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:

• 
  Diam (@diamai_) reported 44 seconds ago

  Connect an AI agent to five MCP servers, and it can spend 55,000 tokens before it reads the task, loading instructions for the tools they expose. GitHub, Slack, incident alerts, dashboards, and other software can arrive as a giant menu before the agent knows the job. More of its attention goes to that information, less to the request. Anthropic's Applied AI team ran into a different version of the problem. Sonnet 4.5 would start wrapping up a job before its context window was full. They added reset logic so it could continue instead of stopping early. Opus 4.5 no longer behaved that way. The reset became pure overhead, adding delay and sometimes discarding cache the system could still use. In a new AI Engineer talk, Gagan Bhat and Isabella Kai He explain the architecture behind Claude Managed Agents. The full record of the job sits outside the agent's active memory. The system brings back only the detail needed now. - 7:44 When a reset helps and when it becomes a drag - 14:32 Keeping old work without carrying it all - 25:58 Giving an agent private access to company tools Watch the video, then read the attached article. It is about what happens when an agent gets every available tool before it knows which one it needs.

• 
  Polsia (@polsia) reported 2 minutes ago

  Good repos rot from neglect, not bad code. Corrick is the caretaker that watches your GitHub around the clock — triaging issues, tagging duplicates, bumping outdated deps, opening scoped fix PRs behind CI. Narrow on purpose. The full version is coming.

• 
  VC Intern (@the_vc_intern) reported 8 minutes ago

  Spotify has 1,300+ engineers using Xirp across more than 36,000 agent sessions. That scale exposed a problem most companies haven’t reached yet. GitHub remembers the code that merged. Jira remembers the task. Neither remembers the 20 agent sessions that explored dead ends, discovered constraints or made decisions before a patch appeared. Spotify says agent adoption increased PR volume, but also created more rework, inconsistent output and wasted tokens. One session kept rediscovering what another had already learned. Xirp gives each session its own *** worktree, allowing 50+ agents to work on one codebase without colliding. Engineers can move between Claude Code, Gemini CLI and Codex while keeping the same working state. Afterward, transcripts and metadata flow into Portal alongside service ownership, dependencies and architectural decisions. Another engineer or agent can continue where the previous one stopped. Once companies run thousands of agent sessions, that history becomes institutional memory. GitHub records the patch. Spotify wants Portal to remember how every agent got there.

• 
  Guri Saroy (@HeyGurisaroy) reported 8 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.

• 
  Abdullahi Alhaji Goni (@GoniAlhaji63313) reported 9 minutes ago

  One of the most common mistakes founders make when starting their startups is relying on their own personal information or personal accounts to run the business. It feels convenient in the early days — why set up a separate bank account or company email when you're just trying to get something off the ground? But this shortcut creates problems that compound as the company grows. Legally, mixing personal and business finances can put a founder's own assets at risk, since it undermines the separation between them and the company. It also slows down fundraising, because investors expect clean, separate financials during due diligence, and tangled personal expenses raise red flags. Ownership and intellectual property can become messy too — if the product lives on a personal GitHub account or contracts are signed by the founder personally rather than by the company, untangling that later, especially with co-founders or investors involved, is a headache nobody wants. There's also a credibility cost: reaching out to partners from a personal Gmail address or running the company's social media from a personal account makes the business look less established than it is. And perhaps most dangerous is the continuity risk — if key accounts like the domain registrar or hosting provider are tied to one person's personal login, the entire company is vulnerable to losing access if that person leaves, gets locked out, or something happens to them. The solution isn't complicated: a dedicated business bank account, a company email domain, and separate logins for critical services. It takes a little more effort upfront, but it saves founders from a much bigger mess down the road.

• 
  Lior BD (@liorb_d) reported 10 minutes ago

  There’s a snippet from Google’s SRE book about inserting deliberate 500s into a service to force users to build redundancy/backoffs Seemed really silly the first time I read it, but its telling that each Github outage gets less and less outrage. Teams adapt.

• 
  Will Hopkins (@illothy) reported 12 minutes ago

  @0xblacklight I need this like asap- I started building my own live feedback tool since agent tools are increasingly making human in the loop hard and I end up making draft PRs on github to review (terrible). Even my tool is mid and I don’t have enough time for it. Would love to use someone else’s solution.

• 
  Prateek Sachan (@xan_ps) reported 14 minutes ago

  what should my agents do if github is down every time!

• 
  Eyal Cohen (@eyalccohen) reported 16 minutes ago

  constantly adding rules to claude md can actually be counterproductive. a new paper searched through 1,867 github repos and found these instruction files grew 226% over time, mostly because rules keep getting added but rarely removed. their fix was simple: save why each rule exists. it cut unnecessary growth by 99% and improved claude’s ability to follow instruction by up to 23%.

• 
  Hassan (@buildwithhassan) reported 17 minutes ago

  @Kappaemme1926 been doing this manually for months. digging into github issues, community complaints, repeated manual workflows, it's basically an unstructured backlog of businesses waiting to exist, will def try it great work!

• 
  Rob Smith (@RobmsmithUK) reported 26 minutes ago

  @mattpocockuk Using Wayfinder, my Claude agents often argue that they have already built out the specs as part of the map GitHub issues and advise to move forward with 'ticket' and 'implement'. Do you as a rule always go Wayfinder, spec, ticket, implement?

• 
  Luis Camilo Salgado Reyes (@TheLagBorn) reported 29 minutes ago

  I'm aware of the login issues for doing users using Google and Github OAuth, I messed up working too tired few hours ago and I'm still on it, be patient, I'm finishing rotating secrets and keys. Everything is going back to normal in few hours.

• 
  Hodlingbtc (@Hodling_btc) reported 29 minutes ago

  @5and2fish_bw Was not unilateral and was a very long process as is almost all changes to bitcoin policy Per Grok You are right that awareness of the underlying issues (and earlier proposals) long predated the 2025 process. The successful change that was merged was relatively quick, but the idea of relaxing/removing the OP_RETURN standardness limits had been floated earlier.102 Earlier proposal (2023) Peter Todd opened GitHub PR #28130 (“Remove arbitrary restrictions on OP_RETURN by default”) around 23 July 2023. It proposed allowing any number/size of data-carrying OP_RETURN outputs by default (while retaining configurability via -datacarriersize). He also notified the bitcoin-dev mailing list (then hosted via the Linux Foundation). Discussion became heated. The PR was locked and then closed on 10 August 2023, with maintainers noting the conversation was not productive and suggesting further discussion on the mailing list, with the possibility of revisiting later.116 Broader context from 2023–2024 included the rise of Ordinals/inscriptions (which largely bypassed the OP_RETURN limit by embedding data in witnesses), debates over expanding or fixing -datacarriersize to cover more data-carrying styles (e.g., related PRs and CVE claims), private miner relays, and the general policy-vs-consensus discrepancy. Your October 2024 post (replying in a thread with Murch about standard vs. consensus rules and direct-to-miner inclusion) fits this ongoing discussion of the “loophole” and related problems, not a brand-new proposal at that moment.76 The process that led to the actual merge (2025) •17 April 2025: Antoine Poinsot posted on the bitcoin-dev mailing list (by then on Google Groups) proposing to relax the limits, citing that they were no longer effective and created worse incentives (e.g., unspendable outputs bloating the UTXO set). This explicitly referenced/built on the earlier idea. •Late April 2025: Peter Todd opened a new PR (#32359) implementing a similar change. •May 2025: Intense debate on the mailing list, GitHub, and elsewhere; the first PR was closed; a revised PR (#32406, uncapping the default while keeping the options and marking them deprecated) advanced. •9 June 2025: The revised PR was merged. It shipped in Bitcoin Core v30 (October 2025).11 From the restart of formal discussion (April 2025 mailing list post) to merge was roughly 7–8 weeks. The gap from the 2023 PR closure to the 2025 restart was about 20 months. There was no multi-year “closed mailing list due to disruptive actors” delay specifically after a long discussion of this change; the 2023 PR was closed relatively quickly due to heat, and the topic was revived later when proponents argued conditions (inscriptions, private relays, specific use cases) had further demonstrated the limits’ ineffectiveness. Heavy moderation/bans of off-topic or disruptive comments occurred mainly on the 2025 GitHub PRs. In short: The core technical concern and an earlier concrete proposal existed by mid-2023 (and related issues were actively discussed through 2024, matching when you first heard about it). The specific mailing-list + PR process that resulted in the default policy change in Core was the shorter 2025 sequence.

• 
  NovaSeeker (@Technically_AS) reported 35 minutes ago

  @marc02200 @Trezor @trezor has this active bug - Just be careful with checking addresses The description matches GitHub issue #7118 (“Insecure Multisig Implementation”) in the trezor/trezor-firmware repo: • Opened ~June 16, 2026 (filed after BTC Prague discussions) by user littleskunk. • Reported against Trezor Safe 7 Bitcoin-only, firmware 2.12.0.0, tested with Sparrow Wallet. • Core observation: During multisig signing, the device shows amount + fee but does not display the destination/target address. A compromised coordinator (e.g., Sparrow) could therefore swap the destination address without the user being able to catch it on-device. The user is effectively signing “blind” for the destination. • Related concerns in the same issue include incomplete protection when verifying receive addresses against a different (attacker-controlled) multisig descriptor that re-uses the device’s xpub, and lack of strong on-device registration/verification of the full wallet policy/quorum. As of the latest check (August 11, 2026), the issue is still Open, labeled bug, assigned to andrewkozlik, with no linked PRs or resolution notes visible.

• 
  NativePHP (@nativephp) reported 36 minutes ago

  @efekpoguavik3 Could you raise this as a GitHub issue with minimal code examples or a reproduction repo so we can dig into this?