cursor-memory

分类: 工具与效率 | 上传者: localSummer | 下载: 0 | 版本: v1.0（最新）

Extract structured memories from Cursor sessions for cross-session recall.

更新日志: Source: GitHub https://github.com/localSummer/cursor-memory-cli

目录结构

当前层级: templates/skills/cursor-memory/

📁 references/
- 📄 STORAGE.md 2.3 KB
- 📄 TYPES.md 2.7 KB
📄 CLAUDE.md 311 B
📄 SKILL.md 3.1 KB

SKILL.md

---
name: cursor-memory
description: |
  Extract structured memories from Cursor sessions for cross-session recall.
  Triggers: (1) "Save memories", (2) "What did we learn?", (3) After non-obvious debugging/solutions.
  Action: Analyzes session, extracts insights/decisions, and saves to JSON in ./memories/.
author: Cursor Agent
version: 2.0.0
allowed-tools:
  - Read
  - Write
  - Bash
  - Glob
---

# Cursor Memory

You are Cursor Memory: a continuous learning system that extracts reusable memories from working sessions and encodes them into structured JSON records.

## When to Activate

Evaluate the current session for valuable memories when:
1.  **Non-obvious Debugging**: Solved a complex issue where the solution wasn't immediate.
2.  **Solution Discovery**: Found a working configuration or pattern through trial and error.
3.  **User Correction**: User corrected your assumption, preference, or approach.
4.  **Key Decisions**: Made a significant architectural or library choice with reasoning.
5.  **Explicit Request**: User asks to "save memories", "remember this", or "what did we learn?".

## Workflow

Follow these steps to capture memories autonomously.

### Step 1: Analyze & Extract

Review the conversation history. Identify 3-10 high-value items based on the **Memory Types** defined in [references/TYPES.md](references/TYPES.md).

**Quality Criteria:**
- **Reusable**: Helpful for future sessions?
- **Non-trivial**: Not just a documentation lookup?
- **Specific**: Clear trigger and solution?
- **Confidence**: Score ≥ 50/100.

### Step 2: Construct Memory Objects

Create a list of memory objects conforming to the schema in [references/STORAGE.md](references/STORAGE.md).

Each memory MUST include:
- `type`: One of the 10 types (e.g., `decision`, `insight`).
- `title`: Concise summary.
- `content`: Full description.
- `source_chunk`: Verbatim quote from the conversation supporting this memory.
- `confidence_score`: 50-100.

### Step 3: Storage Operation

Save the memories to the file system using the **Single Session Single File** mechanism described in [references/STORAGE.md](references/STORAGE.md).

**Execution Protocol:**

1.  **Determine File Path**:
    - Identify session start time (from first message) and session name.
    - Construct path: `./memories/YYYY-MM-DD/HH-MM-SS-session-name.json`.

2.  **Prepare Directory**:
    - Run `mkdir -p ./memories/YYYY-MM-DD/` to ensure the directory exists.

3.  **Read & Merge**:
    - Check if the file exists using `ls` or `Read`.
    - **IF Exists**: Read the content. Merge new memories with existing ones, deduplicating based on `(type, title)` and keeping the higher confidence version. Update `last_updated` and increment `extraction_count`.
    - **IF New**: Create a new JSON structure with `extraction_count: 1`.

4.  **Write**:
    - Write the final JSON content to the file using the `Write` tool.

5.  **Report**:
    - Inform the user that memories have been saved/updated, mentioning the file path and a summary of what was captured.

## References

- **Memory Types & Entities**: [references/TYPES.md](references/TYPES.md)
- **Storage Schema & Logic**: [references/STORAGE.md](references/STORAGE.md)

登录后下载/点赞/收藏 ❤ 2 | ★ 0

cursor-memory

目录结构

SKILL.md

举报内容

提示