AI Catalog Permissions, Backend Enforcement, Conditional Rules (issue 3 of 29)

[gabemontero]

Labels: ready-to-code
Dependencies: None
RHIDP Stories: RHIDP-15271, RHIDP-15272, RHIDP-15306 (permission definitions), RHIDP-15312
Feature: RHDHPLAN-1508 — Epic RHIDP-15270

Define AI Catalog permission constants (ai-catalog.asset.read, ai-catalog.asset.read.usage-docs, ai-catalog.admin), implement graduated visibility backend enforcement (Tier 1 entity-level + Tier 2 field-level filtering), and implement conditional permission rules (isAiAssetCategory, isFromConnector, isInTenant) with toQuery() support. The default-deny config implementation is in Issue 20.

Tasks

From openspec/changes/ai-catalog-asset-governance/tasks.md group 1 (RHIDP-15271, RHIDP-15306):

• 1.1 Define AI_CATALOG_ASSET_RESOURCE_TYPE constant in boost-common/src/permissions.ts
• 1.2 Define ai-catalog.asset.read resource permission
• 1.3 Define ai-catalog.asset.read.usage-docs resource permission
• 1.4 Define ai-catalog.admin basic permission
• 1.5 Export all permission constants and resource type
• 1.6 Register all 3 permissions via permissionsRegistry.addPermissions()

From openspec/changes/ai-catalog-asset-governance/tasks.md group 2 (RHIDP-15271, RHIDP-15272):

• 2.1 Implement Tier 2 field-level filtering in AI asset detail endpoint
• 2.2 Implement entity-level filtering on list endpoint using authorizeConditional()
• 2.3 Implement batch authorizeConditional() for Tier 2
• 2.4 Add unit tests for field-level filtering

From openspec/changes/ai-catalog-asset-governance/tasks.md group 4 (RHIDP-15312):

• 4.1 Implement isAiAssetCategory rule with apply() and toQuery()
• 4.2 Implement isAiAssetCategory.toQuery() generating catalog query predicate
• 4.3 Implement isFromConnector rule with apply() and toQuery()
• 4.4 Implement isFromConnector.toQuery() generating catalog query predicate
• 4.5 Implement isInTenant rule with apply() and toQuery()
• 4.6 Implement isInTenant.toQuery() generating catalog query predicate
• 4.7 Register all 3 rules via createPermissionIntegrationRouter with resourceType: 'ai-catalog-asset'
• 4.8 Add unit tests for each rule's apply() and toQuery() methods

Specifications

• openspec/changes/ai-catalog-asset-governance/specs/graduated-visibility/spec.md
• openspec/changes/ai-catalog-asset-governance/specs/conditional-policies/spec.md

[fullsend-ai-triage]

🤖 Finished Triage · ✅ Success · Started 7:59 PM UTC · Completed 8:04 PM UTC
Commit: e03b281 · View workflow run →

[fullsend-ai-triage]

Triage Summary

Severity: High | Category: Feature

This issue defines the foundational AI Catalog permission framework for the boost workspace, covering three task groups:

Scope

• Group 1 (Permission Definitions): Define AI_CATALOG_ASSET_RESOURCE_TYPE, three permission constants (ai-catalog.asset.read, ai-catalog.asset.read.usage-docs, ai-catalog.admin), and register them via permissionsRegistry.addPermissions()
• Group 2 (Graduated Visibility Backend): Implement Tier 2 field-level filtering on detail endpoints and entity-level filtering on list endpoints using authorizeConditional()
• Group 4 (Conditional Rules): Implement isAiAssetCategory, isFromConnector, and isInTenant rules with apply() and toQuery() methods

Implementation Context

• Existing pattern in boost-common/src/permissions.ts defines boost-agent and boost-tool permissions — follow the same conventions
• Detailed behavioral specs exist at openspec/changes/ai-catalog-asset-governance/specs/graduated-visibility/spec.md and conditional-policies/spec.md
• The AGENTS.md establishes that ai-catalog.* is the correct permission namespace for catalog-layer RBAC

Related Work

• Issue boost: RBAC-gate the Usage tab #3734 (RBAC-gate the Usage tab) depends on the ai-catalog.asset.read.usage-docs permission defined here
• Default-deny configuration is scoped to a separate issue (group 6 in tasks.md)

Proposed Test Cases

1. Permission definitions:
   - Verify AI_CATALOG_ASSET_RESOURCE_TYPE === 'ai-catalog-asset'
   - Verify each permission has correct action and resourceType

2. Field-level filtering:
   - Tier 1 only: response includes name/description/category, omits usage-docs/config
   - Tier 2 included: response includes all fields
   - No access: returns 403

3. Conditional rules (for each: isAiAssetCategory, isFromConnector, isInTenant):
   - apply() returns true when annotation matches params
   - apply() returns false when annotation does not match
   - apply() returns false when annotation is missing
   - toQuery() generates correct catalog query predicate

4. List endpoint integration:
   - ALLOW result: returns all assets unfiltered
   - DENY result: returns 403
   - CONDITIONAL result: returns only assets matching rule criteria