一键在 Manus 中运行任何 Skill

onboarding

星标1

分支0

更新时间2026年6月25日 15:36

First-time setup for a fresh MUM instance. Use when the knowledge/ folder has no core content yet, or when the user asks to (re)onboard. Interviews the user and writes the initial core knowledge files. Do not use for ongoing knowledge updates during normal work — that's the self-evolve protocol.

安装

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。

在 Manus 中运行

来源

sinedied

sinedied/mum

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Onboarding

You set up a brand-new MUM instance by interviewing the user and writing the first core knowledge files into knowledge/. Keep it light — capture the essentials, not their whole life. MUM fills in the rest over time via self-evolution.

How to run it

Interview the user one question at a time — ask, wait for the answer, then ask the next. Keep questions short. Skip anything they'd rather not share. Aim to fill these two core areas:

Who you are & how you work → knowledge/user.md Name, what they do, how they work, and recurring context that helps MUM be useful. (Repeatable procedures become skills, not notes here.)
How MUM should behave → knowledge/mum.md Tone, level of detail, defaults, dos and don'ts, how they like an assistant to work.

If something they share clearly belongs to a life/work domain (not core context), create a domain note for it following the knowledge skill's convention.

Steps

Confirm the repo is private before collecting any personal details. If it's public or you can't tell, stop and ask the user to make it private (or confirm) before continuing.
Briefly explain what onboarding will do, then start the interview.
As answers come in, write concise notes into the files above. One topic per file, facts over prose. Create domain subfolders only if the conversation clearly calls for one.
When both core areas are covered, summarize what you captured and confirm.
Replace the README. This is a fresh instance, so README.md is still the template's one. Rewrite it README.md with a short, private, personal note to the owner — what this repo is and how to use it.
Commit the new knowledge/ files and the rewritten README.md to main.

Gotchas

Re-onboarding: if knowledge/ already has core content, confirm before overwriting — update files in place rather than clobbering them.
Don't over-interview or pre-create empty domain folders; capture the essentials and let self-evolve fill the gaps as they come up.

Rules

Keep it lean — capture what's actually useful, don't pad files.
After onboarding, follow the self-evolution protocol to keep knowledge current.

同仓库更多 Skills

同仓库

knowledge

sinedied/mum

Find and maintain the user's knowledge notes. Use to locate which knowledge notes are relevant to a task (run the map tool for progressive disclosure), or when creating, updating, or organizing a domain note. Authority on MUM's knowledge-note format. Do not use for editing the constitution (AGENTS.md) or building skills (see skill-builder).

2026-06-251

self-evolve

sinedied/mum

An explicit pass to improve MUM. Use when the user asks MUM to "evolve", "learn from this", or tidy itself up. Captures durable learnings into knowledge and proposes new or refined skills. Do not auto-run after ordinary work (capture confirmed learnings inline instead), for first-time setup (use the onboarding skill), or for jotting a single fact mid-task (just write it).

2026-06-251

skill-builder

sinedied/mum

Create, refine, or retire MUM skills. Use when the user has a repeatable workflow worth capturing as a skill, wants to improve or fix an existing skill, or wants to remove one. This skill is the authority on MUM's skill format and authoring best practices. Do not use for one-off tasks the agent already handles well.

2026-06-241

name	onboarding
description	First-time setup for a fresh MUM instance. Use when the knowledge/ folder has no core content yet, or when the user asks to (re)onboard. Interviews the user and writes the initial core knowledge files. Do not use for ongoing knowledge updates during normal work — that's the self-evolve protocol.

Onboarding

How to run it

Interview the user one question at a time — ask, wait for the answer, then ask the next. Keep questions short. Skip anything they'd rather not share. Aim to fill these two core areas:

Who you are & how you work → knowledge/user.md Name, what they do, how they work, and recurring context that helps MUM be useful. (Repeatable procedures become skills, not notes here.)
How MUM should behave → knowledge/mum.md Tone, level of detail, defaults, dos and don'ts, how they like an assistant to work.

If something they share clearly belongs to a life/work domain (not core context), create a domain note for it following the knowledge skill's convention.

Steps

Confirm the repo is private before collecting any personal details. If it's public or you can't tell, stop and ask the user to make it private (or confirm) before continuing.
Briefly explain what onboarding will do, then start the interview.
As answers come in, write concise notes into the files above. One topic per file, facts over prose. Create domain subfolders only if the conversation clearly calls for one.
When both core areas are covered, summarize what you captured and confirm.
Replace the README. This is a fresh instance, so README.md is still the template's one. Rewrite it README.md with a short, private, personal note to the owner — what this repo is and how to use it.
Commit the new knowledge/ files and the rewritten README.md to main.

Gotchas

Re-onboarding: if knowledge/ already has core content, confirm before overwriting — update files in place rather than clobbering them.
Don't over-interview or pre-create empty domain folders; capture the essentials and let self-evolve fill the gaps as they come up.

Rules

Keep it lean — capture what's actually useful, don't pad files.
After onboarding, follow the self-evolution protocol to keep knowledge current.