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:

• 
  Prasenjit Sarkar (@stretchcloud) reported 7 minutes ago

  *** was not built for agents. The protocol assumes a human cloning a repo once a day, maybe a few times. A single agent completing a coding task can trigger dozens of clone operations. Scale that to thousands of agents running concurrently and you have an infrastructure problem that GitHub did not design for. GitHub admitted internally that agent workloads would require 30x their existing *** infrastructure scale by February 2026. Thomas Dohmke built GitHub for eleven years. He saw this coming before most people were talking about it. He left and started Entire. The company raised $60M seed at a $300M valuation in February 2026, backed by Felicis, Madrona, Basis Set, and M12. The pitch: a distributed *** network built from scratch for agent-scale clone traffic. In testing, Entire handled 570,000 clones per hour. That is not a GitHub traffic spike. That is the baseline for what an agent-first development environment actually looks like. There is a second product that gets less attention. Entire records the AI reasoning that produced each code change alongside the commit. Future agents or humans can see not just what changed, but why the model made that choice. Version control for decisions, not just files. The pattern here is straightforward. Every piece of infrastructure in the software development stack was designed for humans. Agents interact with those systems at different frequencies, different scales, different access patterns. The infrastructure needs to be rebuilt layer by layer.

• 
  Dan (@thedansho) reported 7 minutes ago

  @TFTC21 @ODELLXYZ @MartyBent Just switched to radar from Molly last night. Unfortunately there's a bug at the moment and I can't use the payments feature, so I've temporarily shifted back to Molly, but will be keeping an eye on the issue in github to migrate again! Very cool stuff.

• 
  Yoshik (@AskYoshik) reported 12 minutes ago

  15 CI/CD pipeline patterns you should understand before your next build: 1. Artifact Promotion - Build once, push one artifact, promote the same image across dev, staging, and ****. 2. Immutable Build IDs - Tag images with commit SHA or build number, not just 'latest'. 3. Pre-merge Validation - Run tests, lint, security checks, and Terraform plan before code reaches main. 4. Environment Gates - Keep production behind manual approval, SLO checks, or change window rules. 5. Fast Rollback Path - A deploy pipeline without rollback is only half a pipeline. 6. Database Migration Checks - Separate schema changes from app deploys when rollback is risky. 7. Secrets Injection - Pull secrets at runtime from Vault, AWS Secrets Manager, or sealed secrets, not ***. 8. Cache Discipline - Cache dependencies, but include lockfile hash so old packages do not silently survive. 9. Matrix Builds - Test across versions like Node 20/22, Python 3.11/3.12, or multiple OS images. 10. Ephemeral Preview Environments - Spin up short-lived stacks for PRs, then destroy them cleanly. 11. Deployment Health Checks - Wait for readiness probes, 5xx rate, latency, and error logs before calling it done. 12. OIDC for Cloud Auth - Avoid long-lived cloud keys inside CI variables when GitHub/GitLab OIDC works. 13. Policy Checks - Block public S3 buckets, open security groups, and untagged expensive resources before apply. 14. Pipeline Time Budgets - If CI takes 45 minutes, people start bypassing it. 15. Audit Trail - Know who deployed what commit, from which runner, to which environment, at what time.

• 
  hustler one (@hustlerone4) reported 14 minutes ago

  omp's issue:// defaulting to github is driving me insane, and you can't seem to disable it

• 
  Nosyt Labs (@NosytLabs) reported 16 minutes ago

  @gonka_ai Is community brokers closed if we issue a github issue? Or only way is to do the public broker? Thanks @gonka_ai

• 
  Spencer Baggins (@bigaiguy) reported 22 minutes ago

  A self-taught developer from Brazil just cracked the context window problem that's been plaguing RAG systems for 2 years. No PhD. No research lab affiliation. Just 400 GitHub commits and a personal obsession. Here are the 8 techniques from his open-source library that every RAG tutorial gets completely wrong:

• 
  Julian Goldie SEO (@JulianGoldieSEO) reported 26 minutes ago

  AI Studio Update: Google just fixed the one-way door in AI Studio. Old code was stuck outside. Now you can bring it home. The problem before: You could push projects OUT to GitHub. You couldn't bring them back IN. Old project? Rebuild from scratch or copy files by hand. Now it's one button: Import from GitHub. What that unlocks: → That dead project from 6 months ago? Import it. Ask Gemini to fix it up. → Build in Cursor or Claude, polish in AI Studio, push back out. The walls between tools are falling. → Teammate left? Anyone can pick up their code using plain English. And if you can't code at all: Someone built your website. It sits in a repo. You can now just say "change the colors" or "fix it on phones." Here's the move today: Find one old project you gave up on. Import it. Ask AI what it would improve. "I'd have to rebuild it" is no longer an excuse.

• 
  bryson (@brysontang) reported 29 minutes ago

  having gpt-5.6 and fable communicate over github issue comments

• 
  Mikhail Rogov (@i_mika_el) reported 29 minutes ago

  @abhimeeofficial real GitHub issues plus code quality checks should expose agents that only learn to game test suites.

• 
  Koder (@koder0x) reported 30 minutes ago

  A follow-up to something I posted recently: a set of Claude Code subagents I built and refined, and actually use daily, both at work and across side projects. Most of the value isn't any single agent. It's their interaction. Here's the loop I've been running lately, at work against real DevOps user stories, and it holds up almost unchanged on side projects too, swapping the work item for a plan created beforehand. "Understand user story NNNN from DevOps project XYZ and create a multi-step plan" "Fan out to the most appropriate agent for each step, normally task-builder, test-builder, or change-executor, and proceed with plan implementation, tracking progress in a TODO list" "Use complexity-pruner to identify gaps, issues, and bugs in the latest changes, ignoring secondary advice and warnings, then fan out to code-fixer for each finding" Then I do something that turned out to be the most important part of the whole loop. I reset the session. "Understand user story NNNN from DevOps project XYZ, that's the truth. Use fact-checker to compare it against the changed files" The reset is what makes this work. An agent that watched itself write the code tends to justify its own decisions when asked to check them. An agent that only sees the intended outcome and the actual diff has nothing of its own to defend, it's comparing two artifacts, not reviewing its own reasoning. That asymmetry is the whole point of splitting this across agents instead of asking one long-lived session to plan, build, and verify itself. Verification only means something when it comes from somewhere the implementation couldn't reach. Repository on GitHub: gsscoder | claude-coding-agents

• 
  Ryan Moore (@0xElGato) reported 31 minutes ago

  Anyone else having problems with Grok 4.5 lieing and making claims that are clearly untrue? For example, I asked it to check data on an MCP (GitHub) and it routinely claims it reads the data but it absolutely does not? Later it admits it lied or guessed. Not acceptable.

• 
  Techjunkie Aman (@Techjunkie_Aman) reported 33 minutes ago

  Microsoft spent years adding more to Windows. One developer spent years taking it back out. Every fresh Windows install meant repeating the same routine: uninstall bloatware, disable telemetry, tweak privacy settings, install apps, and undo Microsoft's defaults. Chris Titus Tech got tired of doing it manually. So he turned his personal PowerShell scripts into WinUtil. What started as a private toolkit became one of GitHub's biggest Windows projects, trusted by millions of users worldwide. Today, WinUtil can: • Install apps with Winget • Debloat Windows in minutes • Reduce telemetry • Improve gaming and system performance • Control Windows Update • Restore classic Windows behavior • Create restore points automatically • Build custom Windows ISOs With 57K+ GitHub stars, hundreds of contributors, and tens of millions of launches, WinUtil has become the first thing many enthusiasts run after installing Windows. The best utilities aren't created to make money. They're created because someone got tired of solving the same problem every single day.

• 
  Koder (@koder0x) reported 33 minutes ago

  A follow-up to something I posted recently: a set of Claude Code subagents I built and refined, and actually use daily, both at work and across side projects. Most of the value isn't any single agent. It's their interaction. Here's the loop I've been running lately, at work against real DevOps user stories, and it holds up almost unchanged on side projects too, swapping the work item for a plan created beforehand. "Understand user story NNNN from DevOps project XYZ and create a multi-step plan" "Fan out to the most appropriate agent for each step, normally task-builder, test-builder, or change-executor, and proceed with plan implementation, tracking progress in a TODO list" "Use complexity-pruner to identify gaps, issues, and bugs in the latest changes, ignoring secondary advice and warnings, then fan out to code-fixer for each finding" Then I do something that turned out to be the most important part of the whole loop. I reset the session. "Understand user story NNNN from DevOps project XYZ, that's the truth. Use fact-checker to compare it against the changed files" The reset is what makes this work. An agent that watched itself write the code tends to justify its own decisions when asked to check them. An agent that only sees the intended outcome and the actual diff has nothing of its own to defend, it's comparing two artifacts, not reviewing its own reasoning. That asymmetry is the whole point of splitting this across agents instead of asking one long-lived session to plan, build, and verify itself. Verification only means something when it comes from somewhere the implementation couldn't reach. Repository on GitHub: gsscoder | claude-coding-agents

• 
  Welldone (@welldone_tech) reported 36 minutes ago

  🔥 Two recent findings, one lesson. GuardFall showed that 10 of the 11 most popular open-source AI coding agents can be hijacked with shell tricks documented decades ago. And a flaw in Claude Code's GitHub Action let a single malicious issue poison any repo that used it.

• 
  H I K A R U (@Terry3nty) reported 38 minutes ago

  Now imagine an AI agent. Today it needs GitHub. Tomorrow it needs Gmail. Then PostgreSQL. Then Docker. Then your local files. Then AWS. Then Notion. Then a browser. Unlike traditional software, an AI agent isn’t built for one workflow. It’s expected to perform many different tasks across many different systems. That’s where the problem starts. Every tool speaks differently. Every API has different rules. The AI doesn’t just need access to tools… It needs a consistent way to understand and use them.