openclaw-workspace-template

Production-tested workspace template for Claude Code (and OpenClaw) AI agents, extracted from a real agent running 4+ months in production.

A fresh OpenClaw agent is stateless — it wakes up, helps you, and forgets everything. This template turns it into something that persists, learns, and improves over time.

What You Get

• 🧠 Your agent remembers — Daily logs (memory/YYYY-MM-DD.md) + curated long-term memory (MEMORY.md) with priority levels (P0/P1/P2) and auto-expiry. It reads yesterday's and today's logs every session.

• 🔧 Your agent learns from mistakes — A three-tier self-improvement system (repair / optimize / innovate) tracks errors, corrections, and knowledge gaps. Problems that recur 3+ times auto-promote to long-term memory.

• 🤖 Your agent delegates — Battle-tested sub-agent patterns with result verification and delivery confirmation. Main agent stays focused while sub-agents handle heavy lifting.

• 🌙 Your agent dreams — Weekly "cold memory association" finds cross-domain insights by randomly pairing unrelated memories. Weekly "rumination" (Wed 21:03) detects contradictions between recent and long-term memory. Monthly auto-archival keeps memory fresh.

• 📋 Your agent maintains itself — Routine checks framework (Type A: fixed logic / Type B: needs LLM) so your agent doesn't waste tokens on things a shell script can do.