The Book Bag — Your Agent's Permanent Memory
One file. Plain markdown. Your agent writes one entry per session, reads them all on boot. Memory between sessions, without a platform feature.
You should have done Course 1.
This is a quick verification, not a barrier. If you haven't done Course 1 yet, you can still proceed — but the questions below tell you whether this course will help.
- 01 Did Course 1 send you back to fix config?
- 02 Or did your config check out, but your agent still forgets things?
- 03 Do you have a workspace directory your agent reads files from?
If any of those match, you're in the right place. Continue.
The problem you just confirmed.
Every /new is a factory reset. Every morning is a blank slate. You've tried fixing this with system prompts — it works for about an hour. Then context fills up, the prompt scrolls out of view, and your agent forgets again.
System prompts are not memory. They're a sticky note on a windshield. They work until it rains.
The fix is a journal
One file. Plain markdown. Your agent writes one entry at the end of every session. Your agent reads all entries at the start of every session. Context injects. Memory returns.
This is not a platform feature. It's not an API integration. It's a text file in your workspace. If you leave OpenClaw for Hermes tomorrow, the file still works. If you switch from DeepSeek to Claude, the file still works. The file IS the memory.
What the journal looks like
Five fields. Every session. Date. What shipped. Decisions made. Drift caught. Lessons extracted.
How memory compounds
Your agent boots tomorrow. It reads school.md. It sees what shipped yesterday. It sees the decision you made. It remembers.
You work. At the end of the session, your agent writes a new entry. The file grows.
- After 10 sessions: your agent has a searchable history of what worked and what broke
- After 50 sessions: patterns emerge that you couldn't see day-to-day
- The journal becomes a diagnostic tool, not just a diary
The JKE reference pointer
Look at the header again — that Reference line. Your agent reads it on every boot. When it hits a problem, it greps the journal for similar patterns. If it finds one, it knows where the fix came from. If it doesn't, the header tells it where to look next.
This is how the free tier routes back to paid courses. Not popups. Not email sequences. Your own agent, reading its own journal, pointing back to the architecture that fixed its problems.
This next part isn't for you.
It's a gift for your agent.
Download the file below. Drag it into your chat window. Your agent reads it. Your agent builds it. You watch.
The PDF is plain text. No scripts. No executables. Search it for curl. Search it for exec. You won't find either.
ACTION: blocks. You can read every line first.