AIToday for AI

Guides · Beginner

WorkBuddy Practical: 04 - Quick Start for the First Task

About 30 min · No coding required

WorkBuddy In Action

View every guide in this series.

View series

Creating a WorkBuddy Task and Setting Workspaces

  1. Click "New Task".
  2. Select or create an isolated workspace directory.
  3. Select the execution mode (Craft by default).
  4. Specify the large language model.
  5. Input the goal prompt: "Analyze 'E-commerce Sales.xlsx' and compile a summary PowerPoint."
  6. Attach custom Skills, Experts, or Knowledge Bases (optional).
  7. Submit the prompt, and monitor the generated plan, tool logs, and file changes.
  8. Verify and preview the compiled presentation in the Result Area.

You can open the output file locally, upload it to the cloud, or share it. Redact sensitive data before sharing.

How to Structure a Task Specification Prompt

A perfect prompt should answer:

ElementQuestion to Answer
GoalWhat is the target deliverable to produce?
InputWhich files, paths, or URLs should be accessed?
ActionShould the agent analyze, summarize, convert, or generate?
ConstraintsWhat are the limits, formatting rules, or boundaries?
OutputWhere should the files be exported to?
ValidationWhat standards determine a successful completion?

Example A: File Sorting

Goal: Sort the practice files in the input directory.
Input: Process files only within the active workspace.
Action: Detect file types and propose a cataloging plan.
Constraints: Do not delete or overwrite any files. Append suffixes if names conflict.
Output: Output inventory.xlsx and proposed-actions.md.
Validation: File counts must match, actions must be traceable. Do not move files until proposed-actions.md is approved.

Example B: Meeting Summary

Extract action items from input/meeting.txt.
Required: Key decisions, follow-ups, owners, deadlines, open questions.
Constraints: Mark missing owners as "TBD". Do not make up facts.
Output: output/meeting-notes.md and output/action-items.xlsx.
Validation: Every decision must map back to the transcript.

Example C: Document to Presentation Conversion

Convert input/project-report.docx into an internal presentation under 10 slides.
Audience: Department Head. Presentation Time: 8 minutes.
Constraints: Preserve raw facts and metrics. Apply the color palette in reference/brand-guide.pdf.
Output: output/report_v1.pptx and slide list markdown.
Validation: One core takeaway per slide.

Tencent CodeBuddy Engine and Environment Configurations

WorkBuddy runs on the same developer agent engine as Tencent Cloud CodeBuddy ( codebuddy.cn ). During initialization, the client creates a hidden .workbuddy folder in your user root directory to store configuration states.

Offline Local Model (Ollama) Integration Steps

For confidential corporate data, you can configure local models to run entirely offline:

  1. Download and install Ollama.
  2. Pull required models: ollama run deepseek-coder.
  3. Open WorkBuddy's model settings panel, direct the API path to http://localhost:11434, and run a connection check.

Troubleshooting (FAQ)

  • Python Not Found Error: WorkBuddy sandboxes rely on a local Python runtime. Ensure Python is installed and added to your system PATH variable, or specify the path to python.exe manually in client settings.
  • Task Planning Loop Alert: If generated scripts throw runtime errors, the agent might get stuck in an automated correction loop. Click "Stop" to interrupt the task, decompose the prompt into simpler steps, or switch to a model with higher reasoning capabilities.

References

Part of this series

WorkBuddy In Action

View all guides in this series →

Next step

See related AI updates

Keep exploring from this piece.

Browse the latest AI content