Li Yin

First author of attention-is-all-you-have and the designer of @skills, founder and CEO of SylphAI. Known in the Python LLM ecosystem before this as the creator of AdalFlow, the auto-optimizing LLM-application library — which puts the same author on both sides of a line this spoke tracks: fitting the prompt to a metric (prompt-optimization, dspy), and now spending as little of the prompt as possible.

What the paper shows of the position: the argument is a delivery argument rather than a format one, and its force comes from having crawled the corpus first and designed second — 56,804 skills, 54 project directories across 75 agents, 21 providers publishing 958 first-party skills. The design that follows is subtractive on purpose ("the most useful thing to say about this protocol is how little of it is new"), and it leans on borrowed conventions — the @ gesture, .gitignore semantics, Go-style vendoring, git for transport — rather than on new machinery.

The claim most likely to be tested first is not about skills at all: that AGENTS.md won on delivery simplicity rather than on design quality, so a successor has to be at least as easy to use, not merely better.

Connections

• founder of sylphai
• author of attention-is-all-you-have, designer of atskills-protocol

Related

agent-skills · agentskills-spec · prompt-optimization · dspy