active project
Agent Skill Marketplace
A production-minded SaaS platform for the full lifecycle of reusable agent skills. Users browse categorized skills with eval scores and run counts, test them in a live browser sandbox with permissioned execution and detailed trace events (latency, cost, blocked commands), build and publish new skills, compare versions with SKILL.md diffs, run evaluation suites with regression detection, and one-click export/install skills directly into Codex, Claude, Antigravity, OpenCode, Grok, and VS Code. Built as a Next.js App Router application with a Prisma-backed data model on Postgres.
Problem
What it solves.
AI agent capabilities are currently locked inside individual monolithic codebases, making them difficult to discover, trust, safely test, version, or share across teams. The marketplace turns skills into portable, evaluated, sandbox-testable SKILL.md packages with clear governance, execution traces, and easy installation into popular developer tools.
What I built
The specific parts I implemented.
What I owned
My responsibility surface.
Product direction & marketplace positioning
Full-stack implementation (Next.js App Router + Prisma)
Sandbox runtime semantics, trace model & permission system
Skill packaging standards & export/install flows
Developer experience & UI/UX
Architecture
How data moves through the system.
Next improvements
Known gaps & next moves.
- →Real isolated execution sandbox (beyond deterministic mock)
- →Auth, user accounts, publishing workflow & fork tracking
- →Public skill submission + community review process
- →Richer eval harness with LLM-as-judge + custom metrics
- →Analytics dashboard for skill usage & performance trends