Note

Note Skill — RAL Recording Layer

Lightweight personal record keeper. Capture moments that feel significant, gradually add meaning — without turning it into a diary burden.

RAL Model

Record (捕获) → Amplify (放大) → Layer (分层)
  10s              2min             ongoing

• Record: one-line event. Don't overthink. 10 seconds.
• Amplify: add feeling + what this reveals about you. 2 minutes max. Can be done later.
• Layer: active (current experiment) | accumulating (pattern forming) | archived (saved, not active)

When to Use

Integrity on Every File Open

Before any read/write of state/records.jsonl, run integrity checks per references/reflection-protocol.md (State Integrity section):

1. Parse each line as JSON — skip unparseable, count corrupt
2. Check required fields (id, timestamp) — repair if possible
3. Detect duplicate IDs — keep first occurrence
4. Report: "records.jsonl: [N] records, [M] corrupt skipped"
5. If >50% corrupt, warn: "记录文件严重损坏，建议手动检查 state/records.jsonl。"

When writing: serialize first, verify JSON is valid, append with trailing \n. Verify line count after write.

Record Schema

{
  "id": "uuid",
  "timestamp": "ISO 8601",
  "layer": "active|accumulating|archived",
  "type": "event|daily_review|weekly_review",
  "event": "one-line description of what happened",
  "feeling": "excitement|frustration|pride|anxiety|curiosity|satisfaction|disappointment|neutral|null",
  "amplification": "what this reveals — null until amplify step",
  "value_tags": ["autonomy", "creation", ...],
  "domain_tags": ["coding", "design", ...],
  "energy": "energizing|draining|neutral|null",
  "linked_records": [],
  "linked_reflection_id": null,
  "processed_at": null
}

Commands

1. Capture (/note capture)

When user says "记一下" or "take a note":

Step 1: Ask for the event

"今天发生了什么让你有波动的事？一句话就行。"

If user provides context in the prompt (e.g., "take a note: just had a heated debate about API design"), use it directly.

Step 2: Quick feeling (optional)

"什么感受？[兴奋/沮丧/好奇/焦虑/满足/无/跳过]"

Step 3: Save

Write the record to state/records.jsonl.

"已记录。ID: [id] — 有空时可以 /note amplify 展开看看这条记录说明什么。"

At most 3 exchanges. This is not an interview.

2. Amplify (/note amplify)

When user says "展开记录" or "amplify":

Step 1: List unamplified records

"你有 [N] 条未展开的记录："

#	日期	事件	感受

"要展开哪一条？（输入编号，或 'all' 逐条来）"

Step 2: Amplify the selected record

事件: [event]

1. "为什么这件事让你产生这种感受？"
2. "这说明了你的什么特点或在意什么？"

Also invite value and domain tags (see Tag Catalog in references/reflection-protocol.md).

Step 3: Layer

"这条记录的状态？"

• 活跃 (active) — 正在进行的实验/信号
• 累积 (accumulating) — 留待以后联结（默认）
• 归档 (archived) — 记下来就够了

"已展开。ID: [id] — 下次 /reflect 会自动加载这条记录。"

3. Daily Review (/note daily)

30-second end-of-day pulse. ONE question. Not a mini-reflect.

Pulse question:

"今天哪个瞬间让你觉得'这就是我'——让你感到能量、投入、或者对劲的瞬间？"

Optional shadow: "今天有什么让你感到消耗或不对劲的？（可选）"

Write a type: "daily_review" record with the pulse + shadow.

"已记下。今天的方向信号: [pulse]."

4. Weekly Review (/note weekly)

Gather this week's records (last 7 days). If < 3: "本周只有 [N] 条，等至少 3 条再来。"

"本周记录了 [N] 件事：[table]" "这些事件之间有什么关联吗？试着画个箭头——这些事件指向什么方向？" "用一句话总结——本周这些事件告诉你什么？"

Write a type: "weekly_review" record with connections + direction.

"已保存。方向箭头：[one-line insight]。这些联结会在下次 /reflect 时一并分析。"

5. List (/note list)

Show compact table of all records, grouped by layer. Count active/accumulating/archived.

6. Default (/note with no args)

Show status + suggest next action based on time of day and record state.

Integration with /reflect

Records are passive until /reflect runs. Reflect loads unprocessed records as additional signal sources for all three Lens agents. After processing, processed_at is set and linked_reflection_id points to the reflection.

Key Files