TLDR: Record your workflow once, get a reusable SKILL.md file that works with Claude Code, gstack, and other AI coding agents. Free tier available at skillforge.expert.
The Problem
You've spent hours teaching your AI agent how to do something. It finally works. Then you start a new chat, and... it forgot everything.
Sound familiar?
AI coding agents like Claude Code, Codex, and OpenClaw are powerful, but they don't retain skills across sessions. Every time you need them to do something complex (deploy to Vercel, run your test suite, interact with an API), you're either:
- Re-explaining the entire workflow
- Copy-pasting from your notes
- Hoping it remembers from context
There's a better way.
The Solution: SKILL.md
The SKILL.md standard (published by Anthropic in Dec 2025) lets you package workflows as reusable skills that AI agents can load on-demand.
Think of it like this:
-
Without skills: "Claude, please deploy this to Vercel. First run
npm run build, thenvercel --prod, and if it asks about the project, say yes..." - With skills: "Claude, use the deploy-vercel skill"
The agent reads the skill file, understands the context, and executes the workflow correctly.
The Challenge: Writing SKILL.md Files
Here's the catch: writing good SKILL.md files is hard.
You need to:
- Document every step clearly
- Handle edge cases
- Format it for AI consumption (not human reading)
- Test it across different scenarios
Most developers spend 30-60 minutes per skill. And if your workflow changes? Rewrite time.
Introducing SkillForge
SkillForge turns screen recordings into AI agent skills.
Here's how it works:
- Record your workflow (using any screen recorder)
- Upload to SkillForge
- Get a SKILL.md file that works with Claude Code, gstack, OpenClaw, and any other agent that supports the standard
What SkillForge Extracts
- Commands executed (terminal, browser, IDE)
- UI interactions (clicks, form fills, navigation)
- Conditional logic (if you do X, then Y)
- Error handling (what you did when things went wrong)
- Context (what the workflow is for, when to use it)
Example: Deploying to Vercel
You record:
- Opening terminal
- Running
npm run build - Running
vercel --prod - Answering prompts
- Checking deployment URL
SkillForge generates:
# deploy-vercel
## Description
Deploy Next.js app to Vercel production
## When to Use
User says "deploy to vercel", "push to production", "go live"
## Steps
1. Run `npm run build` (wait for completion)
2. Run `vercel --prod`
3. If prompted "Link to existing project?": answer "yes"
4. Wait for deployment URL
5. Confirm deployment at returned URL
## Common Issues
- If build fails: check for TypeScript errors first
- If "no vercel.json": user needs to run `vercel init` first
## Output
Return deployment URL to user
Now any AI agent can use this skill. No more re-explaining.
Real-World Use Cases
People are using SkillForge for:
- Deployment workflows (Vercel, Railway, Fly.io)
- Testing routines (run tests, check coverage, lint)
- API integrations (authenticate, make requests, handle responses)
- Database migrations (backup, migrate, seed)
- Content publishing (build, optimize images, deploy)
- Customer support (check logs, query DB, send updates)
Basically: anything you do repeatedly that involves multiple steps.
Getting Started
Free tier: 3 recordings/month, full SKILL.md output
Paid tier: Unlimited recordings, team sharing, advanced features
👉 Start recording at skillforge.expert
Why This Matters in 2026
The SKILL.md standard is becoming ubiquitous:
- Claude Code: native support
- OpenAI Codex: native support
- OpenClaw: native support
- Other frameworks: rapidly adopting
Over 2,600+ skills are already published (and doubling every quarter).
If you're building with AI agents, you need a skill library. SkillForge is the fastest way to build one.
The Bigger Picture
We're moving from "prompting AI" to "teaching AI workflows."
Skills are the unit of reusable agent knowledge. The better your skill library, the more productive your agents become.
SkillForge automates the teaching part. You focus on doing the work (which you were going to do anyway). The AI learns by watching.
Try it: skillforge.expert
Questions? Drop them in the comments. I'm actively building this and love feedback.
Top comments (0)