Search markdown corpora using BM25 and vector embeddings with optional reranking.
npx skills add https://github.com/Motion-Creative/runneth-apps --skill corpus-search
Please help me install this Agent Skill. Skill: corpus-search Source: https://github.com/Motion-Creative/runneth-apps/tree/main/corpus-search Command: npx skills add https://github.com/Motion-Creative/runneth-apps --skill corpus-search
You need reliable ways to find relevant information inside large collections of markdown notes, transcripts, and summaries without manually reading everything.
Ask to install and index your markdown sources, then immediately run a natural-language query against the indexed kind to get ranked results back in under a second.
High-intent search queries and answers about installing and using this skill.
Markdown semantic search combines vector embeddings with BM25 keyword matching to find relevant passages by meaning rather than exact terms. This hybrid retrieval ranks chunks from your corpus based on conceptual similarity to your query intent.
Markdown search supports filtering by kind, role, workspace, and date ranges by ingesting YAML frontmatter metadata. You can narrow ranked retrieval results to specific document types or contextual attributes within your corpus.
Building a local markdown search index uses a sandbox-only SQLite store with FTS5 and sqlite-vec for BM25 and vector search. Indexing is idempotent, persistently storing corpus metadata and embeddings without requiring a separate server to run.
Searching markdown video scene summaries uses hybrid retrieval with timestamp extraction to trace results back to where they occurred. Querying natural-language intent against indexed summaries returns ranked passages with timestamps in under a second.
No, hybrid retrieval over markdown operates entirely within a workspace sandbox using a local SQLite database. It combines FTS5 for BM25 keyword matching and sqlite-vec for vector embeddings, requiring no separate server setup or external infrastructure.
Keyword search alone misses relevant transcripts because it relies on exact term matching. Hybrid retrieval solves this by adding vector embeddings to capture semantic meaning, ensuring passages with different wording but similar intent are still returned and ranked.
jdonohooCreate a standardized discovery folder structure with input and output directories.
RyderFreeman4LogosAnalyze Rust crates to generate facts.toml, README.md, review_report.md, and blog.md.
Yangel-hideSearch local knowledge stores with hybrid BM25, vector, and LLM reranking.
JoeyJoziahBuilds semantic vector search over document corpora with AgentDB.
