Claude Memory Health — Development skill for Claude Code
Claude Code skill for auditing your MEMORY.md index — size, orphans, broken links, staleness.
How to install Claude Memory Health
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open alexknowshtml/claude-memory-health and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Claude Memory Health does
Claude Code skill for auditing your MEMORY.md index — size, orphans, broken links, staleness.
Alternatives in Development
- Reap Orphans — 扫描并清理本地过期的 PUA agent/state 19.5k ★
- Om Tidy — Self-maintenance pass — acts on every hygiene flag: archives completed work, groups loose clusters, splits ove 4.6k ★
- Memory Gc — Inspect or rotate the autopilot ledger JSONL files (findings.jsonl, negatives.jsonl) 3.8k ★
README
claude-memory-health
A Claude Code skill and scheduler for keeping your memory index focused — auditing `MEMORY.md` for bloat and stale entries, then autonomously demoting lower-priority notes to cold storage so the most valuable context loads every session.
Install via Claude Code
Paste this into Claude Code to install:
Install the memory-health skill from https://github.com/alexknowshtml/claude-memory-health:
Clone or copy the repo into .claude/skills/memory-health/ in the project root.
The scripts and cold-storage templates are included in the repo — no separate copies needed.
To set up cold storage: copy .claude/skills/memory-health/cold-storage/ to cold-storage/
in the project root as a starting point, then customize the domain files for your project.
Set MEMDIR to ~/.claude/projects/[your-project-name]/memory
Set COLDDIR to [project-root]/cold-storage
Once installed, run /memory-health to audit the memory index.
Requires [Claude Code](https://claude.ai/code) and [Bun](https://bun.sh).
The Problem
Claude Code's auto-memory saves a lot. Every correction, fact, and decision ends up in `MEMORY.md` — and it all gets injected at session start. Two problems compound over time:
**Signal dilution.** When everything is always-on, nothing is prioritized. A behavioral rule you need every session competes with an incident fix from three months ago.
**Silent truncation.** Claude Code loads up to 200 lines of `MEMORY.md` at session start. Lines past 200 are silently dropped — no warning, no error. Your oldest memories quietly disappear as new ones accumulate.
The fix isn't to stop saving memories. It's to sort them.
How It Works
**Hot** (`MEMORY.md`) — injected at every session start. Reserve for behavioral rules, communication patterns, safety gates: things Claude needs regardless of what you're working on.
**Cold** (`cold-storage/`) — domain-specific files loaded on demand. Not injected automatically, but searchable when work
Related Skills
Brain Health
Run a health check over the second brain - contradictions, broken/orphan links, duplicates, stale claims, sche
Recall Health
Show total-recall index health — last ingest age, hook fire rate, p95 latency, error count, DB size.
Recall Status
Show total-recall index health — DB size, session count, last ingest, top topics for this cwd.
Claude Ecosystem Hygiene
Three complementary skills for auditing and maintaining Claude Code ecosystem health: skills, memory, handoffs
Memory Ingest
Process new raw content into the agent memory knowledge base. Creates source/concept/entity pages, cross-links
Claudehut Learning Report
Show the ClaudeHut learning scoreboard — measured memory health (store size, reinforcement, effectiveness/recu
Related Agents
Wiki Auditor
Wiki health audit — finds and fixes broken links, missing frontmatter, orphans, near-duplicates. Fixes inline,
Linter
Validates wiki integrity — checks links, staleness, terminology, orphans
Memory Builder
Use when memory-orchestrator reports the context-pack missing or stale and it must be rebuilt — scans the repo