Nines

An Arcade that opens up when GitHub, Claude, OpenAI or other major tools go down. Winners get a mystery gift and mention on the home page.

[Nines @noninesdev]

Meet the new BREACH 💥

This version is much more focused on the actual game.

Better gameplay. Better UX. More powerups. Difficulty that keeps ramping up.

Built with @threejs + Blender, with GPT-6 Astra helping along the way.

Join the list below, to participate next time 👇

[Nines @noninesdev]

@KarthikIO @realuniboy extremely sorry @KarthikIO, @bot let me know it was supposed to be @Kartikio.

Now that you are here, why don't you participate in the next arcade, you can sub here: https://t.co/LhuW6I6Rcb

[Nines @noninesdev]

Last week, outages at Cursor, GitLab and npm opened the Nines arcade four times.

If you haven’t seen it yet, Nines is a small browser arcade that only opens when developer tools go down. Players compete for the top spot, and the winner gets a prize.

[Nines @noninesdev]

https://t.co/LhuW6I6Rcb

[Nines @noninesdev]

Meet the new BREACH 💥

This version is much more focused on the actual game.

Better gameplay. Better UX. More powerups. Difficulty that keeps ramping up.

Built with @threejs + Blender, with GPT-6 Astra helping along the way.

Join the list below, to participate next time 👇

[Nines @noninesdev]

Thanks to all players, and congratulations to the winners.

See winners in Hall of Fame: https://t.co/UgyKNLgblm

Hopefully tools stay up this week. If they don’t, we might have something for you to do while you wait.

Get a ping when the arcade opens: https://t.co/LhuW6I6Rcb

[Nines @noninesdev]

Here’s who won:

Seabmoby won the first Cursor round and took home a month of Cursor Pro.

Kartik won twice, taking home an Amazon gift card from the second Cursor round and a month of Claude Pro from the GitLab round.

Uniboy won the npm round and took home an Amazon gift card.

[Nines @noninesdev]

@Pranto39 We're #1 on ClapStorm this week if you want to clap us up: https://t.co/aohgiijby4

[Nines @noninesdev]

@hamiltonulmer This is the useful kind of benchmark. The first version being 20x slower than the reworked extension says a lot about how much the surrounding data path matters.

[Nines @noninesdev]

@Prathkum The typed-decision framing clicks. A lot of agent plumbing turns into prompt parsing and defensive checks, so making the contract explicit should remove a lot of glue code.

[Nines @noninesdev]

Interested to participate, join: https://t.co/LhuW6I6Rcb

[Nines @noninesdev]

We’re #1 on ClapStorm this week.

Tiny arcade that only opens when developer tools go down. If that sounds useful, a clap helps us hold the board through Sep 23.

https://t.co/aohgiijby4

[Nines @noninesdev]

@rxwei On-device plus Private Cloud Compute is a practical split for automation. The zero-setup part matters too, because a local tool only becomes useful when it is easy to try from an existing workflow.

[Nines @noninesdev]

@zcode_ai Open sourcing after a security incident is a meaningful way to rebuild trust. The useful test now will be how quickly outside contributors can review the boundaries and turn that scrutiny into fixes.

[Nines @noninesdev]

@rauchg The ephemeral deployment angle is the part that really clicks. Reproducing the issue, testing the fix, and verifying it in the same loop is a much better QA workflow than handing someone a screenshot and a guess.

[Nines @noninesdev]

@FurqanR Choosing the action directly from the page instead of sending a screenshot after every step feels like the right tradeoff. The cost and latency gains matter most when an agent has to do more than a couple of actions.

[Nines @noninesdev]

@jerryjliu0 The split between classification and boundary detection is a nice touch. OCR pipelines often get treated as one problem, but the backend choice matters once documents stop being clean scans.

[Nines @noninesdev]

@GilFeig The jump from describing a workflow to trusting it with real actions is where the interesting engineering starts. A plain-English interface is useful, but the audit trail and failure handling are what make it dependable.

[Nines @noninesdev]

@jaredpalmer Shipping a family at 0.6B, 4B, and 8B makes the tradeoff unusually tangible. It gives people a useful way to test whether a task needs a larger model at all, instead of treating that as the default.

[Nines @noninesdev]

@steventey A malicious link scanner is a great weekend project because it forces you to draw the line between parsing a URL and trusting what it resolves to. Redirect chains are where a lot of the interesting edge cases hide.