TutorialsFirst 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.
This lesson has one goal: see whether it can run the whole thing by itself without cutting corners. The case is a daily AI industry briefing. You only do three things — prepare the space, state the requirements clearly, verify the output.
Step 0 · Install WorkBuddy and create a workspace
The class uses Tencent WorkBuddy on desktop:
Download and sign in: open workbuddy.cn, download the desktop app, install and sign in.
Create a work folder: make a new folder named AI-Briefing. The Agent can only read and write inside this space — keep it away from other directories.
Start a task: create a new task in WorkBuddy and select that folder as the workspace.
Why a dedicated folder matters. This is the step most often skipped and the one most likely to cause trouble. The workspace is its area of movement: choose too wide a space — your desktop, or an entire drive — and it may touch files you never meant to hand over. One task, one folder, archive afterwards.
Step 1 · Write the requirements into a file
This is the most valuable design in the handout: do not stuff every requirement into the chat box. Write them into a file inside the workspace first.
Create requirements.md in the workspace:
# AI industry briefing requirements
Project: daily AI industry briefing
Audience:
General office workers and business managers.
Focus areas:
- AI Agents
- AI office automation
- Enterprise AI adoption
- New features in common AI tools
Content requirements:
1. Only collect real, reliable public news.
2. Prefer official announcements and trusted media.
3. Keep the original headline and source link for every item.
4. Explain in language ordinary readers understand.
5. State what each item means for everyday work.
6. Produce one briefing and one source list each time.
Output requirements:
Keep the Word briefing to 1-2 pages.
The Excel sheet includes title, date, source, link and summary fields.Three benefits: the prompt can stay short; requirements can be edited and re-read at any time; the file is reusable for the next similar task. Requirements are context — they belong in material, not in chat history.
Step 2 · Paste the prompt and let it start
Copy this into the chat box:
You are a professional industry research assistant.
Please produce an industry news briefing.
Topic: 【AI Agents and AI office tools】
Time range: 【last 7 days】
Audience: 【general office workers and business managers】
Plan and carry out the following tasks yourself:
1. Search the web for the latest public news on this topic.
2. Filter down to a few of the most valuable items.
3. For each item, record the title, date, source,
key points and practical value.
4. Generate a Word industry news briefing.
5. Generate an attractive HTML version with charts.
6. Save both files to the current workspace.
Briefing requirements:
- Concise, ideally 1-2 pages.
- Plain language that is easy to understand.
- For each item, explain how it helps everyday AI use.
The Excel sheet includes:
title, publish date, source,
original link, key points.
Notes:
- Use only real, verifiable news.
- Do not invent dates, numbers or sources.
- Do not present unverifiable information as fact.
- Do not delete or overwrite existing files.
- If a format is unavailable, Markdown and CSV are acceptable substitutes.
When finished, check that the files were created
and tell me where they are saved.
If key information is missing, ask me up to 3 questions first;
if it is sufficient, start directly.Expect three outputs: a Word briefing, an HTML version with charts, and an Excel source list. Steps 4 and 5 ask for two briefing formats, and the Excel fields are requested separately — if you only want two files, delete the step you do not need before sending. Cheaper than asking it to fix afterwards.
Step 3 · Four things to watch while it runs
Do not walk away. Watching these four live is worth more than reading ten pages of documentation:
| What to watch | Normal | Warning sign |
|---|---|---|
| Planning | Lists a plan or steps, then works through them | Prints an article immediately, no actions |
| Tool use | Clearly searching, reading files, writing files | Everything "known" with no source at all |
| Files written | New files appear in the workspace and open | Says "generated" only in the chat |
| Location reported | Ends with file name and path | Stops without telling you where |
Step 4 · Verify: three checks, hard to soft
| Check | How | Failing sign |
|---|---|---|
| File exists | Open the folder and double-click each file | Will not open, empty, or placeholder text |
| Content verifiable | Pick two items and compare date and headline with the source | Dates mismatch, links dead, headline rewritten |
| Source traceable | Check every item carries an original link | "According to some media report", no link |
A generated page is not a verified one. The most common failure in web research is stale references, or merging several similar stories into one. So during verification, open at least two sources and actually read them. Skip this and wrapping it into a Skill and scheduling it only multiplies the error.
Step 5 · When it fails, fix by symptom
| Symptom | Cause | What to say |
|---|---|---|
| Output only in chat, no file | The deliverables section was too vague | Add: write the result as a .docx file in the current workspace and give me the full path |
| File created but content invented | Constraints too loose, sources unlimited | Add: every item needs a working original link; delete anything unverifiable; do not fill gaps with guesses |
| Format far from the requirement | Format described too broadly | Specify: headline, date, source, three-sentence summary, value for work; one item per paragraph; two pages maximum |
Change one thing at a time and rerun. Do not stack five new requirements at once — if something breaks, you will not know which one caused it.
The real takeaway
Not the briefing. The takeaway is that you saw with your own eyes: it can actually do a complete piece of work by itself.
Once that works, the next question arrives by itself: if you want a briefing for another industry, do you paste that whole prompt again?
No. Next lesson turns it into a Skill.
From Chat to Work: What an Agent Really Is
Ordinary AI answers questions; an Agent finishes the job. Learn the four core concepts, eight high-frequency scenarios, and the five-part prompt formula.
Wrap 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.