Finds groups of classes in the same directory subtree that share the same overall method-shape (same method names, visibilities, and order) and have high per-method body similarity. These groups are candidates for extraction into a shared base class.

What it catches

Existing detectors (duplicate_function, near_duplicate, parallel_implementation) compare pairs of functions. They miss class-level scaffolding — where dozens of classes each follow the same skeleton with small variations.

For example, a plugin with ~90 ability classes all following __construct → registerAbility → execute → checkPermission surfaces as one shared_scaffolding finding, not 90 unrelated parallel-implementation pairs.

Algorithm

1. For each fingerprinted file that declares a class, build an ordered shape signature of (method_name, visibility) tuples using FileFingerprint.methods and FileFingerprint.visibility.
2. Bucket classes by (subtree_root, shape_signature). The subtree root is the first two path components under the component root (e.g. files under inc/Abilities/** share the inc/Abilities bucket).
3. For each bucket with ≥ 3 classes, compute mean per-method body similarity using FileFingerprint.method_hashes. For each method in the shape, the similarity is the size of the largest identical-hash bucket divided by the member count.
4. If the mean similarity is ≥ 60%, emit one finding (severity warning, kind shared_scaffolding) describing the group.

Finding fields

• file: the subtree root (e.g. inc/Abilities).
• description: member count, shape, mean body similarity, identical method-body count, member class list, and an estimated LOC reduction.
• suggestion: proposes extracting a shared base class under the subtree.

Thresholds

Related detectors

• duplicate_function / near_duplicate — function-level body similarity.
• parallel_implementation — similar call patterns across function pairs.
• shadow_module — whole directories that are near-copies.