How do I install the Memory Doctor skill?

Run npx skillmds@latest add lord1egypt/memory-doctor in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Memory Doctor safe to use?

This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: CAUTION, Skill Scanner: PASS. Capability flags: makes network calls. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Memory Doctor?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Memory Doctor free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Memory Doctor?

Lord1Egypt (@lord1egypt) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Memory Doctor by lord1egypt — Agent Skill

Name: Memory Doctor
Rating: 4.5 (4 reviews)
Author: Lord1Egypt

Memory Doctor

Keep an LLM agent's long-term memory healthy — the cure for context rot. Use when a memory store exists and the user wants to check or tidy it ("记忆体检", "整理记忆", "memory health", "memory cleanup", "dedup memory", "fix MEMORY.md index"), or whenever memory has grown long, contradictory, or its index drifted. Runs a read-only health report and safe, reversible repairs. Supports the Claude Code auto-memory format (frontmatter + [[links]] + MEMORY.md) and the mneme format (.mneme cells). Zero dependencies, zero LLM in v0.1.

Lord1Egypt Updated Sep 8, 2026 2 repo stars

File contents

memory-doctor — keep agent memory healthy

Memory skills write memory; almost none maintain it. Over time a store grows long, contradicts itself, sprouts dead links, and its index drifts — the agent then burns tokens parsing stale history instead of acting. That is context rot. This skill governs an existing store; it does not produce memory.

Producers write, memory-doctor keeps it alive. Complements memory skills, doesn't replace them. Source & issues: https://github.com/casperkwok/memory-doctor

When to use

A memory store exists (a MEMORY.md + *.md notes, or a *.mneme file) and the user wants to check or clean it: "记忆体检 / 整理记忆 / memory health / cleanup / dedup / fix the index".
Memory has clearly grown long, repetitive, or the index no longer matches the files.

How to run (zero deps, Python ≥ 3.9)

# read-only — safe anytime; print the health card first
python scripts/curate.py report --dir <memory-dir>

# auto-memory only: preview then apply the index reconciliation (snapshots first)
python scripts/curate.py lint --dir <memory-dir>
python scripts/curate.py lint --dir <memory-dir> --fix

# revert the last write
python scripts/curate.py undo --dir <memory-dir>

The format is auto-detected (*.mneme present → mneme, else auto-memory); override with --format auto-memory|mneme.

What the report checks

dead links, stale notes, oversized files, lexical duplicate candidates, and a freshness score (per-type exponential decay).
auto-memory: index drift (MEMORY.md vs files) and orphans (no inbound link, unindexed).
mneme: history cells (superseded/retired/proposed) excluded from active health, and supersede back-link symmetry; freshness uses each cell's seen date.

What it fixes (v0.1 — deterministic & reversible only)

Index reconciliation (auto-memory): add missing MEMORY.md entries, drop dangling ones, preserving your hand-written hooks. A full-directory snapshot is taken first; undo reverts.
mneme's spine is derived, so there is no index to reconcile — use report.

Semantic fixes that need a model — true duplicate merge, contradiction reconcile, oversized compaction — are flagged in v0.1 and resolved in v0.2 (no embeddings; manual-apply, snapshot-backed). Never auto-edit memory content on the user's behalf without confirmation.

Safety

Every write snapshots the whole directory to .memory-doctor/snapshots/<ts>/ and logs to .memory-doctor/changelog.md; undo restores it. v0.1 only ever rewrites the index file.

Optional: a read-only health nudge

hooks/health_nudge.py (in the repo) can be wired to a Claude Code SessionStart hook to print a one-line nudge when health drops. It is read-only and never edits anything. Opt-in.

---name: memory-doctordescription: Keep an LLM agent's long-term memory healthy — the cure for context rot. Use when a memory store exists and the user wants to check or tidy it ("记忆体检", "整理记忆", "memory health", "memory cleanup", "dedup memory", "fix MEMORY.md index"), or whenever memory has grown long, contradictory, or its index drifted. Runs a read-only health report and safe, reversible repairs. Supports the Claude Code auto-memory format (frontmatter + [[links]] + MEMORY.md) and the mneme format (.mneme cells). Zero dependencies, zero LLM in v0.1.---# memory-doctor — keep agent memory healthyMemory skills *write* memory; almost none *maintain* it. Over time a store grows long,contradicts itself, sprouts dead links, and its index drifts — the agent then burns tokensparsing stale history instead of acting. That is **context rot**. This skill governs an*existing* store; it does not produce memory.> Producers write, memory-doctor keeps it alive. Complements memory skills, doesn't replace them.> Source & issues: <https://github.com/casperkwok/memory-doctor>## When to use- A memory store exists (a `MEMORY.md` + `*.md` notes, or a `*.mneme` file) and the user wants to check or clean it: "记忆体检 / 整理记忆 / memory health / cleanup / dedup / fix the index".- Memory has clearly grown long, repetitive, or the index no longer matches the files.## How to run (zero deps, Python ≥ 3.9)```bash# read-only — safe anytime; print the health card firstpython scripts/curate.py report --dir <memory-dir># auto-memory only: preview then apply the index reconciliation (snapshots first)python scripts/curate.py lint --dir <memory-dir>python scripts/curate.py lint --dir <memory-dir> --fix# revert the last writepython scripts/curate.py undo --dir <memory-dir>```The format is auto-detected (`*.mneme` present → mneme, else auto-memory); override with`--format auto-memory|mneme`.## What the report checks- **dead links**, **stale** notes, **oversized** files, lexical **duplicate candidates**, and a **freshness** score (per-type exponential decay).- auto-memory: **index drift** (MEMORY.md vs files) and **orphans** (no inbound link, unindexed).- mneme: **history** cells (superseded/retired/proposed) excluded from active health, and **supersede back-link symmetry**; freshness uses each cell's `seen` date.## What it fixes (v0.1 — deterministic & reversible only)- **Index reconciliation** (auto-memory): add missing `MEMORY.md` entries, drop dangling ones, preserving your hand-written hooks. A full-directory snapshot is taken first; `undo` reverts.- mneme's spine is *derived*, so there is no index to reconcile — use `report`.Semantic fixes that need a model — true duplicate **merge**, contradiction **reconcile**,oversized **compaction** — are *flagged* in v0.1 and *resolved* in v0.2 (no embeddings;manual-apply, snapshot-backed). Never auto-edit memory content on the user's behalf withoutconfirmation.## SafetyEvery write snapshots the whole directory to `.memory-doctor/snapshots/<ts>/` and logs to`.memory-doctor/changelog.md`; `undo` restores it. v0.1 only ever rewrites the index file.## Optional: a read-only health nudge`hooks/health_nudge.py` (in the repo) can be wired to a Claude Code `SessionStart` hook to printa one-line nudge when health drops. It is read-only and never edits anything. Opt-in.

Frequently asked questions