What problem does it solve? Pre-classified trope candidates from fast local classifiers often include false positives, such as casual metaphors or literal word matches, and cannot distinguish canonical clichés from creative subversions. This Skill adds a second validation layer that uses a local LLM with the full plot synopsis to classify each candidate accurately. ## Core Features & Use Cases - Contextual Trope Validation: Combines the trope's canonical definition, the chapter snippet, and the global story synopsis into a structured prompt for a local instruct LLM (e.g., Phi-3.5-mini-instruct). - Ternary Classification: Labels each occurrence as PRESENTE (canonical cliché), AUSENTE (false positive), or VARIANTE (creative subversion), returning structured JSON with confidence and justification. - Subversion Mapping: Flags VARIANTE results in the writer's interface as creative originality and logs PRESENTE results into cliché density metrics. - Use Case: After the level-1 classifier flags "Mentor Death" in a chapter, this Skill checks the synopsis confirming Valdris is the hero's mentor, then validates whether the scene is a true trope occurrence or a subversion. ## Quick Start Run the level-2 trope validation on the candidates generated for the current chapter using the compiled story synopsis.
