TutorialsWrap It Into a Skill, Then Schedule It
Turn the proven method into a reusable skill, test it on a new topic, then schedule it to run daily — while holding three quality red lines.
Last lesson's result is worth one run. This lesson makes it reusable and self-starting. Two steps: wrap it into a Skill, then hand it to a scheduled task. After that, only verification is left to you.
What a Skill is
Wrap a proven method — its steps and its rules — into a capability the Agent can call again and again.
Industry briefings work for AI, but also for education, e-commerce and finance. The topics differ; the search, filtering, organization and file generation do not. Extract that common method and store it, and next time you only say: use the Industry Briefing Generator for education.
Do not confuse three things:
| Concept | The question it answers | Example |
|---|---|---|
| Prompt | What to do this time | Produce an AI industry briefing |
| Skill | How this kind of job is always done | New topic, same standards and format |
| Scheduled task | When it starts by itself | Every day at 9:00, AI industry |
Step 1 · Turn it into a Skill
Path: open 专家·技能·连接器 on the left of WorkBuddy → Skills → Add skill → Create skill.
Prerequisite: it must be able to reach the workspace and result files from the first run. It summarizes the method from those results; without access it will invent a process.
Then paste this prompt:
Please create a Skill named
Industry Briefing Generator.
We have just completed one
industry news search and briefing task.
Refer to the result files in the current workspace,
summarize the reusable working method,
and wrap it into a Skill.
The Skill needs these capabilities:
1. The user only supplies topic,
time range and target audience.
2. Automatically search relevant public material
and select 3-5 important items.
3. Record headline, date, source, key points
and practical value to a consistent standard.
4. Generate the industry briefing
and the matching Excel source list.
5. Save the final files to the specified location.
Skill requirements:
- No restriction on industry or topic.
- Consistent, readable structure.
- Keep the verifiable original sources.
- Do not fabricate facts or data.
- Do not delete or overwrite existing files by default.
- Use substitute formats when a format is unavailable.
Follow the Skill specification WorkBuddy supports and
create a skill that can actually be recognized,
installed and called — not just a prompt printed out.
After creating it, check that the skill files are complete
and tell me how to invoke it.
If necessary information is missing,
ask me up to 3 questions before starting.Keep the line create a skill that can actually be recognized, installed and called — not just a prompt printed out. Without it, you will likely get a "skill description" printed into the chat box: convincing to read, impossible to install or invoke.
Step 2 · Test it on a new topic
Testing is mandatory, and it must use a topic you have not done before — otherwise you cannot tell whether it follows the Skill or simply reworded the previous output.
Start a new task, enable the Skill, and send:
Use the Industry Briefing Generator Skill
to produce a briefing on
【AI video generation tools】 for the last 7 days.
The audience is ordinary content creators.
Generate a Word briefing and an Excel source list,
and save them to the current workspace.Three checks, easy to hard:
| Check | What to look for |
|---|---|
| Was it invoked | The task shows the Skill loaded, not treated as ordinary chat |
| Did it follow the process | Search, select 3-5 items, consistent fields, both file types — all present |
| Does it survive a new topic | Topic changed, format and standards unchanged, content genuinely new |
Similar-looking files do not prove the Skill ran. It can skip the Skill entirely and still produce a similar briefing. Look specifically for the skill-loading step in the execution trace. Without that step, you tested nothing.
Step 3 · Schedule it to run every day
Final step: let it start by itself. Create a new task and send:
Please configure a scheduled task.
Task name:
Daily AI industry briefing
Run time:
Every day at 9:00 AM
Skill to use:
Industry Briefing Generator
Topic:
【AI Agents, AI office tools】
Time range:
【last 24 hours】
Audience:
【general office workers and business managers】
After creating it, show the run time,
the Skill used and the save location.
Run it once immediately after confirming,
to check that the Skill is invoked and the output is produced.
If the Skill cannot be bound or invoked,
say so explicitly — do not pretend it succeeded.Then hit Run now and check four things:
- Is the task actually created, at 9:00 daily;
- Is it bound to the Skill you built in step 1, or some other one;
- Did the test run really produce new files;
- Are they in the expected folder, or scattered somewhere else.
Local or cloud: know the trade-off
| Mode | Benefit | Cost |
|---|---|---|
| Local scheduled task | Reads and writes files on your own computer — which this case needs | Depends on the computer being on and the client running |
| Cloud execution | Runs even when the computer is off | Cannot read local files; only what the cloud can reach |
This case writes local files, so local scheduling comes first. If your job is "collect information online and send it somewhere", cloud fits better. If it is "organize a batch of files on my machine", only local works.
Three quality red lines
Once automation runs, mistakes scale too. Three rules:
One: verify numbers, dates and sources yourself. Agents fabricate confidently, and web research especially tends to cite stale content or merge several stories into one. Check each scheduled run, do not save them up for a month.
Two: desensitize confidential material. Client data, trade secrets and internal files should not go to an AI without permission. When granting access, check which folders and systems it can reach. The space you give it is the space it can touch.
Three: output is not final, and high-risk actions keep a human in the loop. Deleting files, sending messages, changing system settings — do not fully automate these. Test-run a scheduled task first, then spot-check results regularly.
One sentence for the boundary: an Agent can work on its own, but judging and deciding remain yours. It saves you the hands-on time; it cannot take the consequence for you.
The loop closes here
Agent finishes the task → Skill reuses the method → scheduled task runs it automatically.
After these three steps you do not just have a case; you have a repeatable approach. For the next job, walk the same path: get it done once, confirm the result is trustworthy, wrap it, then schedule it.
The assignment asks you to do exactly that with a job of your own.
First Run: Let the Agent Finish a Task on Its Own
Set up a workspace, write the requirements into a file, paste the prompt, and watch it search, organize and save by itself — then verify the result properly.
Let an Agent Run a Real Job for You
Take a task of your own through the full path — finish it once, wrap it into a Skill, schedule it — and submit all four pieces.