Agent capability matching

[jedisct1]

Hi,

And thanks for starting this simple but useful proposal.

Skills can be used by a wide range of agents, with different capabilities. But with the current proposal, skills can't declare the required capabilities (shell access, network, filesystem, etc.)

As a result, agents might fetch skills they can't execute, or can't execute correctly.

It may be worth adding a way to filter skills by what the agent supports.

For example by adding a capability field:

{
  "name": "data-pipeline",
  "capabilities": ["shell", "filesystem", "network"]
}

[douglasborthwick-crypto]

Adding to the capability discussion: beyond runtime capabilities like shell/filesystem/network, there's a category of verification capabilities that skills might need from agents (or vice versa).

For example, a commerce skill that applies token-gated discounts needs the agent to be able to verify on-chain state. We provide this through InsumerAPI (ECDSA-signed attestation across 31 EVM chains), available as an MCP server.

A skill declaring a dependency like "capabilities": ["on_chain_verification"] would let agents know they need access to a blockchain attestation tool before fetching that skill. This might generalize to a broader verification capability category (identity verification, on-chain attestation, credential checking, etc.) that becomes important as agentic commerce grows.