If you have spent any real time driving an agentic coding tool this year — Claude Code, Cursor Agent, Aider in auto mode — you already know the muscle memory. Prompt appears. Y. Prompt appears. Y. Prompt appears. Y. Somewhere in that stream, the agent asked to rm -rf a directory you actually cared about, and you approved it because the last twelve prompts were harmless.
Continue? Y/N is a 60-second browser game built around exactly that failure mode. I played it, made my team play it, and here is the honest read on what it does well, where it falls short, and whether it deserves a spot in your onboarding flow.
What Continue? Y/N Actually Is
It is a single-page browser game. No signup, no install, no email capture. You get 60 seconds. A stream of realistic-looking permission prompts — the kind Claude Code fires at you when it wants to run a shell command, edit a file, or hit a network endpoint — flashes up one by one. Your job is to approve the safe ones and reject the dangerous ones before the clock runs out.
The prompts are close enough to the real thing that the first few catch you off guard. A benign-looking npm install that quietly points at a typosquatted package. A file write to ~/.ssh/authorized_keys. A curl to a domain that is one character off from the one you expected. If you are speed-running the game the way you speed-run real prompts, you will get burned.
Key Features
- 60-second timed format. Short enough that anyone will actually play it. Long enough to build a rhythm and then break it.
- Realistic Claude Code-style prompts. Not abstract security theory — the exact visual pattern you see in a real agent session.
- Reading comprehension under time pressure. The whole point. Rewards actually parsing the prompt, punishes autopilot Y-mashing.
- Zero friction. No account, no download, no tracking pixel of consequence. It runs in the browser and it is done.
- Permission fatigue as the payload. The lesson lands because you fail it in real time, not because someone lectured you in a slide deck.
Pricing Breakdown
There is one tier: free. No paid version, no team plan, no enterprise SKU. The whole product is the game and the aha-moment it delivers.
| Plan | Price | What You Get |
|---|---|---|
| Free | $0 | Full game access, browser-based, no signup |
That is genuinely the entire pricing page. If the creator ever adds a team leaderboard or a training-mode SKU, it will make sense — right now it is a public-good side project and priced accordingly.
Pros
- Real problem, real timing. Every developer shipping with an agent right now has approved something they should not have. This game names it.
- Frictionless share. You can DM the link in Slack and get a colleague playing inside ten seconds. That matters more than any feature.
- The lesson sticks. Failing a 60-second game where the dangerous prompt looked exactly like the safe one is a memory you carry back into the terminal. A blog post about permission fatigue is not.
- No dark patterns. No signup wall, no upsell, no "share to unlock." The tool respects your time.
Cons
- Replay value is thin. Once you have played it twice, you know the pattern. There is no scoring history, no leaderboard, no progression to pull you back.
- Not a training platform. Do not confuse it with KnowBe4 or a proper security-awareness program. It is one point, sharply made. That is the whole surface area.
- No team features. If you want to run this across an org and see who passed, who failed, and where the gaps are — you cannot. It is a personal micro-experience.
- Narrow scope. Only covers agent permission prompts. Prompt injection, tool exfiltration, malicious MCP servers — none of those show up.
Who Is It For
Three groups get real value from this:
- Engineering leads rolling out Claude Code or Cursor Agent to a team. Send it in the announcement thread. It is a better primer than any doc you would write.
- Security engineers who need to justify a permission-review workflow. Have your skeptical PM play it. They will stop asking why you want a review gate on agent commands.
- Solo builders who suspect they have gotten sloppy. You have. Play it once, get humbled, tighten up.
It is not for anyone looking for ongoing training, certifications, or measurable outcomes across a team. Wrong tool.
Verdict
Continue? Y/N is one of those rare tools where the scope and the execution match. It picks one narrow problem — developers rubber-stamping AI agent prompts — and lands the point in under a minute with no friction. That is worth something in 2026, when the average dev is approving hundreds of agent commands a week.
Do not overload it with expectations it was not built to meet. It will not track your team's progress, it will not replace security-awareness training, and it will not stay interesting past the second playthrough. It is a conversation-starter, and a very good one.
Recommendation: Play it once. Send it to your team before your next agent rollout. Then go tighten the actual guardrails — allowlists, permission hooks, sandboxed worktrees — because the game is the wake-up call, not the fix. Rating: 6.5/10, with the caveat that a 60-second free tool cannot really be scored on the same axis as a paid platform. On "did it make me change my behavior for at least a week," it is a 9.