Problem
ai-catalog defines a publishing format with identity verification (trustManifest), source lineage (provenance[]), legal terms (termsOfServiceUrl, privacyPolicyUrl), and structural federation via nested catalogs (Design Goal #5: Scalable Federation). The recent proposals for dependencies (#63), deployment metadata (#64), and lifecycle (#65) fill additional gaps for single-registry use.
But structural federation (catalog A links to catalog B) has no governance layer. When catalogs get consumed by other registries — federated, mirrored, or aggregated across org boundaries — three governance gaps remain:
- No residency semantics.
provenance[].relation tracks build lineage ("publishedFrom", "derivedFrom") but has no vocabulary for registry-level residency: is this entry owned here, federated from elsewhere, or an unverified mirror?
- No bilateral consent.
trustManifest.trustSchema defines a trust framework but doesn't model bilateral consent between registries — publisher A approving registry B to federate their entries, with scoping and revocation.
- No structured execution terms.
termsOfServiceUrl and privacyPolicyUrl point to legal documents but provide no machine-readable policy for runtime enforcement — quota limits, data residency requirements, SLA targets.
These gaps are acute in B2B scenarios — e.g., a SaaS provider's agents federated into a customer's enterprise registry. Neither MCP nor A2A address cross-registry governance at the protocol level.
Suggested Direction
Building on existing Trust Manifest primitives:
Extend Provenance Link vocabulary for federation residency:
- New
relation values: "federatedFrom", "mirroredFrom"
- New optional field:
residency enum (owned / federated / mirrored)
- New optional field:
lastSynced (datetime — when this copy was last reconciled with source)
mutualTrust — bilateral consent state machine (distinct from trustSchema):
trustSchema = what framework governs an artifact. mutualTrust = who has agreed to federate it.
- State machine:
unverified → one_way → mutual → revoked
- Negotiation via
.well-known endpoints (DNS) or DID Document service endpoints (decentralized)
executionContract — structured complement to existing termsOfServiceUrl:
- Machine-readable runtime policy: quota limits, data handling rules, SLA targets
termsOfServiceUrl remains the human-readable legal doc; executionContract is the machine-parseable counterpart
Interoperability Impact
Related Work
Next Steps
Happy to draft spec text + examples in a follow-up PR once directional alignment is confirmed.
Problem
ai-catalog defines a publishing format with identity verification (
trustManifest), source lineage (provenance[]), legal terms (termsOfServiceUrl,privacyPolicyUrl), and structural federation via nested catalogs (Design Goal #5: Scalable Federation). The recent proposals for dependencies (#63), deployment metadata (#64), and lifecycle (#65) fill additional gaps for single-registry use.But structural federation (catalog A links to catalog B) has no governance layer. When catalogs get consumed by other registries — federated, mirrored, or aggregated across org boundaries — three governance gaps remain:
provenance[].relationtracks build lineage ("publishedFrom", "derivedFrom") but has no vocabulary for registry-level residency: is this entry owned here, federated from elsewhere, or an unverified mirror?trustManifest.trustSchemadefines a trust framework but doesn't model bilateral consent between registries — publisher A approving registry B to federate their entries, with scoping and revocation.termsOfServiceUrlandprivacyPolicyUrlpoint to legal documents but provide no machine-readable policy for runtime enforcement — quota limits, data residency requirements, SLA targets.These gaps are acute in B2B scenarios — e.g., a SaaS provider's agents federated into a customer's enterprise registry. Neither MCP nor A2A address cross-registry governance at the protocol level.
Suggested Direction
Building on existing Trust Manifest primitives:
Extend Provenance Link vocabulary for federation residency:
relationvalues:"federatedFrom","mirroredFrom"residencyenum (owned/federated/mirrored)lastSynced(datetime — when this copy was last reconciled with source)mutualTrust— bilateral consent state machine (distinct fromtrustSchema):trustSchema= what framework governs an artifact.mutualTrust= who has agreed to federate it.unverified→one_way→mutual→revoked.well-knownendpoints (DNS) or DID Document service endpoints (decentralized)executionContract— structured complement to existingtermsOfServiceUrl:termsOfServiceUrlremains the human-readable legal doc;executionContractis the machine-parseable counterpartInteroperability Impact
termsOfServiceUrlis complemented, not superseded; Trust Schema is untouched.Related Work
accessModel). Distinct: Spec Proposal: Discovery-Time Access & Monetization Signaling #83 signals pricing tier for selection;executionContractcovers operational terms post-selection (quota, SLA, data residency).Next Steps
Happy to draft spec text + examples in a follow-up PR once directional alignment is confirmed.