兆鹏于

Posted on Jul 5

AI Skill Practical Training Course — Student Step-by-Step Guide

Target Audience: Students enrolled in the AI Skills Competition

Course Goal: Complete each challenge and take home a ready-to-run Skill

Prerequisite: TeleAgent desktop application installed

Pre-Course Preparation (5 minutes)

1. Verify Your TeleAgent Is Ready

Open TeleAgent desktop and confirm:

Check Item	Action	Expected Result
Version	Click top-right gear icon → About	Version >= 1.2.0
Skill Directory	Left sidebar → Skills	Can see "My Skills" list
Network	Send a test message	Receives normal response

2. Download Course Resource Pack

Download from the course link:

skill-templates.zip — 8 in-class Skill templates
demo-data.zip — Demo data (sanitized)
cheat-sheet.pdf — Quick reference card (one page)

3. Import Your First Skill (Pre-class Exercise)

Steps:

Extract skill-templates.zip
Open TeleAgent → Skills → Import Skill
Select the l1-01-info-extractor-archiver folder
Click "Import" → See "Import Successful" message

If import fails: Check if the folder contains a skill.yaml file — this is the Skill's "ID card"

Challenge Roadmap

Challenge 1: Info Extractor & Archiver  →  Challenge 2: Daily Report Generator
[30 min]                                    [30 min]
Difficulty: *

Challenge 3: Material Audit Assistant   →  Challenge 4: Knowledge Base Q&A
[30 min]                                    [30 min]
Difficulty: **

Challenge 5: Permission Self-Check      →  Challenge 6: Message Linker
[30 min]                                    [30 min]
Difficulty: **

═════════════════════════════════
Challenge 7: Unified Entry Gateway (L2) →  Challenge 8: Multi-Agent Complaint Flow
[45 min]                                    [45 min]
Difficulty: ***

Completion Reward: 8 Skills + 1 Skill Certification

Challenge 1: Information Extraction & Archiving (30 min)

Challenge Goal

Learn to create a Skill that extracts structured information from messy chat records or OCR text and auto-archives it.

Real-World Application

Inspection photos from work group → Auto extract "time, location, issue, handler"
Customer says "My broadband is broken" → Auto extract "fault type, address, contact"
Receive Excel table → Auto extract key fields and organize

Challenge Steps

Step 1: Import Template (5 min)

Path: TeleAgent → Skills → Import Skill → Select l1-01-info-extractor-archiver

You should see:

Skill name: Info Extractor & Archiver
Version: 1.0.0
Level: L1 (Basic)

Step 2: Understand Parameter Configuration (10 min)

Open the Skill config panel, you'll see three parameters:

Parameter	Type	Required	Default	Purpose
input_content	Text	Yes	-	Paste chat records, OCR results, or file summaries
output_format	Select	No	JSON	Output format: JSON/Markdown/Table
strict_mode	Toggle	No	Off	Strict mode: error on missing fields

Try it:

Paste this simulated chat record in input_content:

   [Work Group - Nanchang Maintenance]
   Zhang San 09:15: Optical cable cut on Beijing East Road, Qingshanhu District, affecting 3 neighborhoods
   Li Si 09:20: Sent Master Wang to the site, estimated 2-hour recovery
   Wang Wu 09:45: Site photos sent, excavator construction caused it, police notified

output_format: "JSON"
strict_mode: "Off"
Click "Run"

Expected output:

{
  "status": "success",
  "data": {
    "time": "09:15",
    "location": "Beijing East Road, Qingshanhu District",
    "event": "Optical cable cut",
    "impact": "3 neighborhoods",
    "handler": "Master Wang",
    "estimated_recovery": "2 hours",
    "site_status": "Excavator construction, police notified"
  },
  "missing_fields": [],
  "confidence": 0.92
}

Success marker: See the JSON output above — Skill correctly extracted information

Step 3: Modify for Your Own Scenario (10 min)

Customize extraction fields:

time → keep
location → keep (or change to "area")
event → change to "fault_type"
impact → change to "affected_users"
handler → keep
estimated_recovery → change to "resolution_deadline"

Step 4: Connect to Feishu Table (5 min, optional)

Configure auto-archiving to a Feishu multi-dimensional table:

Configure Feishu MCP connector in TeleAgent
Add output configuration with field mapping

Challenge 2: Daily/Weekly Report Auto-Generation (30 min)

Challenge Goal

Create a Skill that reads data sources, auto-calculates metrics, and generates formatted reports.

Real-World Application

Daily fault ticket summary → "Broadband Fault Daily Report"
Weekly sales data → "Business Weekly Report"
Monthly complaint data → "Service Quality Monthly Report"

Challenge Steps

Step 1: Import Template

Import l1-02-daily-weekly-report-generator

Step 2: Configure Data Source (10 min)

Try with simulated data:

Paste fault ticket statistics
Select "Daily Report" template
Run and see formatted output with key metrics, TOP3 issues, and recommendations

Step 3: Connect Real Data Source (10 min)

Three options:

Option A: Connect Excel file
Option B: Connect Feishu table
Option C: Manual paste (simplest)

Step 4: Customize Report Template (5 min)

Edit the Skill → Find "report_templates" → Copy and modify the daily report template.

Challenge 3: Material Audit Assistant (30 min)

Challenge Goal

Create a Skill that auto-checks submitted materials for completeness and format compliance.

Real-World Application

Review project proposals → Check for "budget table, feasibility report, risk assessment"
Review contract attachments → Check for "signature page, stamp page, attachment list"
Review expense claims → Check for "receipts, approval forms, detail sheets"

Key Concept: Audit Rule Table

Rule Name	Check Content	Required	Format Requirement
Application Form	Is form present	Yes	File exists
Budget Detail	Is budget table present	Yes	Excel format, amount column exists
Feasibility Report	Is report present	Yes	Word/PDF, >= 5 pages
Risk Assessment	Is assessment present	No	Bonus if present, warning if absent
Signature Page	Is signature present	Yes	Image/PDF, contains signature

Try uploading a simulated material pack (3 files, deliberately missing 1) and run the audit.

Challenge 4: Knowledge Base Q&A Assistant (30 min)

Challenge Goal

Create a Skill that connects to a knowledge base and answers employee questions.

Real-World Application

New employee asks "How to apply for VPN?" → Auto find answer from KB
Customer service asks "What does this error code mean?" → Auto query fault KB
Sales asks "What does this plan include?" → Auto query product KB

Two Ways to Configure Knowledge Base

Method A: Upload Documents (Simple)

Prepare Word/PDF/Markdown knowledge docs
Upload to Skill config
Skill auto-parses and builds index

Method B: Connect Existing KB (Advanced)

Configure KB MCP connector in TeleAgent
Specify KB ID or URL
Skill auto-queries and returns answers

Challenge 5: Permission Self-Check (30 min)

Challenge Goal

Learn to check Skill permission configurations, ensuring "least privilege principle" and avoiding security risks.

Why It Matters

A Skill with "delete all data" permission can cause severe damage if misused or attacked
Customer requirement: Skills must have minimized permissions, delete operations require double confirmation

Key Concept: Permission Audit

The Skill auto-checks another Skill's permission config and reports:

Risky permissions (e.g., unnecessary delete or broad data access)
Recommendations for permission tightening
Risk level rating (LOW/MEDIUM/HIGH)

Challenge 6: Message Linker (30 min)

Challenge Goal

Create a Skill that monitors enterprise WeChat group messages and auto-triggers other Skills based on keywords.

Real-World Application

Someone says "fault" in group → Auto invoke fault diagnosis Skill
Someone says "daily report" → Auto invoke report generator Skill
Someone says "audit" → Auto invoke material audit Skill

Keyword Rule Configuration

Keyword	Match Type	Invoke Skill	Reply Method
fault/outage/offline	Contains	KB Q&A	Reply in group
daily report/weekly/stats	Contains	Report Generator	Direct message
audit/check/materials	Contains	Material Audit	Direct message
help/how to	Contains	Usage Guide	Reply in group

Advanced Challenge: L2 Skills (90 min)

Challenge 7: Unified Entry Gateway (45 min)

Challenge Goal

Create a "smart front desk" Skill that understands user requests and auto-routes to the corresponding L1 Skill.

Scenario

User says: "Check yesterday's fault daily report for Nanchang, and also see if there are any complaints in Qingshanhu District"
→ Gateway understands: Two requests (daily report + complaints)
→ Auto invokes: Report Generator + KB Q&A
→ Merged response: One complete report

Key Concepts

Intent Recognition: Determine what the user wants (query data? generate report? audit materials?)
Routing Rules: Based on intent, dispatch to the corresponding Skill
Result Merging: Combine results from multiple Skills into one response

Challenge 8: Multi-Agent Complaint Handling Flow (45 min)

Challenge Goal

Create a workflow Skill that auto-handles customer complaints: Receive → Classify → Dispatch → Process → Respond.

Scenario

Customer complains: "My broadband hasn't been fixed for 3 days, customer service said someone was sent but no one came!"
→ Auto classify: Fault complaint + Service complaint
→ Auto dispatch: Technical dept + Customer service dept
→ Auto process: Check ticket status + Escalate
→ Auto respond: Inform processing progress and estimated resolution

Flow Nodes

[Receive Complaint] → [Sentiment Analysis] → [Classify] → [Dispatch] → [Process] → [Respond] → [Archive]
        ↓                    ↓                  ↓           ↓           ↓           ↓           ↓
   Extract info         Assess urgency     Tech/Service  Find owner  Check system  Generate reply  Write to table

Post-Course Skill Certification

Certification Standards

Level	Requirement	Certification Content
Bronze	Complete L1 x 6	Basic Skill config + simple modifications
Silver	Complete L1 x 6 + L2 x 1	Can configure routing rules
Gold	Complete all 8 challenges	Can independently design multi-Agent workflows

Certification Method

Submit your 8 Skill configuration files
Record a 3-minute demo video (any Skill running)
Complete the student feedback form

Appendix

A. Quick Reference Card

Import Skill: Skills → Import → Select folder → Confirm
Edit Skill:   Click Skill → Edit → Modify parameters → Save
Test Skill:   Enter parameters → Click Run → View output
Debugging:    Check logs → Check parameters → Check permissions

Parameter Types: Text/Number/Toggle/Select/File
Required Mark:   ✅ Must fill  |  ❌ Can leave blank
Default Value:   Auto-used when not specified

Common Errors:
• yaml indent error → Use 2 spaces, not Tab
• Permission denied → Contact admin to enable
• Data source unreachable → Check network/URL/permissions
• Wrong output format → Check output.format config

B. Troubleshooting Guide

Problem	Debug Steps	Contact
Skill import fails	1.Check skill.yaml format 2.Check file encoding (UTF-8) 3.Check indentation (2 spaces)	Instructor/TA
No output on run	1.Check if parameters are filled 2.Check network 3.View run logs	Instructor/TA
Wrong output format	1.Check output.format config 2.Check template syntax 3.Check data source format	TA
Permission denied	1.Check Skill permission config 2.Contact admin 3.Use least privilege	Admin
Feishu connection fails	1.Check Feishu URL 2.Check permissions 3.Check network	Feishu Admin

C. Post-Course Resources

Course Repository: https://github.com/yuzhaopeng-up/openclaw-workspace
TeleAgent Official: https://www.teleai.com.cn/product/super-agent
ClawHub Skill Marketplace: https://clawhub.ai
More Skills by Yu Zhaopeng: financial-ai-skills (104 financial AI skills), skill-framework (208-skill classification), soe-compliant-office (20 SOE-compliant Skills)

Version: v1.0 | Updated: 2026-06-20 | Course: AI Skill Practical Training