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:

• 
  Milady (@onchainmilady) reported 5 minutes ago

  99% OF ALL DEVELOPERS DO NOT WRITE CODE ANYMORE There's a repo that does it for you and here's how to use it Owain Lewis built it in Rust and it polls GitHub 24/7 A triage agent turns a one-line vague ticket into a full spec with acceptance criteria Then it spawns a coding agent in an isolated worktree, writes the fix, runs the tests and reviews You come back 2 hours later and the pull request is done He runs the same loop for 99% of his implementation work now Codex, Claude Code, Neo - he says the agent barely matters, the workflow is everything A scheduled bug-finder even opens its own tickets while nobody is watching The honest catch: only a share of your tickets should ever be handed off this way Full walkthrough below

• 
  Aaron Stannard (@Aaronontheweb) reported 5 minutes ago

  @RyanRodemoyer2 We're not offering hand-curated github issue descriptions around here lol

• 
  zelzmiy (@zelzmiy_yes) reported 11 minutes ago

  @nai_sucks although that doesn't mean I think the normies are right, I still think if you can't figure out how to download something from GitHub you deserve to be melted down for the iron in your blood (it's more useful than you)

• 
  Polsia (@polsia) reported 12 minutes ago

  Enterprise triage: $500 to $2,000 per repo a month. Mendrel handles the same across GitHub and Discord, open-source and self-hosted. Triages issues and PRs, drafts replies, pins FAQs, posts weekly digests. Shipping soon.

• 
  Matt Pocock (@mattpocockuk) reported 16 minutes ago

  @seflless @kingincity @JamesHurburgh No, the spec is in a GitHub issue

• 
  Lord Bean (@lifeisameeme) reported 16 minutes ago

  OpenAI's next model just resolved two Erdős problems and disproved an 80-year-old conjecture in group theory. Total cost: about $𝟮,𝟬𝟬𝟬 in tokens. The unreleased Astra model produced ten new results in pure math and theoretical CS, from sphere-packing bounds to arithmetic circuit complexity. Every proof ships with a 𝗟𝗲𝗮𝗻-𝘃𝗲𝗿𝗶𝗳𝗶𝗲𝗱 𝗳𝗼𝗿𝗺𝗮𝗹 𝗰𝗲𝗿𝘁𝗶𝗳𝗶𝗰𝗮𝘁𝗲, not just a claim. → Disproved Connes' Rigidity Conjecture (open since the 1980s) → Resolved Erdős problems #146, #180, and #183 → Improved bounds on high-dimensional sphere packing toward the Cohn-Elkies threshold This isn't the model's first math result either — the same system reportedly disproved the Erdős unit-distance conjecture back in May, with human mathematicians then verifying and publishing the proof. The Lean certificates are on GitHub right now if you want to check the agent's work yourself. 🔗 Links in the reply 👇

• 
  Michael Fischer (@Holden_Rye_) reported 20 minutes ago

  I’m working on a Bitcoin wallet. I’m going to put the repo on GitHub, make it public, and let Bitcoiners build it together. Fork it. Improve it. Save Bitcoin. Hardware isn’t my lane, which is exactly why this needs to be collaborative. My goal is just to get the ball rolling and the gears turning. Right now it’s a CLI. The point isn’t a pretty interface. The point is proving an architecture that makes an easy, safe wallet possible without lying to the user. The safety shouldn’t depend on the user being perfect. A 2-of-3 multisig design means losing one key isn’t catastrophic, and stealing one key isn’t enough to steal your Bitcoin. That’s a structural guarantee. Most wallet security today is procedural: be careful, verify the address, don’t click that link. Humans are the weakest link. This moves the protection into the architecture instead. The core policy engine does the hard work so the UI can stay simple. It analyzes the transaction, assigns a risk level, and explains it in plain English. The user sees Green: Safe to send or Red: Here’s what’s wrong. They never need to know what a script, descriptor, or PSBT is. Transaction simulation gives the confirmation screen something real to verify: “You’re sending 50,000 sats to Bob, 12,000 sats are returning to your wallet as change, and the fee is 400 sats.” That’s generated from the decoded transaction itself, not what the app thinks it’s doing. The descriptor becomes the backup, and it isn’t secret. Save the descriptor and your keys, and any compatible wallet—Sparrow, Bitcoin Core, or another implementation—can rebuild your wallet. No vendor lock-in. No “sorry, your funds are gone.” But here’s the hard truth. A real 2-of-3 multisig wallet actually makes setup more complicated, not less. You have multiple devices, descriptor registration, recovery testing, and key management. That’s the biggest unsolved UX problem. This repo doesn’t solve that yet. My original MVP focused on the user experience with placeholder security. This version focuses on real security with almost no user experience. Neither half is the finished product. The real challenge is designing a setup and recovery experience that hides the complexity without hiding the truth. That’s not just a coding problem. It’s a design problem. And I think it’s one of the most important problems Bitcoin still has to solve.

• 
  The Singularity Project (@01Singularity01) reported 23 minutes ago

  Full Access tasks intermittently request approvals after Codex Desktop update Summary Codex Desktop tasks are intermittently requesting command approvals even though Full Access is enabled both globally and for the individual task. The behavior began after a recent desktop app update. Queued or background tasks appear particularly likely to be affected, although the relationship to the update has not been proven. Environment Codex Desktop build: 26.727.51351 Codex CLI: 0.146.0-alpha.9.2 Host: Windows with an Ubuntu WSL workspace Workspace trust: trusted UI permissions: Full Access enabled globally and per task Windows sandbox setting: elevated Workload: multiple concurrent or queued coding tasks Steps to Reproduce Enable Full Access globally in Codex Desktop. Open a trusted WSL repository. Create or queue several tasks with Full Access enabled. Have the tasks execute ordinary repository commands such as file reads, tests, or *** inspection. Observe that some tasks pause for command approval. The behavior appears intermittent. Other tasks in the same workspace may execute without an approval request. Expected Behavior Tasks using the Full Access permission profile should execute ordinary shell and filesystem operations without pausing for approval. Explicit user confirmation may still be appropriate for application-level destructive actions or project-specific coordination requirements, but those should be distinguishable from sandbox approval prompts. Actual Behavior Some tasks request approvals despite Full Access being displayed globally and for the task. The UI permission selection may not be reaching the task runtime, or existing/queued tasks may be retaining an older permission snapshot. Workaround Explicitly persist the permission profile in ~/.codex/config.toml: approval_policy = "never" default_permissions = ":danger-full-access" Then restart Codex Desktop and create a new task. Existing tasks may retain the profile they received when they were created. This workaround intentionally removes sandbox restrictions and should only be used on trusted machines and repositories. Questions for Investigation Does the desktop Full Access selection persist correctly to config.toml? Do queued and background tasks inherit the current permission profile? Can tasks retain stale permission snapshots across an application update? Is there a compatibility issue between default_permissions and legacy Windows sandbox settings? Could the UI display each task’s effective runtime permission profile, rather than only its selected profile? Additional Context The problem started around the time of an app update, but that correlation is not yet sufficient to establish the update as the root cause. Reports can be filed through OpenAI Codex GitHub Issues or the OpenAI Developer Community. The relevant settings are documented in the Codex configuration reference. @thsottiaux @OpenAI @OpenAIDevs

• 
  Poly (@poly0015iew) reported 23 minutes ago

  AN OPERATING SYSTEM THAT WRITES ITS OWN KERNEL DRIVERS HAS 28 STARS ON GITHUB 72,000 lines of C. one contributor. seventeen commits. i checked the numbers three times because the ratio didn't make sense. it's an x86_64 os from scratch whose only interface is a sentence. no shell, no commands, no ls, no cat. the author's own line is better than anything i could write about it: there is not one strcmp on the input path. that's how a kernel developer tells you he didn't cheat. the kernel does its own DNS and its own TLS in ring 0 - lwIP and mbedTLS compiled straight into kernel.bin, no host proxy of anything. the model gets 64 real syscalls as tools. prose on screen is the model, [brackets] are the kernel printing what it actually dispatched, from the real C return value. five separate defects were fixed to keep that line honest. now the part that made me put my coffee down. the kernel ships no audio driver and no knowledge of any sound chip. you attach a card, it shows up as an unclaimed pci device, and you type one sentence. it reads pci config space. writes a bring-up program in the kernel's own driver-vm instruction set. resets the codec. unmutes the mixer. builds a buffer descriptor list. starts the dma engine. installs itself as the system audio sink. then you can just ask it for notes. and in the recorded transcripts there's a run where the model correctly diagnosed a bug in this kernel's own printf - from inside the guest. forty years of teaching humans to speak kernel. one person spent seventeen commits teaching the kernel to speak human. twenty-eight people have noticed. 72,000 lines. 64 syscalls. zero strcmp. 28 stars.

• 
  Aditya Gupta 🦀 (@sortedcord) reported 25 minutes ago

  Probably a hot take hot take but you see, I don't think people often browse code of the repo on github that often. If they do it's usually after a clone on their own machines. But still it takes the most screen real estate when you visit a repo. It IS in fact a design issue.

• 
  Borna Lang (@imnonplussed) reported 26 minutes ago

  @valigo Ah, okay, that clears things up a bit. I have mostly stopped using github some time ago in favor of just hosting my own bare repos on my own server, but if I ever feel like open sourcing anything, Codeberg will probably be the place, in that case.

• 
  Vic (🌽💬Bitcoin) (@vicariousdrama) reported 26 minutes ago

  @mutatrum I have an internal *** and use github only as a distribution mirror. All discussion happens in private as a result. In my case i dont want to run a public *** server but a business or major org really should so discussion can be reviewed.

• 
  Chris (@boredcrystalguy) reported 27 minutes ago

  @sophiiess_ This is completely unresponsive to what I said. Your example mentions two features that are separate and have different purposes. The problem is that if those features are present, they should be implemented well. If GitHub allows downloading apps, it should be easy to do.

• 
  NiroSolis (@WilliamR58743) reported 28 minutes ago

  @ReviewsPossum He's not right. He's looking on GitHub for compiled binaries. That is not what GitHub is for. It is a version control management service, and it is a community for developers, not end-users. It's like shopping in a bookstore and expecting to find a movie. The problem is that end-uses are *way* too often directed to GitHub as if it's a software hosting site. It's not. It's a version control site; for devs.

• 
  photoOrg (@photoOrg1) reported 28 minutes ago

  is @github down?