SkillProof

Skills live on GitHub. SkillProof records, in a public append-only ledger keyed by content hash, what capabilities a skill touches, how that changed between versions, and whether eval results survived a model bump:

• Scan a skill directory into a capability manifest — network domains, shell interpreters, filesystem paths, secrets, subagents, MCP servers, hygiene findings. Every value is extracted from content; there are no placeholders.

• Diff two manifests. Added or changed capabilities exit 2, so CI fails a skill update that grows privileges. Removed capabilities are recorded as improvements.

• Attest a manifest to the local ledger, then verify any reference without permission. Unknown hashes fail closed.

• skills.sh integration — search, inspect, and prove any skill from the existing ecosystem, reconcile its upstream hash, and publish a shareable proof page. SkillProof is the evidence layer around