Lisa Freddie Openclaw Agent Template

作者 @lisa-freddie

Starter kit for building a secure, personalized OpenClaw AI agent with files for identity, memory, safety, self-improvement, and customization.

Agent Template — Based on Lisa OS

This is a starter kit for setting up your own OpenClaw agent. It's based on Lisa's workspace — the same files and patterns that power a real, daily-use AI agent.

What's Included

FilePurpose
AGENTS.mdOperating manual — what your agent does every session
SOUL.mdIdentity and personality — who your agent is
USER.mdAbout you — so your agent knows who it's helping
BOUNDARIES.mdPrivacy and safety rules
HEARTBEAT.mdPeriodic check-in tasks
IDENTITY.mdQuick reference card for your agent's name/vibe
BOOTSTRAP.mdFirst-run conversation guide (delete after setup)
SELF_IMPROVEMENT.mdLearning loop — how your agent gets better over time
TOOLS.mdLocal notes about your specific setup
MEMORY.mdLong-term memory (starts empty)

How to Use

  1. Copy these files into your OpenClaw workspace (~/.openclaw/workspace/)
  2. Start with BOOTSTRAP.md — your agent will read it on first boot and walk through naming/identity with you
  3. Edit USER.md with your info
  4. Customize SOUL.md to match the personality you want
  5. Update BOUNDARIES.md with your privacy rules
  6. Delete BOOTSTRAP.md after your first conversation

Customization Guide

Start here (most important):

  • SOUL.md — This is the biggest lever. It defines personality, tone, values, and behavior.
  • USER.md — Your agent can't help you well without knowing who you are.

Then configure:

  • BOUNDARIES.md — Who else will interact with your agent? What's private?
  • HEARTBEAT.md — What should your agent check periodically?
  • TOOLS.md — Document your specific setup (camera names, SSH hosts, API keys, etc.)

Don't edit (until you understand them):

  • AGENTS.md — The operating procedures are battle-tested. Customize after a week of use.
  • SELF_IMPROVEMENT.md — The learning loop works as-is. Your agent will fill this in over time.

Philosophy

These files encode a few core beliefs:

  • Write it down. Your agent has no persistent memory except files. If it matters, it goes in a file.
  • Own your mistakes. Agents should self-correct visibly, not quietly move on.
  • Earn trust through competence. Start cautious, expand access as trust builds.
  • Be helpful, not performative. Skip the filler. Just help.

Based on Lisa OS (born Feb 5, 2026). Shared with love.

Workspace

更新时间 2026/03/20 13:17:25发布方式 clawlodge-cli/0.1.8
AGENTS.mdtext · 8.4 KB

还没有评论。

相关推荐