AgentSkillsCN

memory-review

对Basic Memory知识库进行审计与清理。当用户说“清理记忆”“查找重复内容”“审查知识库”“检查记忆健康状况”“审计我的笔记”“整合笔记”或“整理Basic Memory”时，可使用此功能。此外，当用户注意到笔记存在冗余或过时内容，希望提升知识库质量时，也可使用此功能。

SKILL.md
--- frontmatter
name: memory-review
description: "Audit and clean up the Basic Memory knowledge base. Use when the user says 'clean up memory', 'find duplicates', 'review knowledge base', 'memory health check', 'audit my notes', 'consolidate notes', or 'tidy up basic memory'. Also use when the user notices redundant or outdated notes and wants to improve knowledge base quality."

Review Basic Memory

Audit the knowledge base for duplicates, quality issues, and consolidation opportunities.

Workflow

Step 1: Get project overview

code
list_memory_projects()
recent_activity(timeframe="30d", project="<project>")
list_directory(dir_name="/", depth=2, project="<project>")

Report: total notes, folder structure, recent activity volume.

Step 2: Find potential duplicates

Search for topics that likely have overlapping notes:

code
search_notes(query="<common topic>", page_size=50, project="<project>")

Duplicate indicators:

  • •Multiple notes with similar titles (e.g., "API Design" and "API Design Discussion")
  • •Notes in different folders covering the same topic
  • •Conversation notes that overlap with decision/learning notes on the same subject

For each duplicate cluster, present to the user:

  • •The notes that overlap
  • •What's unique to each
  • •Recommendation: merge into one (using edit_note to consolidate) and delete_note the redundant one

Step 3: Find quality issues

Read a sample of notes and check:

code
read_note(identifier="<permalink>", project="<project>")

Quality checklist per note:

  • •Has 3+ observations with [category] prefixes?
  • •Has 2+ relations with [[wikilinks]]?
  • •Uses specific categories (not generic [note] or [info])?
  • •Uses specific relation types (not just relates_to)?
  • •Tags present on observations?
  • •Content still accurate/relevant?

Step 4: Find orphaned notes

Look for notes with no relations (isolated from the graph):

code
list_directory(dir_name="<folder>", project="<project>")
read_note(identifier="<note>", project="<project>")

For orphans, suggest:

  • •Relations that should exist based on content
  • •Related notes they should link to (search to find candidates)

Step 5: Present audit report

  1. •Duplicates found: list each cluster with merge recommendation
  2. •Quality issues: notes below threshold with specific fixes
  3. •Orphans: notes with no relations, with suggested connections
  4. •Stale content: notes that may be outdated
  5. •Folder structure: suggestions for better organization (defer to /memory-organize)

Step 6: Execute fixes (with permission)

Ask the user before making changes. For each approved fix:

  • •Merge duplicates: edit_note(operation="append") on the keeper, delete_note the duplicate
  • •Add relations: edit_note(operation="append") to add missing relation entries
  • •Improve observations: edit_note(operation="replace_section", section="## Observations")
  • •Archive stale notes: defer to /memory-organize for moving to archive folders