Descriptions:
Delivered as a recorded keynote for the AI Engineer World’s Fair, this talk by the creator of the popular Matt PCO skills repo introduces what he calls a “missing manual” for writing great agent skills — a structured checklist designed to help developers and organizations escape what he terms “skill hell”: the state of having access to many skills without being able to evaluate or compose them effectively.
The checklist covers four areas: trigger design (the difference between user-invoked and model-invoked skills, and why it matters for agent behavior), internal structure (how to compose steps and reference material cleanly within a skill file), steering (how to guide agent decision-making through the skill), and minimization (the discipline of making skill files as small as possible to reduce tokens, maintenance burden, and ambiguity). Throughout, the speaker uses concrete comparisons between his own skills and the competing Superpowers skill set to illustrate the practical consequences of different design choices.
Two detailed worked examples anchor the abstract principles: a PRD-writing skill that includes a human-in-the-loop checkpoint before test seam confirmation, and a domain-modeling skill with branching behavior that justifies splitting reference material into separate files rather than embedding everything in the main skill.md. The speaker also introduces a new meta-skill in his repository — “writing great skills” — that encodes the entire checklist so developers can use an agent to improve their existing skills or author new ones from scratch.
📺 Source: AI Engineer · Published June 29, 2026
🏷️ Format: Deep Dive
