Creating AI Agent Skills Without Writing Code
What if you could teach an AI agent to perform complex web tasks in less time than it takes to drink a coffee? No scripting. No debugging brittle selectors. Just record your screen and let AI do the rest.
That's the promise of demonstration-based skill creation.
The Traditional Pain
If you've ever built browser automation, you know the drill:
- Inspect the DOM to find element selectors
- Write code that depends on those selectors
- Test it thoroughly
- Watch it break when the website updates
- Repeat
A simple "book a meeting" workflow can take hours to script and days to debug across different sites.
The New Way: Record Once, Reuse Forever
What if you could just... show the agent what to do?
Here's the process that changes everything:
Step 1: Record (2 minutes)
Fire up your screen recorder and perform the task normally. Navigate to the website, click buttons, fill forms—just do what you'd normally do.
Step 2: AI Extraction (30 seconds)
Modern AI analyzes the recording to identify:
- Your goal — What you're trying to accomplish
- The workflow — Step-by-step actions
- UI context — How to identify elements by appearance, not fragile selectors
- Decision points — Where choices need to be made
- Error handling — Recovery strategies
Step 3: Get Your SKILL.md (instant)
The output is a structured, human-readable skill file:
# Book a Demo
## Goal
Schedule a product demo through the website
## Workflow
1. Navigate to /book-demo
2. Locate the calendar widget
3. Select an available time slot
4. Fill contact information
5. Submit the booking
## Context
- Look for calendar with time slots
- Form should have name, email, company fields
- Success = confirmation message or email
## Error Handling
- If no slots available → try next day
- If validation fails → check required fields
Step 4: Deploy (any agent, any framework)
Your SKILL.md works with LangChain, AutoGen, CrewAI, or any compatible agent framework.
Why This Matters
Speed: Create skills in minutes, not hours
Resilience: Skills survive UI updates because they describe intent, not implementation
Accessibility: Domain experts can create automation without coding
Portability: One skill file works across multiple frameworks
Real-World Example
I needed to automate expense report submission. The traditional way would require:
- Analyzing the expense portal's DOM
- Writing Selenium/Playwright scripts
- Maintaining those scripts when the UI updates
With demonstration-based creation:
- Recorded myself submitting one expense report (3 minutes)
- Got a structured SKILL.md file
- Deployed to my agent
- Done
Total time: 5 minutes. Maintenance: virtually zero.
Try It Yourself
Want to create your own agent skills?
🚀 SkillForge — Record your screen, get a SKILL.md file
🔥 Support our Product Hunt launch
What tasks would you automate if you could create skills in 5 minutes?
Top comments (0)