Independent product / 2025-2026
AI SkillsBank.
A visual home for the instructions, tools, knowledge, and guardrails that make AI agents useful.
I built it to solve a problem I was already experiencing: my agents were getting more capable, but their reusable intelligence was scattered across terminals, harnesses, and file formats.
The premise
Agent skills should be portable assets, not knowledge trapped inside one chat or one tool.
My role
Product strategy, experience design, agent-directed engineering, testing, deployment, and operation.
The result
A working SaaS product, a much deeper understanding of agent architecture, and a useful lesson in timing and product-market fit.
01 / The problem
I needed a better view than the terminal.
While experimenting with persistent agents in OpenClaw and later Hermes Agent, I asked them to create reusable skills for recurring work. The skills were useful, but inspecting, comparing, and moving them between tools remained awkward.
Git could version the files, but it did not give me the visual, approachable experience I wanted for understanding what an agent could do. I also expected more people to move between agent harnesses and need a neutral place for the capabilities they had developed.
02 / What I built
From prompt vault to portable agent system.
The product began as a prompt version-control experiment. As I explored the market and my own usage, it evolved into a system for creating, organizing, testing, and exporting agent skills.
It grew to include guided skill creation, variables, evaluations, revision tools, public portfolios, and agents composed from skills, knowledge, memory, tools, and guardrails. The goal was not another chat interface. It was an inspectable layer of reusable intelligence that could travel between AI tools.
03 / What I learned
The product solved my problem. That was not enough.
I used AI SkillsBank heavily to move work between agent tools. Visitors from builder communities explored it, but they did not adopt the workflow with the same intensity. Technical users already had Git, terminals, and increasingly capable agents that could translate files between formats for them.
The audience that might benefit most from a visual alternative was not yet operating personal agents at meaningful scale. I had built for a behavior I believed would become common before it actually had.
Building for yourself is a good starting point. It is not proof that enough other people experience the problem in the same way.
04 / Why it matters
A product can be valuable even when the market says “not yet.”
AI SkillsBank changed how I understand agent systems, portability, evaluation, and the gap between technical possibility and adopted behavior. It also demonstrated what a product-minded builder can now make with AI-assisted engineering.
I am still convinced that people will need understandable ways to manage what their agents know and can do. I am less convinced that a standalone marketplace is the form that need will take.