
remember
by astrosteveo
Semantic memory for Claude Code - persistent context across sessions
SKILL.md
name: remember description: This skill should be used when the user asks to "remember this", "save for later", "store this information", "persist this context", or wants to explicitly save important information for future sessions. context: fork hooks: Stop: - hooks: - type: command command: "${CLAUDE_PLUGIN_ROOT}/scripts/validate-remember.sh" timeout: 10
Remember - Persist Important Context
Save important information to engram memory for retrieval in future sessions.
Execution
Step 1: Identify What to Remember
Analyze $ARGUMENTS or recent conversation to determine:
- What information needs persistence
- Why it matters for future sessions
- Category that best fits (general, decision, or lesson)
Step 2: Classify Content Type
| If the content is... | Use this tool |
|---|---|
| General information, preference, or context | memory_remember |
| Architectural or implementation choice | memory_decision |
| Bug fix, gotcha, pattern, or anti-pattern | memory_lesson |
Step 3: Structure the Content
Write content that will be useful when retrieved later:
- Include context ("When working on X...")
- Include the actual information
- Include rationale ("Because Y...")
Example structures:
For preferences:
"User prefers {preference} when {context}. Reason: {rationale}."
For decisions:
"Selected {choice} for {purpose}. Key factors: {factors}. Trade-offs: {trade-offs}."
For lessons:
"When {situation}: {lesson}. Root cause: {cause}. Prevention: {how to avoid}."
Step 4: Save to Memory
For general information:
mcp__plugin_engram-mcp_engram__memory_remember
content: "{structured content}"
tags: ["{relevant}", "{tags}"]
For decisions:
mcp__plugin_engram-mcp_engram__memory_decision
content: "{structured content}"
category: "architecture" # or implementation, tradeoff, tooling
alternatives: ["{other options considered}"]
For lessons:
mcp__plugin_engram-mcp_engram__memory_lesson
content: "{structured content}"
category: "pattern" # or bug_fix, gotcha, anti_pattern
root_cause: "{cause if known}"
Step 5: Confirm Persistence
Report to user:
- What was saved
- Which tool was used
- Tags/categories assigned
- How to retrieve later (search query suggestion)
Tag Guidelines
Use consistent, searchable tags:
| Domain | Example Tags |
|---|---|
| Technology | react, typescript, postgresql |
| Concern | performance, security, accessibility |
| Type | preference, constraint, requirement |
| Feature | auth, payments, notifications |
Examples
User: "Remember that I prefer Tailwind over styled-components"
mcp__plugin_engram-mcp_engram__memory_remember
content: "User prefers Tailwind CSS over styled-components for styling. Reason: utility-first approach, better performance, smaller bundle size."
tags: ["preference", "styling", "tailwind", "css"]
User: "We decided to use PostgreSQL, not MongoDB"
mcp__plugin_engram-mcp_engram__memory_decision
content: "Selected PostgreSQL as the primary database. Reasons: ACID compliance needed for financial transactions, JSONB provides document flexibility, team has SQL expertise."
category: "architecture"
alternatives: ["MongoDB", "MySQL"]
User: "Remember that auth tokens expire silently - we spent hours debugging this"
mcp__plugin_engram-mcp_engram__memory_lesson
content: "Auth tokens expire silently without error - requests return 401 but error message is vague. Always check token expiry first when debugging auth issues."
category: "gotcha"
root_cause: "Token refresh middleware wasn't logging expiry events"
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon