What an agent-skill installer has to get right
Published
Nine months of maintaining Ai-Agent-Skills, reduced to the decisions that mattered: where files go, what a copy owes its upstream, and why a manifest beats a folder.
Writing · Updated
Writing
Long-form pieces by me, Abdul Moiz Shahzad, about the systems I build and the ones underneath them. They come out of the work, when there is something worth showing: what the durable systems around a model have in common, the markdown layer that runs the agents, the tools I maintain for people who run them, and a map of one AI request from the power plant to the screen.
Pieces
Agent skills and MCP servers: what each one is for
Published
One gives an agent a new capability; the other tells it how to use the capabilities it has. Confusing the two produces tools nobody calls and instructions nobody can execute.
The Map of AI
Published ; revised
Most maps begin with the model. This one follows a common cloud AI request from the grid to the application on screen.
Agent engineering in 2026: the harness is the product
Published ; revised
Across the strongest 2026 operator reports and official docs, the same artifacts keep appearing: progress files, decision logs, checkpointers, approval gates, and runtimes that know when to stop.
The markdown is the message
Published ; revised
I spend more time on markdown files than on functions. The code is generated; the markdown is engineered. Here is what that layer is made of.