gist-retriever

What problem does it solve?

This Skill solves the problem of reliably retrieving relevant evidence from a large knowledge corpus by combining lexical, dense, and graph-adjacent neighborhood expansion into a single candidate pipeline.

Core Features & Use Cases

• Multi-tier Retrieval Cascade (L1-first): Routes queries through a tiered fallback approach that escalates from wiki/markdown to local memory retrieval and finally to deep-research.
• Hybrid Seed Retrieval and Fusion: Builds an initial candidate pool using BM25 and dense GIST/semantic retrieval, then fuses results with RRF to preserve ranking diversity.
• L2 Neighborhood Expansion + Late Reranking: Expands seeds into a local semantic neighborhood (BM25 triplet expansion + dense centroid expansion) and applies ColBERT late interaction for higher-precision final ranking.
• Reconstruction and Final Selection: Reconstructs candidates into coherent units (subclass-specific) and selects the final set according to a defined stopping rule.

Quick Start

Use gist-retriever to retrieve a high-quality evidence candidate set for a query before running syllogistic reasoning or answer synthesis.