Four pillars · not one opaque grade

Skill Trust Scorecard

Every listing gets independent signals: functional verification (tiered honesty for sealed packs), permissions / least agency, freshness, and reputation from confirmed unlocks (ERC-8004 additive when configured). Zero sales uses a neutral prior — new is not a red fail.

Skill id

API: /api/trust?skill=… · Index: /trust/index.json · Field on each catalog.json skill: trust

Rollout (honest)

v1 auto (now): integrity hash + inferred scope + freshness + ledger counts for all listings.
Operator verified: opt-in execution_verified for featured skills only.
Ranking: use ranking_boost / freshness among equal peers — don’t bury unscored pillars as fails.
ERC-8004: merge when identity/reputation registries are live; degrade to ledger-only until then.