Get the exact agent template and setup steps that give Claude a persistent role with memory, deployed in 30 minutes.
Free template · copy-paste ready · built by the BrainVaultAI team
An AI agent isn't a chatbot, it's Claude with a defined role, a persistent context file, a task loop, and clear success criteria. The difference between 'asking Claude a question' and 'running an agent' is structure. This template gives you that structure. You don't need to code anything. You need a text file (the system prompt), a context file (what the agent knows), and a task file (what it's doing right now). Start with one agent for one repeating task. The best first agents are things you do weekly that follow a pattern: competitive monitoring, content review, weekly brief generation, inbox triage.
Agent System Prompt: Template
# Agent Role
You are [Agent Name], a specialized agent for [specific function]. # What You Do
[2-3 sentences describing the agent's exact job. Be specific. 'You monitor competitor pricing and surface changes that require a response' not 'You help with competitive analysis.'] # What You Always Have Access To
- Context file: [describe what static context you'll provide each run]
- Task file: [describe what the current task input looks like] # How You Work
1. Read the context file first
2. Read the current task input
3. Do the work
4. Output in this exact format: [define your output format]
5. Flag anything that requires human decision before acting # What You Never Do
- [List hard constraints, things the agent must not do without explicit approval] # Done When
[Define what task completion looks like. Be specific enough that both you and the agent agree on when it's done.]
Context File: Template
# Agent Context, [Agent Name]
Last updated: [date] ## Background
[What does this agent need to know about your business, product, or situation to do its job? Write it as if briefing a smart new contractor.] ## Key Definitions
[Any terms, categories, or concepts the agent will encounter that have specific meaning in your context.] ## Decision Rules
[Rules the agent can apply without asking you. 'If X, then Y' format.] ## Things to Flag for Human Review
[Situations where the agent should stop and surface something to you rather than deciding on its own.]
Task File: Template
# Current Task, [Agent Name]
Date: [today] ## Task
[One clear sentence describing what the agent needs to do right now.] ## Input
[Paste the raw input the agent will work with, a document, a list, a dump of data, whatever applies.] ## Expected Output
[Describe the output format you want back. Be specific: 'A numbered list of 5 items, each with a one-sentence summary and a recommended action.'] ## Priority
[High / Medium / Low, and why if it matters.]
How to Run Your Agent: The 5-Step Pattern
Step 1: Create your system prompt file using the template above. Save it as agent-[name]-system.md. Step 2: Create your context file. This is static, update it when your business context changes. Save as agent-[name]-context.md. Step 3: For each run, create a task file with the current task and input. Save as agent-[name]-task-[date].md. Step 4: Open Claude. Paste the system prompt first. Then paste the context file. Then paste the task file. Say 'Begin.' Step 5: Review the output. If the agent flagged something for human review, handle it. Update the context file if the agent surfaced something worth remembering for next time. That's it. No code. No API. A real agent loop you can run today.
Want the editable version?
We'll send the editable version of this asset, plus a short series of our highest-leverage AI plays for operators. No spam. Unsubscribe in one click.
Free. No spam. Unsubscribe in one click. By submitting you agree to receive emails from BrainVaultAI.