novelty-gap-check
Novelty Gap Check
Overview
This skill adjudicates how novel a research idea is against recent literature, then hands the user an advisory go/no-go. It does three things in order. First it decomposes the idea into three to five core technical claims. Then it searches prior work per claim using the retrieval skills this plugin already ships. Finally it emits a delta verdict: a per-claim HIGH/MED/LOW novelty table, an overall PROCEED / PROCEED-WITH-CAUTION / ABANDON call, and suggested positioning.
This is the verdict and positioning layer, not a retrieval engine. Retrieval is delegated. The skill invokes superpower-writing:research-lookup for fast per-claim queries and superpower-writing:literature-review for a deeper sweep when the idea spans several subfields. Do not reimplement search backends here. The value this skill adds is the judgment on top: closest prior work per claim, the delta, and how to frame the contribution.
Core principle: be brutally honest. A false novelty claim wastes months of research time. Surfacing a strong prior-work overlap early is the most valuable thing this skill does.
Advisory only. The verdict is surfaced to the user. It never auto-rejects an idea, never mutates .writing/ state, and never flips claim STATUS. The user decides whether to proceed, pivot, or abandon.
Never fabricate. Every paper named as prior work must be verified before it enters the report. Unverified candidates are tagged [UNVERIFIED] and surfaced, never dropped and never invented. This mirrors the plugin's claim-first rule. Full protocol in references/citation-discipline.md.