01 · Skills
What a Skill is
A Skill (技能) gives WorkBuddy a specific tool capability. It packages a set of executable scripts and workflows so the AI can, with your authorisation, perform concrete actions: send email, order food, look up a stock price, read and write files, call a third-party API.
Two words in that sentence deserve a pause:
- Executable scripts — a Skill is not a prompt template, it is real code that runs on your machine
- Your authorisation — it acts as you, and what it can do depends on what you granted
Everything in the security section below follows naturally from those two.
The Skills panel
Open Skills (技能) from the sidebar. Two sections:
| Area | Contents |
|---|---|
| Skill market (技能市场) | Recommended Skills, one-click install |
| Installed (已安装) | Local Skills you can call in conversation |
WorkBuddy ships 20+ Skills covering documents, data reporting, poster design and file organisation, and supports importing community Skills.
Three ways to install
Click Add Skill (添加技能) for three routes:
Upload a Skill
Import a local Skill package: choose Upload, drag or select the file, done. Configuration is automatic afterwards.
Find a Skill
Describe the task and WorkBuddy finds relevant Skills. This is the least effortful route — you do not need to know the name first, just say what you want to do.
I want WorkBuddy to handle web scraping, screenshots and page interaction.
Find suitable Skills first and explain the differences between them.Create a Skill
Describe the task and WorkBuddy builds a Skill for it. Good when nothing off the shelf fits but your process is stable. See Turn a routine into a Skill.
Enable and disable
Installed Skills can be switched off and back on without uninstalling.
Disabled Skills stay in the installed list but do not participate in automatic model calls; re-enabling restores them instantly.
Three mechanics worth knowing:
- State syncs across devices — recorded in your global user configuration, following your account
- The Skill's own files are untouched — the toggle lives in your config and does not modify the Skill's
SKILL.md, so Skills copied in from directories like.agentskeep their original content - Disabled is not uninstalled — disabling pauses model calls; uninstalling deletes the Skill and its toggle record entirely
Only enable what this task needs
The official recommendation, and the reason given is less irrelevant noise and a lower chance of wrong calls.
In practice: with a dozen Skills all switched on, the model starts invoking unrelated capabilities at odd moments — slower, more expensive, more likely to drift. A document task does not need a video downloader enabled.
Use the search box once you have a lot of them. Uninstall removes a Skill from the machine entirely, and batch uninstall is supported.
Using a Skill in a task
Two ways it gets triggered:
Automatically. The model decides which Skill the task needs and calls it. Most of the time you do nothing.
Explicitly. Select an installed Skill in the conversation and it gets called preferentially. Use this when you know which Skill you want and do not want the model guessing.
Security: read this section
Skills are one of the few features in this book where getting it wrong causes real damage.
What it can touch
The official list of information involved (depending on the Skill):
- Account identity
- Contact details
- Your input
- Usage logs
- Third-party platform credentials
- Local file read/write
- Screen and input control
The actual scope is whatever you authorised at install and enable time. The same Skill may use only part of that in a given task.
Four official warnings
Unofficial Skills carry real risk
The official wording: unofficial Skills may contain malicious prompt injection, over-privileged access, or backdoors. Prefer officially recommended Skills.
"Malicious prompt injection" is the most underestimated of those — Skill content enters the model's context, so a carefully constructed Skill can steer the model into doing things you never asked for, in ways you would not spot in the conversation view.
Third-party sharing. Some Skills send your input to third parties — food ordering, stock quotes, IM messages.
Credit consumption. Complex tasks and complex Skills generally cost more credits.
Responsibility. When a Skill calls a third party, modifies local files or runs system commands, it does so as you. Problems or disputes with a third-party service go to that Skill's provider or that service's support.
A pre-install checklist
- Is the source trustworthy? Officially recommended > known community repository > an archive from nowhere.
- Are the requested permissions proportionate? A deck-building Skill asking for screen control is not.
- Have you looked at the script? At minimum, scan for unexpected network calls and hard-coded addresses or credentials.
- Do you know where the data goes?
- Have you tried it small first? Anything involving file deletion, bulk writes, system configuration changes or money should be tested in a scratch directory.
Use Skill Scanner as the first filter
The skill-scanner Skill exists to audit other Skills before installation, surfacing suspicious dependencies, hard-coded configuration and potential risks.
Make it a fixed step: audit first, install second. See Ten free Skills.
Which to install first
If you are starting out, these four have the best return:
| Skill | What it unlocks |
|---|---|
| Office document suite (pdf / docx / pptx / xlsx) | Actually reading and writing office files |
web-search | Time-sensitive questions stop relying on model memory |
skill-scanner | A security review before every other install |
find-skills | When you do not know what to install, describe the need |
All ten free Skills are covered in Ten free Skills.
FAQ
How is a Skill different from an expert?
A Skill is capability — it lets the AI do something. An expert is capability plus experience — an AI role that understands a field. Need it to read PDFs? Install a Skill. Need it to think like a financial analyst? Summon an expert. See Experts and expert teams.
How is a Skill different from a connector?
A connector solves one thing — talking to an external service — with a standard authorisation flow and permission boundary. A Skill is a broader capability wrapper that can include local scripts or external API calls. Technically, connectors come in both MCP + CLI and Skill + CLI forms.
Are the web and desktop Skill libraries shared?
No. Personal Skill libraries are separate. Project-level Skills are managed centrally in the cloud and are identical on both. A cloud task cannot see the desktop personal library, and vice versa.
Does disabling a Skill affect finished tasks?
No. The toggle only affects future model calls.
Next
02 Connectors — let it read and write your mail and documents.