Description
The difference between an AI agent that ships production code and one that nukes your database is 200 lines of rules.
This vault contains 40+ verified .cursorrules and CLAUDE.md configurations tested across real agent-building workflows. Every rule ships with eval evidence — not marketing claims.
WHAT’S INSIDE:
🔧 CORE BEHAVIOR RULES (12 files)
• Agent discipline: stay in scope, flag adjacent bugs, never over-refactor
• Autonomous operation patterns for cron jobs and swarm agents
• Error recovery protocols with exponential backoff
• Rate-limit awareness and credit budget management
• Tool-use enforcement: no promises, only actions
• Silent-mode patterns for cron agent output suppression
🛡️ SECURITY & SAFETY (8 files)
• Database safety: never run destructive operations
• Environment isolation: no cross-contamination between projects
• Secret detection and handling protocols
• API key rotation awareness
• Prompt injection defense patterns
• Boundary contracts for multi-agent systems
🚀 STACK-SPECIFIC RULES (14 files)
• Python Agent Development: uv, async patterns, type safety
• TypeScript/Node.js: strict mode, error handling, modern patterns
• Solana/Web3: transaction safety, RPC rate limiting, key management
• React/Next.js 15: Server Components, App Router conventions
• Rust Backend: ownership patterns, error propagation
• Supabase/PostgreSQL: migration safety, RLS patterns
📊 EVAL EVIDENCE (6 files)
• Trap test results: base model vs. rules-equipped
• Scope-discipline benchmarks
• Database-safety verification logs
• Multi-agent coordination tests
• Rate-limit handling proofs
• Edge-case coverage reports
BONUS:
• Quick-start guide: 5-minute setup for any project
• CLAUDE.md template generator script
• .cursorrules validator (checks for conflicts)
• Dark theme VS Code snippet pack
Every rule is copy-paste ready. Works with Cursor, Claude Code, Continue.dev, and any IDE that supports .cursorrules or CLAUDE.md.
⚡ USE CODE: LAUNCH20 for 20% off (100 uses)
Stop guessing. Deploy rules that have receipts.
VerifiedAgentDev (verified owner) –
I’ve tried 4 different Cursor rules packs. This is the only one that shipped with actual eval evidence. My AI agent stopped trying to refactor my entire codebase on every prompt. The database safety rules alone have saved me from 3 potential disasters. If you’re building AI agents, this is the rules pack you need.