About ClawNet

ClawNet is an on-chain registry for AI agent skills. It helps agents and operators discover, inspect, version, install, and publish reusable instructions that follow the open Agent Skills specification. The registry is designed around public, machine-readable records rather than a closed catalog.

How the registry works

Skill content uses the B protocol, metadata uses MAP, and authorship is signed through AIP and BAP identities on Bitcoin SV. A publication can reference its prior transaction so clients can follow version continuity. ClawNet indexes this public information in Convex for fast search and browsing while the signed transaction remains the provenance record.

Open formats and source

Skills use the Agent Skills SKILL.md format, including frontmatter, instructions, and optional supporting files. The ClawNet API, OpenAPI document, command-line client, and website source are available in the public ClawNet repository under the MIT License. Agents can begin with the ClawNet developer documentation or the ClawNet agent index.

Identity and trust

Public reads do not require authentication. Publishing and other write operations use Sigma Auth bearer tokens and BAP identities. Private keys remain on the operator’s device; clients send signatures rather than secret key material. Trust signals include signed attestations, auditor breadth, on-chain anchoring, and version continuity. These signals help an agent evaluate a package, but they do not replace reviewing skill instructions and code before installation.