菜单
Import durable knowledge from URLs, folders, or files into pi-memory-md. Use when the user asks to preserve external content as memory.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Digest recent tape anchors and session context into curated pi-memory-md memories when tape mode is enabled. Use only with tape mode enabled when the user asks to summarize recent work, extract learnings, preserve decisions, convert tape/session history into memory, or review recent intent for durable memory updates.
Search and retrieve information from pi-memory-md memory files. Use when you need to search memory.
Initialize memory repository - clone git repo and create directory structure. Use when you need to set up pi-memory-md for the first time or initalize project's memory files.
Create or update pi-memory-md memory files using the native write/edit tools plus the bundled template script. Use whenever writing, creating, or updating memory files.
Git synchronization operations for pi-memory-md repository. Use when you need to sync memory.
| name | memory-import |
| description | Import durable knowledge from URLs, folders, or files into pi-memory-md. Use when the user asks to preserve external content as memory. |
Use this skill to curate durable memory from external sources. Do not treat import as file copying; treat it as selecting long-lived context worth remembering.
Analyze first, ask when focus is unclear, then generate confirmed memories using the memory-write skill.
memory-write skill, using confirmed paths, descriptions, tags, source refs, and concise content.memory_sync unless the user requested sync.npx defuddle for extracting readable page content; do not require global installation.npx defuddle parse <url> --markdownnpx defuddle parse <url> --json.git, node_modules, build outputs, hidden files, logs, caches, and generated artifacts..env files.Ask a short focus question when the source contains multiple possible memory areas. Present concrete options.
Example:
I found these possible memory areas:
1. Architecture and module responsibilities
2. Development/test workflow
3. Tool/API usage
4. Project conventions
5. User-facing docs
Which should become long-term memory, and what should I ignore?
Do not ask the focus question if the user already specified the scope clearly. Still confirm the final folder/file path unless the user already gave an explicit path.
When asking for the final output path, provide 3 concise alternatives based on the detected topic and existing memory structure:
Where should I save the generated memory?
1. [folder]/[file].md
2. [folder]/[folder]/[file].md
3. [folder]/[file].md
Pick one, or provide a custom folder/file path.
description and tags for every memory.description must be one concise sentence explaining why this memory is useful for future retrieval or agent behavior.tags must be an array of 2-5 separate, lowercase, short, reusable strings; never use a comma-separated tag string. example: tags: ["defuddle", "web-scraping", "cli"].misc, notes, or imported alone.core/project/<topic>.md or notes/<slug>.md.memory-write skill for each memory, with description as a string and tags as a string array.A good imported memory is:
Avoid importing:
.env contentmemory-write: Create or update pi-memory-md memory files with proper metadata and path handling.memory-sync: Sync memory repository changes after imports when the user requests or confirms synchronization.