mm

name	mm
description	Use this skill when the user wants to work with multimodal files (images, video, audio, PDFs, binary formats) — find, inspect, extract content, search, or count files using UNIX-style commands with multimodal powers.

mm Plugin

Fast, multimodal context for agents. Familiar UNIX CLI tools like find, grep, cat — with multimodal powers. Rust core for speed, Python for dev-ex, UNIX philosophy for composability.

mm enables agents to work with file types that LLMs can't natively read, including images, video, audio, PDFs, and other binary formats. Indexing is implicit — every command auto-builds a metadata index on first use.

Commands

Discovery

• mm find list --path <dir> — Find/list files with multimodal metadata (tree, tabular, schema, JSON)
• mm find list --path <dir> --tree — Hierarchical tree view with sizes
• mm find list --path <dir> --kind image — Filter by kind (image, video, audio, document, code, text)
• mm wc count --path <dir> --by_kind — Count files, bytes, lines, tokens by kind
• mm peek metadata --path <file> — Raw file metadata (dimensions, EXIF, codec, duration, mime, hash)

Content extraction

• mm cat extract --path <file> — Extract content (PDF text, image captions, video descriptions, audio transcripts)
• mm cat extract --path <file> --mode accurate — Full LLM pipeline (requires configured profile)
• mm cat extract --path <file> --lines 20 — First N lines (head)

Search

• mm grep search --pattern <query> --path <dir> — Text + semantic content search
• — Semantic (vector) search