Maintenance goals
Maintenance is not busywork. It protects the three things that matter most:
- startup clarity
- durable signal quality
- trust that the agent is reading the right things
If those break down, the /Brain stops improving sessions.
A practical cadence
Use a simple rhythm. Update context/now.md daily or near-daily when priorities change. Consolidate raw input and prune stale detail weekly. Review me/WORKING_WITH_ME.md and MEMORY.md for drift monthly.
The exact cadence matters less than making it real and repeatable.
Monthly audit checklist
Use this once a month (or every two months if your Brain changes slowly). Should take 15-20 minutes.
Identity layer
- Open
me/WORKING_WITH_ME.md— is this still how you work? Have your preferences or decision patterns changed? - If you use agent-specific files (e.g.
SOUL.md), check whether those instructions are still relevant.
Context layer
- Open
context/now.md— are these priorities current? Remove anything that finished or no longer matters. - Check for stale project references that should be archived.
Memory layer
- Open
MEMORY.md— anything stale, wrong, or no longer useful? Remove it. - A wrong memory is worse than no memory.
Structure check
- Any empty or unused folders? Remove or merge them.
- Are startup files still small enough to scan in one sitting? Split if they have grown too large.
When to run this
Run this monthly if you use your Brain in most AI sessions and your work changes often. Every two months is fine if your Brain is stable and workflows don't shift much. Run it immediately after any major change — new job, new project, new tool.
The goal is not perfection. The goal is that the next AI session starts from files you still believe.
File-size discipline
Startup files should be easy to scan in one sitting. When they grow too large, split details into deeper files and link outward.
Good pattern: small startup files for session start, deeper files for optional reference, and search only when the task needs broader recall.
Identity refresh
Update identity files only when a change is real:
- your decision style changed
- your communication preferences changed
- recurring mistakes show the agent is reading stale instructions
Do not rewrite the identity layer every week just because you can.
Common mistakes
- keeping stale project context in
now.md - letting
MEMORY.mdbecome a second inbox - adding folders faster than they earn their place
- assuming an agent can infer your current priorities from old files
Next steps
- Read Obsidian as the Foundation for how agents read and write to your vault
- Read FAQ for conceptual questions that come up often