hirokidaichi/marp-lens

Overview

This skill provides semantic, vector-based search for Marp presentation slides so you can find relevant slides by topic, retrieve slide content, and explore presentation materials quickly. It indexes Markdown slides into embeddings and returns similarity-ranked results with previews, slide numbers, and metadata. Use it from the CLI to search, fetch specific slides, reindex files, or view database stats.

How this skill works

The skill parses Marp Markdown files into individual slides, generates vector embeddings for slide text (optionally including image descriptions), and stores them in a local vector database. A semantic search query is converted into an embedding and matched against stored vectors to produce ranked results with similarity scores and content previews. Commands allow fetching a specific slide by file and number, rebuilding or updating the index, and exporting results as table or JSON.

When to use it

• Locate slides covering a concept when building a talk or study notes
• Quickly retrieve the full content of a specific slide by reference
• Explore which presentations mention a keyword or topic across a slide library
• Reindex after adding or changing slide files to keep search results current
• Export search results as JSON for automation or integration

Best practices

• Index your slide directory after adding or editing files to ensure up-to-date results
• Use concise, topic-focused queries to improve match relevance
• Adjust the similarity threshold to filter noisy matches for high-precision results
• Include image descriptions when indexing slides that rely on visuals for better semantic coverage
• Use JSON output for programmatic workflows and table output for quick human review

Example use cases

• Search a slide library for "machine learning basics" and open top-matching slides for reuse
• Run a query like "what slides cover deployment" to prepare a training session or checklist
• Fetch a specific slide with its full content using the file path and slide number for copying into another deck
• Rebuild the index after reorganizing a folder of presentations to refresh embeddings and counts
• Integrate search output (JSON) into a documentation site or knowledge base for slide discovery