Skip to content
Open
Show file tree
Hide file tree
Changes from 2 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
104 changes: 104 additions & 0 deletions adr/0021-separate-artifact-identity-from-registry-coordinates.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,104 @@
# ADR-0021: Separate Artifact Identity from Registry Coordinates

**Status:** Proposed

**Date:** 2026-08-26

**Related:** [Issue #102](https://github.com/Agent-Card/ai-catalog/issues/102)

**Supersedes if accepted:** The federated AIR requirement in
[ADR-0015](0015-agent-identifier-naming.md)

## Context

Existing registries commonly identify artifacts with native coordinates,
such as a registry namespace and artifact name. When projecting those
records as AI Catalog entries, a registry may have no publisher-assigned
`urn:air` identifier.

Using the registry's domain in `urn:air` is easy, but the current format
defines that domain as the artifact publisher. Requiring the publisher's
domain is portable, but prevents automatic projection when the publisher
has not supplied or authorized an identifier.

The existing model already separates most roles: `publisher` identifies
the artifact publisher, `host` identifies the catalog operator, `url`
locates the artifact, and `extensions` can preserve registry-specific
coordinates.

## Decision

`entry.identifier` identifies the artifact represented by the entry and
remains stable across versions. A publisher-authorized `urn:air`
identifier is also portable across catalog locations. Other
identifier schemes remain valid but have no cross-catalog portability
guarantee. Consumers that do not recognize an identifier scheme treat
the value as opaque. Identifier syntax alone does not verify publisher
identity or establish trust.

The base format does not require a particular identifier scheme. A
globally unique absolute URI is recommended for open or federated use.
An artifact publisher that wants an identifier to be preserved when the
artifact appears in other catalogs should use the AI Catalog-specific
`urn:air` format in a namespace it controls.

A publisher-authorized `urn:air` identifier is one assigned by the
artifact publisher or its authorized delegate, with the artifact
publisher's domain in the `{publisher}` segment.

A registry uses a stateful preserve-or-mint policy:

1. If the registry previously published an identifier for the artifact,
reuse it, even if another identifier becomes available later.
2. Otherwise, if the source entry contains a publisher-authorized
`urn:air` identifier, preserve it exactly.
3. Otherwise, the registry may preserve or replace a non-`urn:air` source
identifier. Non-`urn:air` identifiers have no guaranteed portability
across catalogs.
4. When assigning a new identifier, a registry that becomes the artifact
publisher should use `urn:air` with its own domain in the `{publisher}`
segment. A registry acting as an authorized delegate should use
`urn:air` with the delegating publisher's domain in that segment. An
independent registry must use a non-`urn:air` identifier under its own
control.

Merely hosting or aggregating an entry does not make a registry the
artifact publisher.

Changing a previously published primary identifier requires an explicit
migration mechanism, which this decision does not define.

The registry assigning an identifier, the catalog `host`, and the
artifact `publisher` are independent roles. A registry-issued identifier
does not imply that the registry published the artifact.

A registry may retain replaced source identifiers or registry-native
coordinates. When retained, they should be stored in a namespaced entry
extension. They do not affect catalog uniqueness or establish publisher
identity, trust, or equivalence with another identifier.

## Consequences

- Publisher-authorized `urn:air` identifiers are preserved across
registries and mirrors.
- Non-`urn:air` identifiers remain valid, but registries may replace them
when they are unsuitable for the destination catalog.
- Legacy records can be projected without publisher enrollment by using
a registry-issued identifier.
- Two registries may assign different identifiers to the same artifact
when no publisher-authorized `urn:air` identity is available. The model
does not claim equivalence it cannot establish.
- No new core field is added; native coordinates use `extensions`.
- Generic aliases and primary-identifier migration remain future work.

## Alternatives Considered

Always using a registry-domain `urn:air` was rejected because it makes the
registry appear to be the artifact publisher under the current format.

Requiring a publisher-domain `urn:air` was rejected as a universal rule
because it makes automatic projection depend on publisher enrollment.

Adding `aliases`, `nativeIdentifier`, or `identifiers[]` was deferred
because registry coordinates do not necessarily assert logical identity
equivalence, and `extensions` is sufficient for the immediate use case.
78 changes: 72 additions & 6 deletions specification/ai-catalog.md
Original file line number Diff line number Diff line change
Expand Up @@ -203,9 +203,9 @@ A Catalog Entry object describes a single AI artifact in the catalog.
It MUST contain the following members:

`identifier`
: A string uniquely identifying this artifact. This field is an open text format (e.g., any valid URI or URN is accepted). However, to ensure interoperability, identity uniqueness, and discoverability, the standard `urn:air` naming structure is **HIGHLY RECOMMENDED** and **MUST** be used for open or federated systems.
: A string uniquely identifying this artifact. This field is an open text format (e.g., any valid URI or URN is accepted). Consumers that do not recognize an identifier scheme MUST treat the value as opaque. Identifier syntax alone does not verify publisher identity or establish trust. For open or federated systems, a globally unique absolute URI is RECOMMENDED. An artifact publisher that wants an entry identifier to be preserved when the artifact appears in other catalogs SHOULD use the AI Catalog-specific `urn:air` naming structure with its own domain in the `{publisher}` segment. A catalog incorporating an entry for the first time with a publisher-authorized `urn:air` identifier MUST preserve that identifier exactly.

**Standard Naming Format:**
**AI Catalog Publisher Naming Format:**
Comment thread
jonathanhefner marked this conversation as resolved.
Outdated
`urn:air:{publisher}:{namespace}:{name}`

- `{publisher}`: The domain name of the organization publishing the artifact (e.g., `example.com`).
Expand All @@ -219,7 +219,7 @@ It MUST contain the following members:

For closed or local systems where a different identifier format is used, client implementations are responsible for parsing and processing the custom format as appropriate.

See [Multi-Version Entries](#multi-version-entries) for uniqueness rules when multiple versions are present.
See [Multi-Version Entries](#multi-version-entries) for uniqueness rules when multiple versions are present, and [Registry Projection](#registry-projection) for identifiers assigned while projecting an existing registry.

`type`
: A string containing the identifier that specifies the type of the
Expand Down Expand Up @@ -433,9 +433,9 @@ single artifact — similar to a package registry.

When `version` is present, the combination of `identifier` and `version`
MUST be unique within the catalog. When `version` is absent, `identifier`
alone MUST be unique. The `identifier` SHOULD be stable across versions
and catalog locations so that the same logical artifact can be
recognized wherever it appears.
alone MUST be unique. The `identifier` SHOULD be stable across versions.
Only publisher-authorized `urn:air` identifiers receive a cross-catalog
preservation requirement, as defined in [Registry Projection](#registry-projection).

Clients that need only the latest version SHOULD sort entries
sharing the same `identifier` by `version` (when parseable as a semantic
Expand Down Expand Up @@ -469,6 +469,72 @@ For example, a catalog listing two versions of the same agent:
Both entries share the same `identifier` but have distinct `version`
values, so the combination is unique.

## Registry Projection
Comment thread
jonathanhefner marked this conversation as resolved.
Outdated

A publisher-authorized `urn:air` identifier is one assigned by the
artifact publisher or its authorized delegate, with the artifact
publisher's domain in the `{publisher}` segment.

A registry creating an entry from an existing record or catalog entry

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

existing record

Does this mean a record in a non-ai-catalog registry/market place?

MUST select its primary identifier by applying these rules in order:

1. If the registry previously published an identifier for the artifact,
it MUST reuse that identifier, even if another identifier becomes
available later.
Comment thread
jonathanhefner marked this conversation as resolved.
Outdated
2. Otherwise, if the source entry contains a publisher-authorized
`urn:air` identifier, it MUST preserve the identifier exactly.
3. Otherwise, the registry MAY preserve a non-`urn:air` source identifier
or replace it. Non-`urn:air` identifiers have no guaranteed portability
across catalogs.
4. When assigning a new identifier, a registry that becomes the artifact
publisher SHOULD use `urn:air` with its own domain in the `{publisher}`
segment. A registry acting as an authorized delegate SHOULD use
`urn:air` with the delegating publisher's domain in that segment. An
independent registry MUST use a non-`urn:air` identifier under its own
control.
Comment thread
jonathanhefner marked this conversation as resolved.
Outdated

Merely hosting or aggregating an entry does not make a registry the
artifact publisher.
A registry MUST NOT infer publisher authorization from the artifact's URL
or an unsigned `publisher` field.

Adopting a different primary identifier after publication requires an
explicit migration mechanism, which this specification does not define.

A registry-assigned identifier identifies the artifact, not a particular
version or registry record. It MUST remain stable across
versions, registry-coordinate changes, and retrieval-URL changes, and
MUST NOT be reassigned to another artifact. The registry assigning the
identifier, the catalog `host`, and the artifact `publisher` are
independent roles.

A registry MAY retain replaced source identifiers or registry-native
coordinates. If retained, they SHOULD be stored in a namespaced entry
extension. They do not participate in catalog uniqueness or establish
publisher identity, trust, or equivalence with another identifier.

```json
{
"identifier": "https://registry.example/ids/artifacts/7bf4a8c2",
"type": "application/a2a-agent-card+json",
"url": "https://registry.example/apis/registry/v3/groups/payments/artifacts/fraud-agent",
"publisher": {
"identifier": "did:web:acme.example",
"displayName": "Acme"
},
"extensions": {
"com.example.registry.coordinates": {
"registryUri": "https://registry.example",
"sourceIdentifier": "foo",
"namespace": "payments",
"name": "fraud-agent"
}
}
}
```

The extension key is illustrative.

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As we're aware that AI catalog is designed for federation, do you think it makes sense to formalise a native/official extension for federation source / entry origin?

This could then be used to build chain of origin if someone wants to find the normative source of a entry, e.g. resolve the entry origin then check that entries origin until you find an entry with no origin.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I pushed a commit to address the other comments, but I think this comment should be addressed in a separate PR if we want to pursue it.


## Publisher Object

The Publisher object identifies the entity responsible for an artifact.
Expand Down
Loading