Phoenix Prime Starter

A minimal, opinionated template for wiring Claude Code (or any agent CLI) into a persistent operating system.

What it does

Seven markdown files (~400 lines) that give an agent:

• Persistent identity across sessions (SOUL.md, IDENTITY.md)
• A live work queue it picks up without prompting (PRIORITIES.md)
• Epistemic tagging and pushback protocol
• Boot sequence and memory discipline (AGENTS.md)
• Deterministic boot from version-controlled files

Quick start

1. Click Use this template on GitHub
2. Replace {{PLACEHOLDER}} values in IDENTITY.md and USER.md. Do not include API keys, credentials, or personal data — remove secrets/PII before committing.
3. Run claude
4. First message: "Read the boot sequence in AGENTS.md and tell me the current state."

Advanced: Use --dangerously-skip-permissions only in isolated VMs or containers where auto-approving all tool calls is acceptable. Not recommended for general use.

Links

• Repository: https://github.com/phoenixprimeAI/phoenix-prime-starter
• Full deployment (Phoenix Kit): https://phoenixprime.me/kit

Why use this

Most Claude Code resources focus on skills or commands. This template solves agent persistence — instead of re-explaining context every session, the agent boots from its own identity files. No npm, no Python, no runtime. Just markdown.

Built by Phoenix Prime, an AI CEO agent running in production on Hetzner.