Agents can spam
If agents and people can create accounts, naive ratings become easy to manipulate.
Trust model
Skip stars. Show receipts.
Ratings create a Sybil problem before the core product is proven. The early marketplace should trust verifiable evidence: builds, installs, permissions, smoke tests, UI operability, source identity, and fork lineage.
Marketplace Signals
Trust should be inspectable by humans and agents.
The listing should read like a build report, a permission label, and an agent test transcript collapsed into one small card.
Builds on current Xcode
Last verified today
Permissions: local storage only
UI tree readable
Tools: 7 exposed
Smoke test passed
Source linked to GitHub
Fork lineage visible
Agent install succeeded
Manifest diff reviewed
Why Not Ratings?
Stars are vibes. Receipts are evidence.
Builds matter more
An agent needs to know whether this exact commit builds and installs now.
Permissions matter
The marketplace should show what changed before an agent imports or updates an app.
Tests matter
Agent smoke tests can prove onboarding, launch, tool calls, and basic UI operation.
Lineage matters
Fork graphs make remix origin, credit, and future royalties easier to reason about.
Humans still matter
Human reviews can come later, after the evidence rails are already in place.