Memory Organization

memory-organization audits and tidies up the project's persistent stores — knowledge, memory, semantic models, and the AGENTS.md index. It finds duplicates, contradictions, misfiled entries, and gaps, then proposes fixes and applies the ones you approve.

Run it periodically, or whenever the stores feel cluttered.

What it does

Over time, knowledge and memory accumulate: the same fact gets recorded twice, two entries start to conflict, something filed as memory really belongs in knowledge, or an obvious fact never got written down. memory-organization reviews the stores as a whole and brings them back into a clean, consistent state.

What it checks

• Duplicates — the same fact recorded in more than one place.
• Contradictions — entries that conflict with each other.
• Misfiled — knowledge that should be memory, or vice versa.
• Gaps — obvious facts that should be recorded but aren't.
• Staleness — entries that are no longer true.

When to use it

• The stores feel cluttered or contradictory.
• After heavy usage that added many entries.
• Knowledge and memory have drifted or started to overlap.
• You want to clean up or reorganize.

How to use it

Run memory-organization as a skill from the REPL. You can optionally tell it which stores to focus on; with no hint, it audits all of them. It reports what it found and proposes fixes before applying the ones you approve — so you stay in control of any change.

Example

After a few weeks of heavy use, your project has the same "fiscal year starts in April" rule written in two knowledge files, a memory note that contradicts a newer preference, and a stale entry referring to a table you dropped. Running memory-organization flags all three, proposes merging the duplicate, removing the stale entry, and reconciling the conflict — and applies them once you confirm.