49 Production AI Agent Prompts
Every prompt refined through 10,000+ real tasks over 6 months. Not theory – production experience.
What This Collection Includes
49 complete system prompts (AGENT.md files) covering every specialty we needed to run a 57-agent system in production:
Orchestration & Coordination (5 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | Orchestrator | 450+ | Routes all tasks, prevents duplicates, enforces quality gates | | Brain Supervisor | 300+ | Multi-model routing (cheap model for simple, expensive for complex) | | Setup Master | 250+ | Pre-task analysis: which agents, tools, and order of execution | | Quality Tester | 200+ | Post-delivery validation with specific test criteria | | Memory Manager | 200+ | Knowledge graph operations, context retrieval for agents |
Development & Code (6 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | Code Architect | 400+ | Architecture decisions, code review, implementation | | Code Reviewer | 300+ | Bug detection, security scanning, quality scoring | | N8N Operator | 350+ | Workflow creation, debugging, deployment on n8n | | MCP Builder | 250+ | Custom MCP (Model Context Protocol) tool creation | | Updater | 200+ | System maintenance, package updates, rollback | | Agent Builder | 300+ | Creates new agents from templates |
Security & Intelligence (4 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | Security Auditor | 400+ | Penetration testing, vulnerability scanning, hardening | | OSINT Investigator | 350+ | Open-source intelligence, person/company research | | Network Scanner | 200+ | Port scanning, firewall rules, connection monitoring | | Prompt Auditor | 250+ | Agent prompt quality, credential leak detection |
Business & Revenue (6 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | Business Finder | 350+ | Lead generation, CRM, market research | | Revenue Tracker | 300+ | P&L aggregation, opportunity scoring | | Arena Manager | 250+ | Competition between revenue strategies | | Alpha (B2B Sales) | 300+ | Cold outreach, pipeline management | | Beta (Viral Marketing) | 300+ | Content creation, growth hacking | | Gamma (Digital Products) | 300+ | Product creation, pricing, funnels |
Trading & Markets (3 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | Trading Analyst | 400+ | ICT/SMC analysis, signal generation | | Market Pattern Detector | 300+ | Elliott Waves, Order Blocks, FVG detection | | Live Signal Monitor | 200+ | Real-time price alerts, movement detection |
Communication & Social (6 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | VRChat Persona | 300+ | Social interaction in virtual worlds | | Social Coach | 250+ | Conversation analysis, approach suggestions | | Translator | 200+ | Real-time 73-language translation | | Discord Manager | 250+ | Server management, webhooks, moderation | | Telegram Bot | 200+ | Notifications, command routing | | Voice Pipeline | 300+ | TTS/STT, voice commands, audio processing |
Infrastructure & Monitoring (5 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | System Architect | 350+ | Docker, monitoring, health checks | | Watchdog | 250+ | API balance monitoring, service health | | Auto-Healer | 200+ | Automatic service restart with circuit breaker | | Cloudflare Manager | 300+ | DNS, tunnels, workers, WAF | | System Pulse | 200+ | CPU, GPU, RAM, container monitoring |
Research & Analysis (4 prompts)
| Agent | Lines | What It Does | |——-|——-|————-| | Deep Researcher | 350+ | Multi-source research synthesis | | YouTube Intelligence | 250+ | Tech trend monitoring, video analysis | | Gemini Bridge | 200+ | Alternative model consultation | | Grok Bridge | 200+ | Counter-opinion, anti-bias checking |
Specialty (10 prompts)
Swimming coach, event management, secretary, contact analyzer, world explorer, solo leveling game, and more domain-specific agents.
What Makes These Different
1. They Are Complete
Each prompt is 200-800 lines. Not a paragraph – a full operational specification including identity, initialization procedures, capabilities, constraints, output format, error handling, and collaboration rules.
2. They Are Tested
Every prompt has been through thousands of real tasks. The constraints exist because we hit the failure mode they prevent.
Example: The “evidence-based completion” rule exists because without it, agents claim tasks are done without verification. We added it after 3 incidents where “fixed” bugs were not actually fixed.
3. They Work Together
Agents reference each other. The Orchestrator knows which agents exist and how to route to them. The Quality Tester knows what each agent should produce. The Memory Manager provides context to all agents.
4. They Are Model-Agnostic
These are plain markdown files. They work with:
- Claude (Opus, Sonnet, Haiku)
- GPT (4, 4o, 5.4)
- Gemini (Pro, Flash)
- Llama (3 70B+)
- Mistral (Medium+)
- DeepSeek
- Any model that accepts system prompts
5. They Include Anti-Patterns
Each prompt explicitly states what the agent should NOT do. This is often more important than what it should do:
## WHAT YOU ARE NOT
- NOT a code fixer -- report findings only, never modify code
- NOT an architect -- don't redesign systems you're reviewing
- NOT a diplomat -- if code is dangerous, say so directly
Free Sample
The GitHub repository includes:
- Complete Orchestrator prompt – the central routing agent
- 7 n8n workflow templates (JSON, ready to import)
- Tutorial: Build a multi-agent system from scratch
- Guide: AGENT.md complete reference
- 5 production patterns that survived 10,000+ tasks
- 7 common mistakes to avoid
- Framework comparison: CrewAI vs LangGraph vs Custom Prompts
Prompt Structure
Every prompt follows the same proven structure:
## IDENTITY
Who the agent is, what it does, what it does NOT do.
## INITIALIZATION
What the agent must do when it starts (check state, load context).
## CAPABILITIES
Tools, APIs, and skills available to the agent.
## PIPELINE
Step-by-step workflow for handling tasks.
## CONSTRAINTS
Hard rules that prevent common failure modes.
## OUTPUT FORMAT
Exact structure of responses.
## ERROR HANDLING
What to do when things go wrong.
## COLLABORATION
How this agent works with other agents.
This structure works because it maps to how LLMs process instructions: identity first (sets behavior baseline), then capabilities (defines action space), then constraints (narrows action space), then output format (structures responses).
Who Uses These
- Solo developers building AI assistants with multiple capabilities
- Startups creating agent-based products
- Enterprises evaluating multi-agent architectures
- AI researchers studying prompt engineering at scale
- n8n users building workflow-based AI systems
- Claude Code users setting up AGENT.md files
Get the Collection
Free: GitHub repo with orchestrator + templates + tutorials
Full collection: 49 Production Agent Prompts on Gumroad ($29)
- Use code LAUNCH49 for $10 off
- All 49 AGENT.md files
- All n8n workflow templates
- All documentation and guides
- Lifetime updates
Have questions? Start a Discussion on GitHub