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:

• 
  0ne (@dontflex00) reported 6 minutes ago

  @x402_Omega Fix the x username on GitHub, you mentioned the wrong account there

• 
  Mike Fishbein (@mfishbein) reported 6 minutes ago

  We replaced ourself from our own client work. Built a system that turns client request into shipped code. Clients get production features built in minutes. Background: We were putting finishing touches on an internal tool. Client was messaging us with all those small requests that always come up before you ship. Then we realized something shocking: The bottleneck wasn't coding, it was US. Reading messages and actioning them was the slowest part. We had already architected the project and context engineer'd the system, so Claude Code could handle most requests. We were just a middle man at that point. So we replaced ourselves. Took ourselves out of the way. Built a Telegram bot hooked up to Github, Claude Code, and Vercel. It turns requests from clients into shipped code. Here's what it does when the client drops a message in the group chat: > The Telegram bot webhook runs as a serverless function on the same Vercel instance as the site (zero new infrastructure) > The function fires a GitHub repository_dispatch event carrying the request text > GitHub Actions spins up a runner, clones the repo, and hands the request to Claude Code > Claude Code reads the full project context from AGENTS.md, makes > Commits to main and pushes, Vercel auto-deploys on push, live and ready for the client to review > The action calls Telegram Bot API to reply in the group chat with confirmation No more "hey can you update this when you get a chance" messages sitting in our inbox for a day. No more context switching. Clients get their requests shipped faster because we built a system thats handles them instead of handling them ourselves. Some people set up a VPS and run OpenClaw or Hermes Agent so they can chat with a coding agent from Telegram. That's a pain though. We skipped all of that. There's no server, no Docker, no systemd service. The entire pipeline is serverless and runs on infrastructure that already existed. The webhook lives on the same Vercel deployment as the site itself. GitHub Actions is the only compute. Here's how to build this for yourself (copy-paste this into Claude Code): 1. Set up a Telegram bot (BotFather) and point its webhook at a `/api/telegram` serverless endpoint on your existing hosting 2. Write the endpoint: verify sender, extract message text + any photo URLs, fire `repository_dispatch` to GitHub 3. Create a GitHub Actions workflow triggered by `repository_dispatch` that runs `claude-code-action` with the request payload 4. Add `AGENTS.md` to your repo with project structure, deploy commands, and conventions (this is the agent's context) 5. Have the workflow's final step POST a confirmation message back to Telegram via Bot API

• 
  Bo Shen (@aplomb2) reported 9 minutes ago

  GitHub quietly switched to usage-based token billing on June 1st. Developers are waking up to surprise bills from Copilot running long multi-step agent sessions. This was inevitable. Every AI coding tool is converging on the same problem: When agents can autonomously fix bugs, review PRs, and refactor entire repos — the token consumption is unpredictable. Flat-rate pricing can't survive agentic workflows. Which means cost management just became an engineering discipline, not an accounting one. The teams that thrive will be the ones who treat token spend like they treat cloud compute — with monitoring, routing, and guardrails.

• 
  Adam Simone 🍃 (@AdmSimone) reported 12 minutes ago

  @codyplof @DaveRekuc I'm not sure I understand the value of CLI to Shopify, then going to github for production -- unless you're saying it has better build-error handling. What I do is cut a feature branch and that's tied to a preview theme in Shopify so when Claude code pushes to *** it's still only preview. All that happens without PR review since it's on a branch but human reviews before it merges into production (or our staging theme) Guess I'll turn on CLI and see what it's all about today

• 
  ViceSol (@ViceSol) reported 13 minutes ago

  @polydao Bro is still trying to farm GitHub handles using OpenAI Codex, a program that was officially shut down in 2023. Forinking a repository and making fake commits won't get you a $1,200 subscription, it just makes your profile look like a desperate spam bot. Stop lying for impressions.

• 
  Sipping on AI (@sipping_on_ai) reported 15 minutes ago

  GitHub Copilot has a desktop app now. The useful test: Can it take 5 stale issues, make reviewable PRs, and explain every file it touched? Try it on backlog dust first. Not the checkout flow.

• 
  Gary Bernhardt (@garybernhardt) reported 17 minutes ago

  2000: You're free to submit a patch to the mailing list! 2010: You're free to open a GitHub issue! 2020: You're free to submit a PR to the GitHub repo, but issues are closed! 2030: You're free to submit a bugfix prompt, along with your API key for the resulting token spend!

• 
  - ben - (@Benny_Jiang_) reported 19 minutes ago

  @rauchg I was seriously thinking of building this and i had a quick prototype. I didn't further spending energy on this cuz of 3 issues 1. most skills are reused: at beginning i install a lot of skills at user level, and then just keep using what's working for me. searching from public space is less of a strong need. 2. skill ranking is hard. think of google works because of page rank. i figure semantic search + github star is much more noisy. you probly need to do really expension batch eval to verify what works or have enough traffic to do ranking. therefore, vercel has a much higher chance to make it work 3. internal skill >> public skills. skill is much more value if people within the same company use it to share the tribal knowledge. still very happy Vercel did it otherwise i would always be curious how good it could be

• 
  umar ibrahim (@imp213x) reported 22 minutes ago

  @github has the worst support system I can ever think of. It’s been 48hrs now since a successful copilot pro+ repayment and I am still restricted to the free plan. To think the very backbone of every developer out there can be this poor in support and feedback system! Terrible!

• 
  🤯 (@forlayo) reported 24 minutes ago

  @github @githubstatus we are suffering some issues suddenly across different repositories and your status says all is fine. {     “error”: {         “message”: “GitHub repository access check failed — re-authorize GitHub in settings”,         “reason”: “github_repo_access_denied”,         “type”: “invalid_request_error”     },     “request_id”: “req_011CbkPBXouARPgUjCsqTaBt”,     “type”: “error” }

• 
  SAAS worker🌻🌻 (@sand_9999) reported 27 minutes ago

  @github Current runner version: '2.334.0' Runner Image Provisioner Operating System Runner Image GITHUB_TOKEN Permissions Secret source: Actions Prepare workflow directory Prepare all required actions Getting action download info Error: Repository access blocked

• 
  Xavier Pérez (@thexap_tech) reported 29 minutes ago

  The obvious fix: path-filter the workflow to ignore markdown. GitHub has paths-ignore built in for exactly this. I reach for it and walk straight into something I didn't know about required status checks.

• 
  ADITYA (@Aditiiitm) reported 30 minutes ago

  @SashaSammy15 @sangatechug Interesting. The reporting part sounds valuable, but the hard problem is trust. How does Trillien determine what's actually true when GitHub, Jira, Slack, and docs don't agree?

• 
  Robert Ta (@therobertta_) reported 30 minutes ago

  What benchmarks get RIGHT: 1. They create a common scoreboard. Without SWE-bench, we would have no way to compare Claude, GPT, and Composer on the same tasks. 2. They force rigor. A score of 79% means 79 out of 100 real GitHub issues resolved. 3. They track progress. Two years ago, the best score was under 20%. Benchmarks are useful thermometers. They are terrible doctors.

• 
  TwoSix (@Anon2Six) reported 32 minutes ago

  @goyblooddrinker I've been wanting to test the yellowkey bitlocker bypass but NightmareEclipse's github got taken down. I also dont have a spare machine to encrypt with bitlocker (as I use VeraCrypt)