iatomicreactori/fovea/spatial

viagent_routing:
  version: 1
  subskills:
    - id: ela
      when: >-
        Error Level Analysis along face/foreground boundaries. Use for almost all
        videos; best default for composite edges and halo artifacts. Safe general
        spatial check.
    - id: patch
      when: >-
        Local patch texture inconsistency inside facial regions. Prefer when faces
        or skin are visible; less informative for distant landscapes with no faces.
    - id: boundary
      when: >-
        Edge melting / abnormal face-to-background transition. Prefer when there is
        a clear subject and background; skip or deprioritize for flat abstract
        graphics or text-only slides.
    - id: edge
      when: >-
        Local edge coherence (soft/noisy boundaries, possible floating objects).
        General spatial cue; useful for natural scenes and talking-head videos.
    - id: multi_face
      when: >-
        Multi-face geometric collapse / inconsistency across frames. Run only when
        multiple faces or crowded people are plausible; omit for single-subject
        tutorials, screens, gameplay without faces, or nature-only footage.
    - id: blur
      when: >-
        Laplacian blur uniformity (face vs background sharpness). Requires face
        regions for full value; pair with multi_face when possible. Omit when no
        faces expected (reduces misleading ratios).