What problem does it solve? Research projects accumulate dozens of markdown artifacts—experiment logs, lab notebooks, postmortems, paper drafts—that are hard to navigate as plain text: loss curves stay invisible in JSONL files, 100+ notebook entries cannot be reverse-looked-up by tag, and phase progress is not visible at a glance. This Skill converts those artifacts into a browsable static HTML site while keeping the original markdown files untouched as the source of truth. ## Core Features & Use Cases - MkDocs Material site generation: Auto-generates mkdocs.yml, copies and transforms source MD files into viz-src/docs/, and builds a static site with search, dark mode, and tag-based reverse lookup via uvx (no fixed installation). - Chart.js time-series visualization: Parses events.jsonl to render loss curves and metric charts on per-run pages, plus a Phase 1-8 progress bar injected from STATE.json into every page header. - Sortable tables and tag inversion: Converts run indexes, survey matrices, and results tables into sortable HTML tables, and transforms lab notebook Tags into mkdocs frontmatter for automatic tag index pages. - Use Case: After completing several experiment runs in an LLM research project, invoke the skill to produce .research/<slug>/viz/index.html, open it in a browser, and navigate brief, survey, ideas, runs, postmortems, results, and paper draft sections with full-text search. ## Quick Start Run /auto-research:notebook-viz <slug> to build the HTML research notebook for your project, optionally adding --serve to preview it at localhost:8000.
