Starter templates for building AI agents with Claude Code.
Each blueprint is a self-contained prompt + configuration that you can drop into your project and customize. They're designed to work standalone or integrate with TinMan for scheduled execution and C3Poh for mobile notifications.
| Blueprint | Use Case | Complexity |
|---|---|---|
| Customer Support Triage | Route incoming tickets, draft responses, escalate edge cases | Low |
| Code Review Assistant | Pre-review PRs before human review, catch common issues | Low |
| Research Synthesizer | Summarize documents, extract insights, build knowledge bases | Medium |
| Lead Qualifier | Score inbound leads, enrich data, draft personalized outreach | Medium |
| Meeting Notes to Action Items | Extract tasks, decisions, and follow-ups from transcripts | Low |
1. Copy the blueprint to your project:
curl -o AGENT.md https://raw.githubusercontent.com/andyuninvited/agent-blueprints/main/blueprints/customer-support-triage.md2. Customize the configuration section at the top of the file with your specific context.
3. Run it with Claude Code:
claude --print "$(cat AGENT.md)"4. (Optional) Schedule it with TinMan:
If you want the agent to run on a schedule, rename it to HEARTBEAT.md and configure TinMan:
pip install tinman-for-claudecode
tinman initEvery blueprint follows this format:
# Agent Name
## Configuration
- inputs: what data this agent expects
- outputs: what it produces
- integrations: external systems it works with
## Context
Background the agent needs to do its job effectively.
## Instructions
Step-by-step process the agent follows.
## Response Format
How the agent should structure its output.
## Constraints
What the agent should never do.This structure makes blueprints predictable, composable, and easy to audit.
Use blueprint-template.md as a starting point. The key is being specific about:
- What inputs the agent receives — files, data, context
- What outputs you expect — format, structure, destination
- What constraints apply — things it must never do, decisions it can't make alone
Vague blueprints produce vague results. Specific blueprints produce reliable agents.
| Tool | How it works |
|---|---|
| TinMan | Run blueprints on a schedule. Perfect for monitoring, reporting, and proactive agents. |
| C3Poh | Get agent outputs sent to Telegram. Great for alerts and summaries you want on your phone. |
PRs welcome. One blueprint per file, follow the structure in blueprint-template.md.
GNU GPLv3
Built by @andyuninvited