How will competing Agent Skills be evaluated and ranked?

In the same essay, O'Reilly extends the protocol framing to Agent Skills, the packaged, reusable instructions Claude and similar systems use to perform a task. He describes a Skill as "a piece of structured knowledge: here's how to create a Word document; here's how to extract the text from a PDF; here's how to publish on the Hugging Face Hub," then argues Skills are themselves "a 'stratum of codified behavior,' and perhaps even a set of competing 'engineered arguments' for how to do work with AI." Because "there can be many Skills that attempt to codify the same knowledge, but some may be better than others," he poses the question directly: "As Skills multiply, how will we find the best ones?"

The mechanism is the same chain of definitions he builds earlier in the essay: a protocol is Ryan's "stratum of codified behavior" and Rao's "engineered argument," distinct from an API's "engineered agreement" enforced by one dominant actor. O'Reilly's move is to notice that a Skill fits that same definition — a codified way of doing something that competes with other codifications of the same task — without needing a formal standards body to sanction it. That places Skills in his de facto category: protocols, in his usage, that win adoption through market contention rather than through a designed specification process.

This cuts against the grain of the library's other protocol notes, which treat formal semantics as the fix for agentic AI's problems: Do LLM agents misinterpret delegated requests more than formal protocols? argues LLM agents guess meaning statistically where classical agent-communication languages fixed it by specification, and Can classical agent architectures fix agentic AI accountability? calls for reintroducing that kind of formal commitment structure. O'Reilly's Skills argument runs the other way: he treats natural-language, informally codified practice as itself protocol-like and market-adjudicated, without claiming it needs formal semantics to count as a protocol — codification by example and adoption, not by specification.

The essay does not propose or test a mechanism for comparing competing Skills; it ends on the open question rather than an answer, and offers no criteria (correctness, efficiency, safety) by which "better" Skills would be identified. It also does not establish that Skills will in fact proliferate and compete the way MCP and other communication protocols did — that is an analogy carried over from the preceding argument, not a separately demonstrated case. The implication, left unstated but following from his own disclosures framework, is that if Skills do function as de facto protocols, the same observability concern he raises for AI deployment generally — that governance requires visibility into what's actually adopted and why — would apply to Skill marketplaces too.