feat(models): manifest-driven model capabilities and curated labels

.github/workflows/ci.yml

@@ -28,7 +28,7 @@
       web: ${{ steps.filter.outputs.web }}
       mobile: ${{ steps.filter.outputs.mobile }}
     steps:
       - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
         with:
           fetch-depth: 2
       - uses: dorny/paths-filter@7b450fff21473bca461d4b92ce414b9d0420d706 # v4.0.2
@@ -130,6 +130,50 @@
       - name: Unit tests
         run: just test-unit
 
+  model-capabilities:
+    name: Model Capabilities (regen + corpus + schema + buzz-agent tests)
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    permissions:
+      contents: read
+    steps:
+      - uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6
+      - uses: cashapp/activate-hermit@cea9af7913204a965fd488637a8d1811bba2e616 # v1
+      - uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
+        with:
+          save-if: ${{ github.event_name != 'pull_request' }}
+
+      - name: Set up Node.js
+        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        with:
+          node-version: '24'
+          package-manager-cache: false
+
+      - name: Regenerate artifacts
+        run: node scripts/generate-model-capabilities.mjs
+
+      - name: Diff check — fail if generated files are stale
+        run: |
+          if ! git diff --exit-code \
+            crates/buzz-agent/src/generated_model_capabilities.rs \
+            desktop/src/features/agents/ui/modelCapabilities.ts; then
+            echo ""
+            echo "ERROR: Generated model-capability files are stale."
+            echo "Run: node scripts/generate-model-capabilities.mjs"
+            echo "Then commit the regenerated files."
+            exit 1
+          fi
+          echo "✓ All generated files are up to date."
+
+      - name: Run corpus (TS interpreter via --experimental-strip-types)
+        run: node --experimental-strip-types scripts/run-corpus.mjs
+
+      - name: Validate manifest (schema-negative tests)
+        run: node --test scripts/test-manifest-validator.mjs
+
+      - name: Run buzz-agent unit tests (normative corpus + generated interpreter)
+        run: cargo test -p buzz-agent --lib
+
   desktop-core:
     name: Desktop Core
     runs-on: ubuntu-latest

crates/buzz-agent/src/catalog.rs

@@ -14,6 +14,8 @@
 
 use std::sync::Arc;
 
+use crate::generated_model_capabilities::DATABRICKS_MODEL_NAMES;
+
 use reqwest::Client;
 
 use crate::{
@@ -23,8 +25,25 @@ use crate::{
     types::AgentError,
 };
 
+/// Returns the curated display name for a Databricks endpoint ID, or the raw
+/// ID when no entry exists in the registry.
+///
+/// The registry (`DATABRICKS_MODEL_NAMES`) is generated from the manifest
+/// (`scripts/model-capabilities.json`) exact records for the `databricks_v2`
+/// provider and covers the ~30 managed Databricks endpoints. Any custom or
+/// workspace endpoint not in the table is returned untouched — no heuristic
+/// guessing.
+pub(crate) fn databricks_model_name(id: &str) -> &str {
+    DATABRICKS_MODEL_NAMES
+        .iter()
+        .find(|(k, _)| *k == id)
+        .map(|(_, v)| *v)
+        .unwrap_or(id)
+}
+
 /// A discovered model entry: `id` is the picker value, `name` is the display
-/// label (same as `id` for Databricks — the API has no separate display name).
+/// label (curated from models.dev for known managed endpoints; raw ID for
+/// custom/unknown endpoints).
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct ModelEntry {
     pub id: String,
@@ -34,8 +53,10 @@ pub struct ModelEntry {
 /// Known Databricks AI Gateway v2 models — used only when an authenticated
 /// `api/ai-gateway/v2/endpoints` call succeeds with an empty list.
 /// Mirrors goose's `DATABRICKS_V2_KNOWN_MODELS`.
-pub const DATABRICKS_V2_KNOWN_MODELS: &[&str] =
-    &["databricks-gpt-5-5", "databricks-claude-opus-4-7"];
+///
+/// Phase 2 cutover: this is now a re-export of the generated constant in
+/// `generated_model_capabilities`. Phase 3 removes the old hand-maintained list.
+pub use crate::generated_model_capabilities::DATABRICKS_V2_KNOWN_MODELS;
 
 const AUTHENTICATED_EMPTY_CATALOG_SUFFIX: &str = " (default catalog)";
 
@@ -44,7 +65,10 @@ fn authenticated_empty_v2_catalog() -> Vec<ModelEntry> {
         .iter()
         .map(|id| ModelEntry {
             id: id.to_string(),
-            name: format!("{id}{AUTHENTICATED_EMPTY_CATALOG_SUFFIX}"),
+            name: format!(
+                "{}{AUTHENTICATED_EMPTY_CATALOG_SUFFIX}",
+                databricks_model_name(id)
+            ),
         })
         .collect()
 }
@@ -206,7 +230,7 @@ pub(crate) fn parse_v1_endpoints(json: &serde_json::Value) -> Result<Vec<ModelEn
 
             Some(ModelEntry {
                 id: name.clone(),
-                name,
+                name: databricks_model_name(&name).to_string(),
             })
         })
         .collect();
@@ -371,7 +395,7 @@ pub(crate) fn parse_v2_endpoints_page(
             Some(V2Endpoint {
                 entry: ModelEntry {
                     id: name.clone(),
-                    name,
+                    name: databricks_model_name(&name).to_string(),
                 },
                 created_ms: endpoint_created_ms(endpoint),
             })
@@ -649,8 +673,12 @@ mod tests {
         let ids: Vec<&str> = models.iter().map(|model| model.id.as_str()).collect();
 
         assert_eq!(ids, DATABRICKS_V2_KNOWN_MODELS);
+        // Each entry carries the curated display name (not the raw ID) plus the
+        // " (default catalog)" suffix to distinguish this authenticated-but-empty
+        // slate from a live discovery result.
         assert!(models.iter().all(|model| {
-            model.name == format!("{}{AUTHENTICATED_EMPTY_CATALOG_SUFFIX}", model.id)
+            let curated = databricks_model_name(&model.id);
+            model.name == format!("{curated}{AUTHENTICATED_EMPTY_CATALOG_SUFFIX}")
         }));
     }
 
@@ -665,4 +693,50 @@ mod tests {
         assert!(!is_chat_capable_endpoint("databricks-gte-large-en"));
         assert!(!is_chat_capable_endpoint("databricks-qwen3-embedding-0-6b"));
     }
+    // ---------------------------------------------------------------------------
+    // Databricks model name registry
+    // ---------------------------------------------------------------------------
+
+    #[test]
+    fn databricks_model_name_known_id_returns_curated_name() {
+        assert_eq!(databricks_model_name("databricks-gpt-5-5"), "GPT-5.5");
+        assert_eq!(
+            databricks_model_name("databricks-claude-opus-4-7"),
+            "Claude Opus 4.7"
+        );
+        assert_eq!(
+            databricks_model_name("databricks-gpt-oss-120b"),
+            "GPT OSS 120B"
+        );
+    }
+
+    #[test]
+    fn databricks_model_name_unknown_custom_endpoint_returns_raw_id() {
+        // Custom workspace endpoints must never be guessed — pass through unchanged.
+        assert_eq!(
+            databricks_model_name("databricks-team-2025-01"),
+            "databricks-team-2025-01"
+        );
+        assert_eq!(
+            databricks_model_name("databricks-finance-2025-01-30"),
+            "databricks-finance-2025-01-30"
+        );
+        assert_eq!(
+            databricks_model_name("some-unknown-endpoint"),
+            "some-unknown-endpoint"
+        );
+    }
+
+    #[test]
+    fn v2_known_models_fallback_entries_get_curated_names() {
+        // The DATABRICKS_V2_KNOWN_MODELS constant lists IDs that are in the
+        // registry, so their fallback entries must carry curated names.
+        for id in DATABRICKS_V2_KNOWN_MODELS {
+            let name = databricks_model_name(id);
+            assert_ne!(
+                name, *id,
+                "known model {id} should have a curated name, not raw ID"
+            );
+        }
+    }
 }