What problem does it solve? After running repo analysis, findings candidates sit in analysis docs with no structured path into the knowledge base, risking duplicates and lost insights. This Skill bridges that gap by reading candidates, deduplicating against existing findings, and writing formally structured Research Finding entries only after user approval. ## Core Features & Use Cases - Candidate Extraction & Deduplication: Reads Findings Candidates sections from single, all, or cross-repo comparison analysis docs and greps the KB to classify each as New, Partial match, or Full duplicate. - Human-Gated KB Writes: Presents candidates in a selection table; even with --auto, a single batch confirmation gate precedes any write, so nothing enters the KB without in-run approval. - Schema-Compliant Finding Generation: Writes findings with full YAML frontmatter (evidence strength, priority triage P1-P3, related_findings links) and standardized body sections, then annotates every candidate in the source analysis doc with its disposition. - Use Case: After running /repo-analyzer on three watched libraries, invoke this Skill to review 12 candidates, promote 8 as new findings, link 2 partial matches, and mark 2 duplicates — all traceable via bidirectional Obsidian links. ## Quick Start Ask the AI to promote findings candidates from the gsd analysis doc into the Research Findings KB and review the deduplication table before approving.
