feat: add central capability metadata schema definition

adm/gdl/dev/schemas/capability-meta.schema.yaml

@@ -0,0 +1,140 @@
+# Capability Metadata Schema
+# Location: adm/gdl/dev/schemas/capability-meta.schema.yaml
+#
+# Authority: This schema is the single source of truth for valid capability
+# metadata fields. All .meta.yaml files in crp/cap/ must conform to this schema.
+#
+# DSL Semantics:
+#   - type: scalar type (string, integer) or structural type (array, object).
+#   - required: whether the field must be present in every .meta.yaml file.
+#   - items: element schema for array types.
+#   - properties: exhaustive property declarations for object types.
+#     Objects must contain ONLY the declared properties — no additional
+#     properties are valid. (Resolves AC-09 "exactly" constraint.)
+#   - enum: closed set of allowed values.
+#   - format: semantic format constraint (e.g., "date" = ISO 8601 YYYY-MM-DD).
+#   - description: human-readable purpose for governance readability.
+#
+# Note (CA-1): The `distinguish_from` field currently exists in ~15 non-empty
+# entries across 3 formats in the corpus. Only the object-array format declared
+# here is the canonical target format. Migration risk for S2/S3 is Medium.
+#
+# Traceability (CA-3): The internal structure of `sources` entries (5 required
+# properties) is an architecture decision (U2), not an explicit AC requirement.
+
+schema:
+  version: "1.0.0"
+
+fields:
+  id:
+    type: string
+    required: true
+    description: >
+      Unique capability identifier. Must match the filename stem of the
+      corresponding .md file.
+
+  version:
+    type: string
+    required: true
+    description: >
+      Semantic version (semver) of the capability definition.
+
+  owner:
+    type: string
+    required: true
+    description: >
+      Semantic Authority over the cognitive contract. Identifies the party
+      responsible for the meaning, boundaries, and evolution of this capability.
+
+  do_not_use_when:
+    type: array
+    required: false
+    items:
+      type: string
+    description: >
+      Negative applicability boundaries. Each entry describes a situation
+      where this capability must not be selected.
+
+  distinguish_from:
+    type: array
+    required: false
+    items:
+      type: object
+      properties:
+        id:
+          type: string
+          required: true
+          description: >
+            Identifier of the capability to distinguish from.
+        boundary:
+          type: string
+          required: true
+          description: >
+            Explanation of the semantic boundary between this capability
+            and the referenced one.
+    description: >
+      Neighbor disambiguation entries. Each object identifies a related
+      capability and the semantic boundary separating them.
+
+  requires:
+    type: array
+    required: false
+    items:
+      type: string
+    description: >
+      Capability IDs that must be co-selected when this capability is active.
+
+  conflicts:
+    type: array
+    required: false
+    items:
+      type: string
+    description: >
+      Capability IDs that must not be co-selected with this capability.
+      Conflicts must be symmetric (if A conflicts B, B must conflict A).
+
+  # Architecture Decision U2: sources internal structure.
+  # All 5 properties are required per entry. This is an architecture decision
+  # (not an explicit AC requirement). The sources array itself is optional —
+  # authors who do not have full citations omit the field entirely.
+  sources:
+    type: array
+    required: false
+    items:
+      type: object
+      properties:
+        title:
+          type: string
+          required: true
+          description: >
+            Title of the reference material.
+        organization:
+          type: string
+          required: true
+          description: >
+            Publishing organization or authoring body.
+        url:
+          type: string
+          required: true
+          description: >
+            Location where the source can be accessed.
+        kind:
+          type: string
+          required: true
+          enum:
+            - standard
+            - architecture-guidance
+            - security-guidance
+            - adoption-guidance
+            - domain-reference
+          description: >
+            Classification of the source material type.
+        accessed_at:
+          type: string
+          required: true
+          format: date
+          description: >
+            Date when the source was last accessed or verified.
+    description: >
+      Authoritative sources underpinning the capability's cognitive contract.
+      Each entry must be fully specified for citation traceability.

adm/pbl/ai4x-context-budget-model.md

@@ -1,5 +1,7 @@
 # Context-Budget-Modell fuer Capability-Komposition
 
+> **Status:** Partially absorbed into Epic #37 (Story #72: `token_estimate` field). Budget ceiling and tiered truncation strategy remain parked — depend on `curate` pipeline being conceptually defined.
+
 ## Motivation
 
 Wenn `curate` Capabilities aus dem Corpus (`crp/cap/`) zu einem CCC komponiert, kann das resultierende Instruktionsset das Context-Fenster des Ziel-Agent-Hosts sprengen. Aktuell gibt es kein Modell, das dieses Risiko adressiert:

adm/pbl/ai4x-lifecycle-feedback-loop.md

@@ -1,5 +1,7 @@
 # Lifecycle Feedback Loop: evolve / assess
 
+> **Status:** Parked — depends on `curate` being conceptually defined. Revisit when `curate` design work begins.
+
 ## Motivation
 
 Das ai4X Operating Model ist aktuell ein Open-Loop-System: `curate` produziert ein Team, `spawn` materialisiert es, danach endet der Lifecycle. Es gibt keinen designten Mechanismus, um:

adm/pbl/ai4x-organization-design.md

@@ -1,5 +1,7 @@
 # Organization Design: Collaboration Topology Selection Capability
 
+> **Status:** Parked — depends on `curate` pipeline being conceptually defined. Revisit when `curate` design work begins.
+
 ## Motivation
 
 `curate` must perform a critical step that no current team member can execute: