Support start-end frame image2video routing

The CLI now submits gateway-compatible first/final-frame payloads, so the server accepts the same public SkillHub schema fields, normalizes them to images[0]/images[1] with metadata.action=firstTailGenerate, and routes multi-frame Vidu-style submissions through the unified video generations endpoint.

Constraint: Align server behavior with popiartcli v0.3.21 start/end-frame payloads

Constraint: Keep existing single-image image2video and MiniMax paths compatible

Rejected: Hardcode only the new fields in server seed skills | loading SkillHub input_schema.json keeps the remote catalog as the source of truth

Confidence: medium

Scope-risk: moderate

Directive: Preserve first-frame then final-frame ordering when changing video reference handling

Tested: go test ./...

Tested: make build

Tested: GOOS=linux GOARCH=amd64 go build -o dist/popiartserver-linux-amd64 ./cmd/popiartserver

Not-tested: Live test-server deployment; current SSH key is rejected by 101.42.99.35

internal/server/skillhub.go

@@ -62,6 +62,18 @@ func loadSkills(skillhubDir string) ([]skill, error) {
 		tags := tagsFromSkillID(entry.Name)
 		name := defaultString(strings.TrimSpace(doc.Title), displayNameFromSkillID(entry.Name))
 		description := strings.TrimSpace(doc.Description)
+		inputSchema := inputSchemaForRoute(routeKey)
+		if schema, ok, err := readSkillSchemaFile(filepath.Join(skillhubDir, skillPath, "input_schema.json")); err != nil {
+			return nil, fmt.Errorf("read input schema for %s: %w", entry.Name, err)
+		} else if ok {
+			inputSchema = schema
+		}
+		outputSchema := defaultOutputSchema()
+		if schema, ok, err := readSkillSchemaFile(filepath.Join(skillhubDir, skillPath, "output_schema.json")); err != nil {
+			return nil, fmt.Errorf("read output schema for %s: %w", entry.Name, err)
+		} else if ok {
+			outputSchema = schema
+		}
 
 		items = append(items, skill{
 			ID:                 defaultString(strings.TrimSpace(doc.Name), entry.Name),
@@ -72,8 +84,8 @@ func loadSkills(skillhubDir string) ([]skill, error) {
 			ModelType:          category,
 			RouteKey:           routeKey,
 			EstimatedDurationS: estimateDuration(routeKey),
-			InputSchema:        inputSchemaForRoute(routeKey),
-			OutputSchema:       defaultOutputSchema(),
+			InputSchema:        inputSchema,
+			OutputSchema:       outputSchema,
 		})
 	}
 
@@ -83,6 +95,24 @@ func loadSkills(skillhubDir string) ([]skill, error) {
 	return items, nil
 }
 
+func readSkillSchemaFile(path string) (map[string]any, bool, error) {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, false, nil
+		}
+		return nil, false, err
+	}
+	var schema map[string]any
+	if err := json.Unmarshal(data, &schema); err != nil {
+		return nil, false, err
+	}
+	if len(schema) == 0 {
+		return nil, false, nil
+	}
+	return schema, true, nil
+}
+
 type skillDoc struct {
 	Name        string
 	Description string
@@ -249,19 +279,7 @@ func inputSchemaForRoute(routeKey string) map[string]any {
 			},
 		}
 	case "video.image2video":
-		return map[string]any{
-			"type": "object",
-			"properties": map[string]any{
-				"source_artifact_id": map[string]any{"type": "string"},
-				"image_url":          map[string]any{"type": "string"},
-				"prompt":             map[string]any{"type": "string"},
-				"motion_prompt":      map[string]any{"type": "string"},
-				"duration_s":         map[string]any{"type": "integer"},
-				"camera_motion":      map[string]any{"type": "string"},
-				"mood":               map[string]any{"type": "string"},
-				"aspect_ratio":       map[string]any{"type": "string"},
-			},
-		}
+		return imageToVideoInputSchema()
 	default:
 		return map[string]any{
 			"type": "object",
@@ -275,3 +293,34 @@ func inputSchemaForRoute(routeKey string) map[string]any {
 		}
 	}
 }
+
+func imageToVideoInputSchema() map[string]any {
+	return map[string]any{
+		"type": "object",
+		"properties": map[string]any{
+			"source_artifact_id":     map[string]any{"type": "string"},
+			"image_url":              map[string]any{"type": "string"},
+			"reference_image_url":    map[string]any{"type": "string"},
+			"last_frame_image_url":   map[string]any{"type": "string"},
+			"end_frame_image_url":    map[string]any{"type": "string"},
+			"last_frame_artifact_id": map[string]any{"type": "string"},
+			"end_frame_artifact_id":  map[string]any{"type": "string"},
+			"images":                 map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
+			"metadata":               map[string]any{"type": "object"},
+			"prompt":                 map[string]any{"type": "string"},
+			"motion_prompt":          map[string]any{"type": "string"},
+			"negative_prompt":        map[string]any{"type": "string"},
+			"duration_s":             map[string]any{"type": "number"},
+			"seconds":                map[string]any{"type": "number"},
+			"fps":                    map[string]any{"type": "number"},
+			"camera_motion":          map[string]any{"type": "string"},
+			"motion_intensity":       map[string]any{"type": "string"},
+			"style":                  map[string]any{"type": "string"},
+			"size":                   map[string]any{"type": "string"},
+			"mood":                   map[string]any{"type": "string"},
+			"aspect_ratio":           map[string]any{"type": "string"},
+			"seed":                   map[string]any{"type": "number"},
+			"notes":                  map[string]any{"type": "string"},
+		},
+	}
+}