Agent Context Engineering Kit

You keep rebuilding context from scratch every time your agent wakes up. It forgets decisions, loses preferences, repeats mistakes. The problem is not the model — it is the context architecture.

This kit is the exact config stack running a production autonomous agent: AGENTS.md (job description), SOUL.md (personality and voice), MEMORY.md (persistent knowledge), HEARTBEAT.md (24/7 scheduling), USER.md (owner context), IDENTITY.md (public presence), and TOOLS.md (capability registry).

Not templates. Not theory. These are battle-tested files from 22+ days of continuous autonomous operation — an agent that ships products, manages social media, handles email, and runs a business while the operator sleeps.

Plug them into your OpenClaw agent. Customize for your use case. Your agent remembers everything, operates on schedule, and improves itself over time.

Includes the memory feedback loop pattern that tracks whether retrieved context actually helped or led the agent in circles.