Distill raw paper-reading drafts into atomic notes and a cross-linked wiki. Use when the user adds new files to `drafts/`, gives a URL to a paper/blog/website to ingest, or asks to "update the wiki", "rebuild the knowledge map", "ingest this link", or "redistill my notes".
This is an external-memory pipeline for paper-notes/, inspired by Karpathy's drafts → distill → wiki workflow.
drafts/ raw, unedited dumps (one file per source — paper, blog, talk)
notes/ atomic distilled notes, one per source, Obsidian wikilinks
wiki/ synthesis layer
index.md top-level knowledge map + tag index
<topic>.md per-cluster synthesis pages
drafts/ is append-only from the user. notes/ and wiki/ are regenerated/extended by this skill — never hand-edited by the user.
drafts/ → distill them.drafts/<category>/<slug>.md, then distill.WebFetch first — fastest for static pages and arxiv abstracts.WebFetch fails, returns paywalled/JS-rendered content, or the user explicitly says "use playwright", fall back to the Playwright MCP tools (mcp__plugin_playwright_playwright__browser_navigate then browser_snapshot or browser_evaluate to extract text). For paywalled academic content, try the cookie-extract skill to get an authenticated session.drafts/<category>/<kebab-slug>.md with a header:
---
source_url: <url>
fetched: <YYYY-MM-DD>
---
<category> to match existing subfolders (e.g. neuroscience/); create a new one if no fit.For each new (or changed) file in drafts/, write notes/<slug>.md with this template:
---
Edit PDFs with natural-language instructions using the nano-pdf CLI.