Quick Start
When the user says "weekly review", execute the following process:
Important: Before executing the weekly review, you must first call the mem-file-scan skill to scan files modified this week and supplement L1 records.
- File scan (preliminary step) - Call mem-file-scan to scan files modified this week and supplement L1 records
- Read this week's L1 records - Retrieve all context records from this week
- Identify recurring patterns - Count behaviors/preferences that appear 3 or more times
- Generate candidate list - Create a candidate list of items to be distilled
- Wait for user confirmation - Present to the user for selection
- Update L2 files - Update the behavior layer based on confirmation
Workflow
Preliminary step: File scan to supplement L1
Before starting the weekly review analysis, you must first call mem-file-scan skill：
调用 mem-file-scan skill
Why this step is needed：
- The AI memory system can only record Claude conversation content
- Files the user edits directly in Obsidian will not be automatically recorded
- File scanning can identify files modified this week and extract important events and decisions
- After supplementing L1 records, the weekly review can analyze more comprehensive behavior patterns
How to execute：
- First call
mem-file-scan - Wait for the file scan to complete and supplement L1 records
- Then continue with the subsequent weekly review analysis process
Step 1: Read this week's L1 records
Read the current month's L1 file:AI_MEMORY/L1_情境层/{年}-{月}.md
Extract this week's records:
- Determine this week's date range (e.g., 12-22 to 12-28)
- Extract all entries from this week
Step 2: Identify recurring patterns
Use Grep to search keywords and count occurrences:
Grep "关键词" AI_MEMORY/L1_情境层/
Criteria for determining recurring patterns：
- The same preference/behavior appears 3 or more times
- The same tool choice appears 3 or more times
- The same expression style appears 3 or more times
Step 3: Generate candidate list
Create a candidate list of items to be distilled (maximum 5):
Format:
📊 本周记忆复盘（{月份}第{X}周）
━━━━━━━━━━━━━━━━━━━━━━━━
🔍 发现 {N} 个重复模式：
1️⃣ "{模式名称}"
- 出现次数：{X}次
- 来源：[日期1] [日期2] [日期3]
- 建议：记录到 L2_行为层/{文件}.md
- 你的选择：[ ] 记录 [ ] 忽略 [ ] 修改后记录
2️⃣ "{模式名称}"
...
Step 4: Wait for user confirmation
Present the candidate list and wait for the user to choose.
The user can:
- Record - Record directly to L2
- Ignore - Skip this pattern
- Record after modification - Modify the content before recording
Step 5: Update L2 files
Based on user confirmation, update the corresponding L2 files:
-
Check whether the file exists
- Exists: append or update
- Does not exist: create a new file
-
Update format
## 📋 {行为/偏好名称}
### 习惯描述
{具体描述}
### 统计信息
- 首次记录：{日期}
- 出现次数：{次数}
- 稳定性：{高/中/低}
- 最后确认：{日期}
### 来源
- {来源1}
- {来源2}
- {来源3}
- Update statistics
- Update INDEX.md statistics
- Update this week's review record
Output example
📊 本周记忆复盘（12月第4周）
━━━━━━━━━━━━━━━━━━━━━━━━
🔍 发现 3 个重复模式：
1️⃣ "处理复杂任务前先用思维导图"
- 出现次数：4次
- 来源：[12-22] [12-24] [12-26] [12-27]
- 建议：记录到 L2_行为层/工作习惯.md
- 你的选择：[ ] 记录 [ ] 忽略 [ ] 修改后记录
2️⃣ "早上9-12点深度工作，不回消息"
- 出现次数：5次
- 来源：[Daily] 5次记录
- 建议：记录到 L2_行为层/工作习惯.md
- 你的选择：[ ] 记录 [ ] 忽略 [ ] 修改后记录
3️⃣ "重要决策会睡眠一晚后再决定"
- 出现次数：3次
- 建议：记录到 L2_行为层/决策习惯.md
- 你的选择：[ ] 记录 [ ] 忽略 [ ] 修改后记录
━━━━━━━━━━━━━━━━━━━━━━━━
等待你的确认...
Post-completion actions
After the review is complete, update AI_MEMORY/Meta/复盘记录.md：
### {年} {月}第{X}周
**复盘日期**：{日期}
**本周关键事件**：
1. {事件1}
2. {事件2}
3. {事件3}
**识别出的行为模式**：
- [模式1] - 出现次数：X次
- [模式2] - 出现次数：X次
**更新到L2**：
- ✅ [文件1]
- ✅ [文件2]
**下周关注**：
1.
2.
3.
Frequently asked questions
Q: What if there are no recurring patterns this week?
A: Output "No obvious recurring patterns were found this week; just keep observing."
Q: How do I determine this week's date range?
A: Use the current date and go back 7 days. For example, if today is 12-27 (Friday), this week is 12-21 to 12-27.
Q: How should the "record after modification" selected by the user be handled?
A: Ask the user how to modify it, then update it according to the user's requirements before recording it to L2.