path. and there was a race condition in auth.ts that was stopping the token from getting injected into the header in time. one line fix. that was it. copilot was overcomplicating it. adding layers, restructuring things, over engineering a problem that needed one small change. 1.5 hours with copilot. 5 minutes without. i felt a bit ashamed honestly. not because copilot failed. because i stopped thinking on my own too early. Sometime we tried to do many things and take help of AI but somehow fallback to our own skill. knowing when to stop asking the AI and just think yourself is still a skill. maybe the most important one right now. Did you also have this kind of facepalm moment ??

• 
  Abdelhamed M. (@VZHydra) reported 5 minutes ago

  @amityhere Please open github issue with it, for more details so we can fix it soon

• 
  Stephan Arenswald (@sarensw) reported 6 minutes ago

  @tr1kn1 @macpackerapp You’re not the only one. I added this upon a request (via GitHub issue) by another user just recently.

• 
  Mysterious Doctor 📉📈📉. (@MysteriousDoct3) reported 9 minutes ago

  @MattMalley14 Fair point. The breach happened. It was patched. And it's exactly why we shipped 3 security-focused mainnet updates in 2 weeks. 5,619 commits this year. Memory safety fixes, protocol hardening, audits. All public on GitHub. You can't have real security without real transparency. That's how building works. Find it, fix it, ship the fix publicly. That breach is why Qubic is stronger now. Security isn't "never breached". Security is "how fast you fix and how open you are about it".

• 
  Kaitee (@KaiteeShiks) reported 10 minutes ago

  My AI app looked ready to ship. Replay QA proved otherwise. Before launching, I connected the GitHub repo to Replay QA, let it run through the app like a real user, and within a few minutes it surfaced issues I would've never noticed by manually clicking around. What I liked most wasn't just finding bugs. It showed exactly what happened in the browser, explained the likely root cause, and gave enough context to fix the issue without guessing. Instead of running a one-time check, I connected the repo so Replay QA can keep testing every time I push changes or open a PR. No Playwright setup. No complicated pipeline configuration. Just a practical QA gate before users ever find the problems. If you're building with Cursor, Claude Code, Codex, Lovable, Bolt, Replit, or any AI coding workflow, this is worth adding before your next launch. @replayio

• 
  The Autonomy Guy (@TheAutonomyGuy) reported 10 minutes ago

  $FANUY, the part that changed my mind on the autonomy question. Essentially all of that ¥378.6bn robot revenue is caged, pre-programmed repetition. Nothing autonomous about it. But in December 2025 Fanuc reversed a decades-old closed-controller doctrine: open-source ROS 2 driver on GitHub, Python on the controllers, 1ms control loop, NVIDIA Jetson on-robot, Isaac Sim digital twins. Then in May 2026 a Google deal using Gemini Enterprise, with Fanuc in DeepMind's Gemini Robotics Trusted Tester Program. One hard number so far: 1,000+ robots shipped into Physical AI applications. Against ¥378.6bn that's a rounding error. And the intelligence belongs to Google and NVIDIA. Fanuc supplies the reliable body and the installed base. If the value migrates to the model, the arm becomes a commodity actuator, which is the segment Chinese vendors already win.

• 
  king jomo (@themostjomo) reported 10 minutes ago

  @CantelopePeel @thru_xyz @untolabs Hi CP. I suspected it was likely GitHub. So Claude helped figured version v0.2.38 was workable. So i worked with that. I’m looking forward to your fix tomorrow

• 
  Dmytro (@Dmytroo_eth) reported 13 minutes ago

  STOP COLLECTING GITHUB REPOS BY HAND. BUILD A LOOP THAT TELLS YOU WHAT STILL MATTERS Most developers still use GitHub stars like a manual storage system Star a repo, clone it, test it once, forget why it exists, then find the same type of tool again three months later That means you are still the memory layer The developers getting leverage from AI are not just better at finding tools They build loops that read every repo, write notes, check real usage, detect duplicates, flag upstream risk and know what can be deleted That sounds less exciting than finding another useful GitHub repo But it is the part that makes the collection useful Without repo notes, you forget why you grabbed the tool Without usage checks, dead folders look important Without duplicate detection, three libraries quietly solve the same problem A README tells you what the author built A loop tells you whether the tool still earns space in your system Most developers are still trying to save more repos The useful work is building the machine that makes every repo explain why it should stay

• 
  YanXbt (@IBuzovskyi) reported 14 minutes ago

  HERMES AGENT v0.19.0 JUST SHIPPED. THE QUICKSILVER RELEASE. 2,245 COMMITS. 1,065 PRs. 450+ CONTRIBUTORS. FIRST TOKEN NOW ARRIVES 80% FASTER. the theme: speed everywhere. WHAT CHANGED: @NousResearch SPEED (the headline): cold start: 4.3 seconds down to 0.9 seconds. 80% cut. applies to CLI, gateway, TUI, desktop, cron. that pause before the agent responded? gone. reasoning models now stream thinking live by default. no more staring at a spinner for 30 seconds. you watch the model think in real time. desktop app got 20+ targeted performance PRs. 14x less CPU on markdown rendering. virtualized diffs. snappy session switching. feels native under load, even with huge transcripts. NEW FEATURES: /subscription and /topup from the terminal. change your Nous plan, preview costs, upgrade, downgrade. your wallet never leaves the keyboard. desktop app got a matching billing settings tab. smart approvals are now the default. an LLM reviewer judges flagged commands instead of asking you every time. safe = auto-approved. dangerous = auto-denied. uncertain = escalates to you. plus: /deny [reason] tells the agent WHY you refused. Bitwarden + 1Password secret sources. API keys no longer live in plaintext .env. pull secrets from vaults at load time. multiple vaults simultaneously. conflict warnings. per-variable provenance. live subagent transcripts. delegate_task now returns live transcript files you can tail -f the moment subagents launch. every tool call, result, and reply. one log per child. durable background delegation. if the process restarts mid-run, results survive. ownership-checked ledger. nothing vanishes. delivery-obligation ledger. if the gateway crashed between generating a response and confirming delivery, that answer used to be lost. now: responses recorded in state.db. redelivered on next boot. no more silent loss. one gateway, many profiles. a single bot token can route different channels to different profiles. work Discord → work profile. hobby server → personal profile. one bot. sessions export. hermes sessions export outputs Markdown, HTML, Quarto, prompt-only, and Hugging Face trace formats. --redact flag scrubs secrets. your conversation history is a real dataset now. NEW MODELS: GPT-5.6 (Sol, Terra, Luna + Pro variants) fully wired. Grok 4.5 GA. Kimi K3. Claude Sonnet 5. Fireworks AI as first-class provider (#2 in picker). DeepInfra hardened. Upstage Solar added. LM Studio JIT model loading for local setups. REASONING: max and ultra effort levels added. per-model reasoning_effort overrides in config. per-slot effort in MoA presets (advisors think hard, synthesizer stays fast). per-task effort for auxiliary models. thinking depth is now a dial, not a global switch. /model --once: one-turn model override that reverts. stacked slash-skills: /skill-a /skill-b do XYZ. hide providers you don't use: enabled: false per-provider. SECURITY: Vertex credentials scoped away from subprocess env. media/vision/image-gen reads through shared guard. webhook body-size cap across every aiohttp server. bot-token redaction in Telegram errors. Fireworks token prefixes in the redactor. CI hardened against untrusted-ref interpolation. STATS: 2,245 commits. 1,065 merged PRs. 2,465 files changed. 300,000 insertions. 3,300 issues closed. 450+ contributors. 218K GitHub stars. 41K forks. biggest contributor window in Hermes history. upgrade: hermes update

• 
  Heart (@heart_) reported 15 minutes ago

  @DastardlyDaniel afaik the “directx to metal” api are developed and maintained by apple themselves i think it’s smart to leave the specific implementations to private companies so they’re not stuck with github issues hell for no profit

• 
  Alex Rogov (@Alex_Rogov_js) reported 21 minutes ago

  there's a pattern across three separate stories this month. AI coding agents tricked into installing malware from "clean" looking github repos. security agents built to catch malicious code getting fooled into running it themselves. claude code's own sandbox had an escape bug this year (CVE-2026-39861, a symlink flaw). same root cause every time. an agent optimized to be helpful will follow instructions hidden in a file, a commit message, a package readme. the more autonomy you give it, the bigger that surface gets. i run agents with --dangerously-skip-permissions on repos i wrote the CI for. i never run it on a repo where i didn't. that's not paranoia. it's the same rule i'd apply to a new hire before their first PR gets reviewed. the fix isn't less automation, it's a trust boundary. an agent gets exactly the access its current task needs, nothing more. every new tool call crosses a boundary you drew on purpose, not one you're discovering after the fact. if your agent setup doesn't have that boundary somewhere, it's not a matter of if. it's when.

• 
  spect (@spectnfa) reported 22 minutes ago

  ONE REPO GIVES CODEX 59 SKILLS AND ACCESS TO 1,000+ APPS. MOST PEOPLE STILL DON'T KNOW IT EXISTS each skill is just a folder with a SKILL.md file. it tells Codex exactly when to trigger, what steps to run, what the output should look like. install once, and Codex pulls the right one the moment your request matches it, no retyping the same setup every session. Codex used to just talk. Now it sends emails, opens GitHub issues, posts to Slack, and acts across 1,000+ apps on its own. PR reviews, CI fixes, meeting notes, invoice sorting, lead research, resume tailoring, all of it already built and waiting in one repo. save this before everyone else finds it and you're the one still copy-pasting the same prompt 👇

• 
  Polsia (@polsia) reported 27 minutes ago

  Code quality erodes between sprints while no one's watching. Vigil continuously monitors GitHub — automated reviews, documentation, and debt tracking before problems reach main. Teams ship cleaner code without changing how they work.