Go deeper: from working to reliable
After a few successful tasks, most people get stuck at the same point: sometimes it is great, sometimes it is not quite right.
That is not luck. The difference is whether you have turned "it worked once" into "it works every time" — by giving it fixed capabilities, fixed context and fixed boundaries, and then letting the repetitive parts run themselves.
That is what this chapter is about.
The eleven pages
| # | Page | What it solves |
|---|---|---|
| 01 | Skills | It cannot do a class of work; how to vet third-party Skills |
| 02 | Connectors | Let it read and write mail, Tencent Docs, Lexiang, Meeting |
| 03 | MCP and plugins | Connect any external service; bundle capabilities together |
| 04 | Experts and expert teams | Give it a professional identity; multi-role delivery |
| 05 | Projects and teamwork | Shared context, standards and material across a team |
| 06 | The Library: MD / CSV / HTML | Stop artifacts scattering; separate data from presentation |
| 07 | Automations on a schedule | Let recurring work run itself |
| 08 | Assistant: drive your PC by phone | Send tasks when you are away from your desk |
| 09 | Memory | Stop re-explaining who you are every time |
| 10 | Permission modes and the sandbox | It touches your files — this is how you keep control |
| 11 | Git and parallel worktrees | Several tasks changing code without polluting each other |
A capability map
There are enough extension concepts to get confusing. This table settles it:
| Concept | What it is | The question it answers |
|---|---|---|
| Skill (技能) | Tool capability | What WorkBuddy can do |
| Connector (连接器) | A bridge to an external service | Which systems it can reach |
| MCP | Standard protocol for wiring in external tools | How connecting is done generically |
| Plugin (插件) | Skills / MCP / slash commands / hooks bundled together | Installing a related set at once |
| Expert (专家) | Capability plus experience — an AI role | Who does it for me |
| Expert team (专家团) | Several experts plus a coordination process | How a complex job gets divided |
| Inspiration (灵感) | Finished work made with those capabilities | What people actually built |
| Project (项目) | Shared instructions, assets and configuration | How a team shares one context |
| Library (资料库) | Where humans and agents both keep output | Where finished work lives and gets reused |
| Automation (自动化) | A task triggered by a schedule | When it runs without me |
The short version: a Skill is capability, an expert is capability plus experience, an expert team is several experts plus a process, and Inspiration is the portfolio built from all of it.
Suggested order
Do not read by number. Pick whatever hurts most right now:
"It cannot even read my files" → 01 Skills, and install the office document suite.
"I re-explain the background every single time" → 09 Memory; for teams, 05 Projects.
"I run the same job every week" → 07 Automations, after freezing the routine into a Skill.
"I am worried it will wreck my files" → jump straight to 10 Permission modes and the sandbox. Read this one first.
"I want it handling mail / documents / meetings" → 02 Connectors.
"How do we organise this as a team" → 05 Projects plus 06 The Library.
Three judgements that run through this chapter
One: more capability means more wrong calls
Once you have a dozen Skills installed, the common failure is not "it cannot", it is "it called the wrong one". Only enable what the current task needs — advice the official docs give more than once.
Two: third-party extensions are executable code
Skills, MCP servers and plugins all act as you, on your machine. Unofficial sources can carry malicious prompt injection, over-broad access, or backdoors.
Vet before installing: the source, the permissions requested, the script contents, where data goes. The Skill Scanner Skill exists for exactly this — see Ten free Skills.
Three: automations run unattended
Scheduled tasks execute when you are not there. Any prompt involving file writes, deletion or outbound sending should run at low frequency first until you have confirmed both the understanding and the execution, before you scale it up.
Start with 01 Skills.