Guide · Agent skill
How to use the memory-review agent skill (Auto-Memory Review)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The memory-review skill is an agent skill for Claude Code that audits the notes the agent has saved about your project in its auto-memory, which live in MEMORY.md and optional topic files. It reports which notes look like proven patterns worth promoting into permanent rules, which look stale, which overlap, and which conflict with your CLAUDE.md. This guide covers what the skill does, how to install it, how to trigger it, and what to watch for.
Auto-Memory Reviewalirezarezvani/claude-skills · MIT Download the skill Quick start
- 1Download the ZIP. Get the memory-review skill ZIP from this page.
- 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
- 3Confirm auto-memory is on. Make sure Claude Code has auto-memory enabled and the project has a MEMORY.md file.
- 4Run a quick review. Type /si:memory-review --quick to get counts and the top three promotion candidates.
- 5Run the full review. Type /si:memory-review to get the complete health report, stale entries, consolidation groups and conflicts.
- 6Act on the recommendations. Promote proven patterns into CLAUDE.md or .claude/rules/, delete stale entries, and merge overlapping notes.
What is an agent skill, and what does memory-review do?
An agent skill is a folder of instructions (centered on a SKILL.md file) that teaches an AI coding agent how to handle a specific task. When a request matches the skill, the agent follows its steps.
The memory-review skill teaches the agent to audit Claude Code's auto-memory for a project. It reads MEMORY.md and any topic files, checks them against CLAUDE.md and .claude/rules/, and produces a report with a health summary, promotion candidates, stale entries, consolidation groups, conflicts and recommendations.
- Promotion candidates: notes that appear in 2+ sessions, are broadly useful, are actionable, and are not already in CLAUDE.md or .claude/rules/.
- Stale entries: notes that reference files that no longer exist, outdated tools or versions, or that contradict CLAUDE.md.
- Consolidation: several entries about the same topic that could merge into one rule.
- Conflicts: MEMORY.md entries that contradict CLAUDE.md.
- Health: MEMORY.md line count against the 200-line startup limit, plus topic file and rule file counts.
How do you install the memory-review skill?
Download the ZIP for the skill from this page. Then choose one of the two install locations below.
For a personal install that applies across your projects, unzip the download into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project-only: .claude/skills/ inside your project
What do you need before running it?
The skill depends on Claude Code with auto-memory enabled. Without auto-memory there is nothing to review; the skill will report that memory may be disabled and suggest checking with /memory.
The agent needs Bash or shell access so it can list files and run find and sed. The project should contain a MEMORY.md, and ideally a CLAUDE.md and a .claude/rules/ directory, since the skill cross-references them.
- Claude Code with auto-memory enabled
- Bash/shell access to list files and run find and sed
- A project with MEMORY.md, and optionally CLAUDE.md and .claude/rules/
How do you trigger the skill?
You can run the slash command directly. The skill supports three focused options in addition to the full review.
You can also ask in plain language, such as asking what has been learned and what should be promoted or pruned. The skill's description is written to match those requests.
Use --quick for a fast summary of counts and the top three candidates. Use --stale to focus on outdated entries, and --candidates to see only promotion candidates.
- /si:memory-review — full review
- /si:memory-review --quick — summary only (counts and top 3 candidates)
- /si:memory-review --stale — stale and outdated entries
- /si:memory-review --candidates — promotion candidates only
/si:memory-reviewWhat does a review look like?
The report is organised into sections with counts. Each promotion candidate includes a suggested target, either CLAUDE.md or a file in .claude/rules/. Stale entries include the line number and the reason they were flagged.
The skill is written to recommend changes rather than edit files on its own. Review the suggestions and make the changes yourself or ask the agent to act on specific items.
- Memory Health: MEMORY.md line count out of 200, topic files, CLAUDE.md lines, rule files
- Promotion Candidates with suggested targets
- Stale Entries with line numbers and reasons
- Consolidation groups, Conflicts and Recommendations
When should you run it?
The source suggests running a full review after a major feature or debugging session, when /si:memory-status shows MEMORY.md is over 150 lines, weekly during active development, before a new project phase, and after onboarding a new team member.
For a low-overhead check, run --quick often. Run the full review when MEMORY.md is getting crowded.
- After a major feature or debugging session
- When MEMORY.md is over 150 lines
- Weekly during active development
- Before a new project phase
- After onboarding a new team member
Safety notes: scripts and data
This skill runs scripts. It does not use the network. The report quotes memory entries, so review output may contain sensitive text.
Do not share the output outside your team if the memory directory holds sensitive data.
- Runs scripts: yes
- Uses network: no
- Avoid sharing reports that quote sensitive memory entries outside the team
What people use it for
- Keeping MEMORY.md under the 200-line startup limit by pruning outdated notes.
- Turning a lesson the agent has repeatedly learned, such as a testing convention, into a rule in CLAUDE.md.
- Checking whether saved memory contradicts current CLAUDE.md rules after the rules changed.
- Running a weekly cleanup of a project's agent memory during active development.
- Reviewing what the agent learned after a new team member onboards.
Frequently asked questions
How do I install the memory-review skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
What do I need to use memory-review?
You need Claude Code with auto-memory enabled, Bash or shell access to run find and sed, and a project with a MEMORY.md file. CLAUDE.md and .claude/rules/ are optional but the cross-referencing depends on them.
Is memory-review free?
The skill is published under the MIT licence. The source does not mention pricing for anything beyond the licence terms.
What if the skill says memory is disabled?
The skill reports that auto-memory may be disabled when the memory directory does not exist. The source suggests checking with /memory to confirm the feature is on.
Does memory-review change my files?
According to the skill's instructions, it reports findings and recommendations rather than editing files itself. You decide what to promote, merge or delete.
Does memory-review need an internet connection?
No. The skill is listed as not using the network, but it does run scripts, so review what it runs before use.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use memory-review if you work in Claude Code with auto-memory, have a MEMORY.md that keeps growing, or want to move repeated lessons into shared CLAUDE.md or .claude/rules/ entries. Skip it if you do not use auto-memory, or if your memory holds sensitive data you would not want quoted in a shared report.
Download the skill