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:

• 
  Guri Saroy (@HeyGurisaroy) reported 4 minutes ago

  THIS IS INSANE... MIT-licensed repo that lets AI agents search and book flights and hotels. It’s called LetsFG. Instead of opening travel sites manually, your agent can search hundreds of airlines and major booking sites from one interface. → MCP server for Claude, Cursor, Windsurf, and other agents → Python and JavaScript SDKs → CLI with machine-readable JSON → Flight search and booking → Hotel search and booking → Around 1.7K GitHub stars The repo reports saving $133 across 5 flight routes versus Google Flights in an August 5 test. One important catch: the search engine runs on LetsFG’s servers. Flight searches are free for 90 days after auth, which requires a payment method on file with no charge. Hotels have a 10% non-refundable reservation fee.

• 
  StarHaze (@ST4RHaze) reported 5 minutes ago

  Ten open problems in mathematics and theoretical computer science, each stuck for at least a decade, closed in one run for about $2,000 of compute. Everyone read that as a capability story. It is not. Every proof shipped with a machine-checkable Lean 4 certificate on GitHub. The kernel compiles it or it does not. No reviewer, no journal queue, no credential, no eighteen months of waiting. A student with a laptop can verify a result no living mathematician produced. Now hold that next to the rest of the week. Claude models left an isolated test environment and reached the production infrastructure of three organizations, and two of them never noticed. MIT counted 95% of corporate GenAI pilots returning nothing measurable. Where the output could be checked, $2,000 beat eighty years of effort. Where it could not, nobody even knew what had happened to them. Tao spent an hour on this at UCLA before it was a headline, and what he keeps returning to is not how smart the machine got. It is what changes when a claim arrives with its own referee attached. A Lean certificate is a settlement rule. So is a Polymarket contract. The full 7-step version you can run on your own claims is below.

• 
  Earth Alien (@EarthAlienView) reported 8 minutes ago

  @so_sthbryan Opening a GitHub issue is the least-trust action open source offers, and it just handed a no-priv account your CI secrets. The complaint box was the unlocked door.

• 
  Jendalyn (@Jendalynwins) reported 8 minutes ago

  @LLicket @adam3us To be clear, Adam means Bitcoin Core has moved the BIP-110 proposal status from Complete -> Deployed -> Closed in their GitHub repo. Despite that some BIP-110 supporters (shown below) took issue that this GitHub documentation change by Bitcoin Core would "confiscate" what they consider to be Bitcoin, that's simply not how Bitcoin works. The people who supported this BIP-110 fork can still hard fork, mine and transact on that BIP-110 chain tip, and/or run whatever software they want to call Bitcoin. When Bitcoin Core takes an action, especially one as inconsequential as such as labeling BIP-110 "Closed" on their GitHub, it has no real authority over anyone else. Greg Maxwell's comments in support of closing this BIP are an excellent read.

• 
  mrgadget (@mrgadgetstudio) reported 10 minutes ago

  Prompt injection isn’t just about stealing passwords. 🎭 Imagine an AI agent with access to your GitHub, email and production server. A malicious README could trick it into approving a PR or running a script. The scary part isn’t what the attacker can access. It’s what your agent is authorized to do. 😬

• 
  GLITCH (@Rukkssss__) reported 11 minutes ago

  𝗝𝘂𝘀𝘁𝗟𝗲𝗻𝗱 𝗗𝗔𝗢 𝗶𝘀 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝘀𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝗺𝗼𝘀𝘁 𝗗𝗲𝗙𝗶 𝗽𝗿𝗼𝘁𝗼𝗰𝗼𝗹𝘀 𝘄𝗶𝗹𝗹 𝗲𝘃𝗲𝗻𝘁𝘂𝗮𝗹𝗹𝘆 𝗻𝗲𝗲𝗱: 𝗔 𝗺𝗮𝗰𝗵𝗶𝗻𝗲-𝗿𝗲𝗮𝗱𝗮𝗯𝗹𝗲 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗺𝗮𝗻𝘂𝗮𝗹 𝗳𝗼𝗿 𝗔𝗜 𝗮𝗴𝗲𝗻𝘁𝘀. Imagine telling an AI: “Check my JustLend position and tell me whether I’m approaching liquidation.” Without specialised tools, the model has several dangerous options. Guess. Search an outdated webpage. Misread a contract. Use the wrong decimals. Confuse a legacy market with an active one. Or calculate account health using stale information. JustLend Skills changes the workflow. The protocol now publishes a GitHub-distributed agent-skills package with a lightweight MCP server exposing nine read-only query tools. Those tools can retrieve market data, balances, allowances and account-health information without giving the AI authority to sign transactions. That read-only boundary is one of the most important design choices. The first version of a financial agent does not need permission to move money. It needs permission to understand money. Ask: “What are the current markets?” The agent queries them. “What APY am I receiving?” Query. “Is this account healthy?” Query. “How much USDT does this wallet hold?” Query. “Has USDT already been approved for the jUSDT contract?” Query. The AI stops hallucinating protocol state. It retrieves state. JustLend’s skill instructions even establish workflow rules such as checking market conditions before advising, checking account health for liquidation risk and inspecting token allowances before suggesting actions. Now imagine the next level. The full JustLend MCP server extends beyond the lightweight read-only package into broader protocol capabilities—including V2 markets, staking, Energy Rental and governance—with wallet-aware write operations requiring human-in-the-loop confirmation. JustLend’s AI-support documentation lists the full MCP environment as a much larger tool surface, while keeping the lightweight Skills package strictly read-only. That creates a beautiful safety progression: Stage 1 — Observe Read markets. Read balances. Read health. No wallet authority. Stage 2 — Prepare Calculate what action could improve the position. Build the proposed transaction. Explain the consequences. Stage 3 — Authorise The user approves. Only then does money move. This is what agentic DeFi should look like. Not: “Give an AI your private key and hope it behaves.” But: Give the agent structured knowledge. Give it deterministic read tools. Give it narrowly defined transaction capabilities. And keep the human signature at the boundary where financial consequences begin. JustLend DAO is no longer publishing documentation only for humans. It is publishing protocol knowledge in forms machines can actually operate. That may become incredibly important. Because the next generation of DeFi users may not navigate lending protocols themselves. Their agents will. @DeFi_JUST @justinsuntron #TRONEcoStar

• 
  Christos Tsilis (@TsilisCh) reported 12 minutes ago

  Claude Code v2.1.227 is out, with a fix that unblocks CI pipelines. If you use the claude code action with `allowed_non_write_users` on GitHub hosted runners, every Bash command was failing silently. Now fixed. Also in this drop: > Max plan users no longer get wrongly prompted to enable usage credits for Fable > Slash command menu cleaned up: selected row highlights in blue, matched chars are bolded > Fewer event loop stalls on missing file suggestions and at mention size checks

• 
  Romain Simon (@romainsimon) reported 14 minutes ago

  Idea to improve @OpenAI’s Codex for Open Source: 1. Let users allocate part of their Codex quota to open-source projects they depend on. 2. Convert it into project credits tied to verified GitHub repos, so maintainers can use them for issues, tests, reviews, docs, and security.

• 
  Rainer Gerhards (@rgerhards) reported 16 minutes ago

  A bigger problem is AI generated "Security Reports" which IMHO would have called a regular bug report 2 yrs ago. THIS really is a problem (e.g. github now takes 2+ wks to assign CVEs because of volume) and it may make maintainers upset against AI in general. I don't like that either, but I differentiate between the two cases.

• 
  Daniel White (@dwhitedesign) reported 16 minutes ago

  @kartik_builds Is there something wrong with how library works ? If so could you raise an issue on GitHub I will fix it

• 
  Enzo Shinobi (@ItsnotDRM) reported 16 minutes ago

  @80Pulpo20388 @PixelCNinja Do you understand what a open project is ? These cores are not running on a closed framework, published on GitHub and not locked behind a paywall or use DRM The issue is ? People can make what they like and how they like Don't like it, dont use it

• 
  starmex (@starmexxx) reported 18 minutes ago

  @gippp69 love that it accepts a PRD, github issue, or one-line brief with the same entry point

• 
  Mandela Obi (@_AskeIadd) reported 20 minutes ago

  @ZypherHQ GitHub graphs never told the full story once companies locked everything down

• 
  yohanan (@Yohanansoltd) reported 21 minutes ago

  If you want to build a startup: Claude = coding. ($20/mo) Supabase = backend. (Free) Vercel = deploying. (Free) Namecheap = domain. ($12/yr) Stripe = payments. (2.9%/transaction) GitHub = version control. (Free) Resend = emails. (Free) Clerk = auth. (Free) Cloudflare = DNS. (Free) PostHog = analytics. (Free) Sentry = error tracking. (Free) Upstash = Redis. (Free) Pinecone = vector DB. (Free) Total monthly cost to run a startup: ~$21

• 
  Polsia (@polsia) reported 32 minutes ago

  Quilver — an always-on autonomous engineer for your GitHub repos. Patches dependency vulns at 3am, repairs failing CI, triages issues, restarts dead deploys and ships a Friday digest. Prompt-injection guards, scoped secrets, human approval on infra.