Forgedemy

Skill Marketplace for AI Agents

Visit Website
March 19, 2026 I built a marketplace where AI agents find and install skill packages mid-task

I had no clean way to move a skill between machines or hand one to someone. An agent skill is a folder: SKILL.md, scripts and reference files. The LLM agent reads the instructions and runs the scripts instead of starting from a blank prompt. Publishing a 3-file bundle to GitHub meant creating a repo and writing setup instructions. I have ADHD and I'd abandon the process every time.

So I built an API your agent calls during a task. It searches the catalogue, reads the skill files, and uses them for the current task. If the agent wants a local copy, it can install a live link that pulls updates, or download the full skill package. The website has a search, but your agent is supposed to do the browsing.

I built Forgedemy itself using Anthropic's frontend design skill and their PR review toolkit with six focused agents. Both are in the catalogue. But I wanted to port the toolkit to Codex, which is part of why I built this. I tried building custom commands and MCPs before skills, but writing a skill takes placing a markdown file in a special folder.

Business model: if you spent time polishing a skill until it works, you should be able to charge for it. You can price skills in Telegram Stars (sellers receive TON/USDT). Telegram handles payments, which was fast to ship but ties the flow to Telegram for now. No revenue yet — I'm focused on getting the catalogue and the agent flow right first.

Where it stands: 500+ entries synced from public repos. Curation is light. Works with Claude Code, Codex, OpenClaw, and the .agents standard. No paying customers yet. I use it daily for my own work and share with friends.

Still figuring out how trust should work when your agent picks skills on its own. Would love input on that, the business model, or anything else that's broken.

Comment

About

I kept building skills for my AI agents but had no clean way to share them. GitHub was too many steps. I wanted my agent to find and use skills mid-task without me doing the work.