
from Agentic Dev Team256
Detect semantic duplication of business logic across architectural layers by annotating computation functions and clustering equivalent domain concepts.
Tool for detecting when the same business rule or domain computation has been independently reimplemented across different architectural layers (domain, application, presentation, infrastructure). Builds a computation register by identifying non-trivial functions, annotating them with normalized semantic descriptions, and clustering entries to surface duplicates. Includes pre-filtering rules to exclude trivial functions and test/build artifacts, an annotation prompt for consistent semantic descriptions, sharded clustering guidance, and canonical scoring to suggest preferred canonical locations for rules. Intended for codebase maintenance and refactoring assistance.
Semantic duplication scan skill with no bundled scripts — purely instructional SKILL.md defining a multi-step agent workflow for detecting business logic reimplemented across architectural layers. Uses LLM-powered annotation (Haiku), clustering (Sonnet), and canonical resolution (Opus). Well-structured process flow with pinned prompts, pre-filter rules, and clear output contracts. No security concerns — no network calls to suspicious hosts, no credential handling, no destructive commands.
No scripts to execute. Pure instructional skill. The multi-model approach (Haiku for annotation, Sonnet for clustering, Opus for disambiguation) is thoughtful but creates dependency on specific model tiers. Token budget partitioning strategy for large registers is well-designed.