memory-garden

Scan MEMORY.md and entry files to detect stale memory items.

8|2|Updated May 3, 2026
One-click install
npx skills add https://github.com/ejoongseok/claude-settings --skill memory-garden-ejoongseok
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: memory-garden
Source: https://github.com/ejoongseok/claude-settings/tree/main/claude-code/skills/memory-garden
Command: npx skills add https://github.com/ejoongseok/claude-settings --skill memory-garden-ejoongseok

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Claude Code auto memory can accumulate outdated project closures, already-resolved feedback, contradictions, and stale references, which reduces retrieval quality and causes the system to keep acting on obsolete information.

Core Features & Use Cases

  • Meaning-based cleanup of auto memory entries: Scans MEMORY.md plus individual memory entries, builds relationships, and assigns types such as completed, resolved, outdated, promoted, duplicate, or stale.
  • Evidence-first validation with confidence: Requires concrete evidence in the form path:line + quote, assigns confidence (High/Med/Low), and degrades anything weak into Stale for user review.
  • Dry-run and gated execution: Produces a diff-style dry-run preview for archive/delete/edit/consolidate actions and waits for explicit user approval before modifying memory.
  • Index readability improvements: Optionally groups MEMORY.md into categories (structure only) to keep activation effective without changing entry text.

Quick Start

Run memory-garden to perform a dry-run cleanup of your Claude Code auto memory by scanning MEMORY.md and memory entry files, producing an evidence-backed report, and applying only your approved archive/delete/update/consolidation changes.

Frequently Asked Questions about memory-garden

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I clean up outdated entries in Claude Code auto memory?

To clean up outdated entries in Claude Code auto memory, you can scan MEMORY.md and individual entry files to detect completed, resolved, duplicate, or stale items. The tool then proposes safe, evidence-backed actions like archiving or deleting after a mandatory dry-run preview.

What is evidence-based memory management for Claude Code?

Evidence-based memory management for Claude Code requires concrete proof, specifically a file path, line number, and quote, to validate changes to auto memory entries. This method assigns confidence levels to prevent unsafe modifications and degrades weak evidence to stale status for manual review.

Can I preview changes to MEMORY.md before applying them?

Yes, you can preview changes to MEMORY.md before applying them because the system enforces mandatory dry-run diffs. It generates a diff-style preview of all archive, delete, edit, or consolidate actions and waits for your explicit approval before modifying any memory files.

How do I remove duplicate or stale references from Claude Code memory?

You remove duplicate or stale references from Claude Code memory by scanning the auto memory directory to build relationships between entries. The process identifies duplicates and outdated references, requiring high confidence evidence to automatically handle them while blocking low-confidence stale cases for user review.

Does Claude Code memory cleanup work without manual confirmation?

No, Claude Code memory cleanup does not work without manual confirmation because the system uses gated execution. It produces a dry-run diff and explicitly waits for your approval before applying any archive, delete, update, or consolidation changes to the memory files.

What are the limitations of automated auto memory archiving?

A key limitation of automated auto memory archiving is that it blocks automatic handling of low-confidence stale cases. Any memory entry lacking concrete evidence like a path, line number, and quote is degraded to stale status and requires explicit user review before modification.