What problem does it solve?
Long-term memory files in an agent workspace grow unbounded over time, causing bloated auto-loaded context, stale rules, and broken topic pointers. This Skill performs periodic 72-hour consolidation passes that delete outdated entries, demote stories to topic files, and merge duplicate rules so the auto-loaded memory layer stays within budget.
Core Features & Use Cases
- Memory Linting: Runs
memory_lint.py to detect budget overruns on MEMORY/USER/SOUL files, missing daily logs despite git activity, overdue gardener or molt runs, stale verified markers, and broken topic pointers.
- Structured Consolidation: Deletes obsolete rules, demotes case details into
memory/topics/, merges same-root rules, and writes a dated run report under memory/gardener/.
- Safe Git Handling: Records the working-tree baseline, commits only changes attributable to the current maintenance run, and never pushes or touches pre-existing uncommitted edits.
- Use Case: An agent workspace whose
04-MEMORY.md has grown past its 50KB budget can be audited and trimmed in one run, with stories moved to topic files and a lint report confirming all pointers remain valid.
Quick Start
Ask the agent to run the myagents-memory-gardener skill on this workspace to lint and consolidate the long-term memory files.