These three open-source Claude skills audit, measure, and repair an Obsidian PKM vault ("second brain"). They were battle-tested on a real 9,700-note vault before release: orphan rate dropped 32.2% to 12.0%, link density rose 2.59 to 3.49 links/note, and the worst super-hub's edge-share fell 10.9% to 1.14%. Total LLM spend for the full cleanup was about $6, because most operations are $0 deterministic.
All three are MIT licensed, authored by @build-with-dhiraj. Added to Productivity & Organization, alphabetized.
What each one does:
- obsidian-graph-auditor answers "is my Obsidian graph healthy, or is it a hairball?" It scores a vault on an 8-dimension PKM rubric: link density, orphan rate, near-orphan rate, connected-greater-than-2 share, top-hub edge-share, top-hub-to-next ratio, Louvain modularity, and frontmatter-wikilink adoption. Read-only, deterministic, JSON output for CI.
- obsidian-brain-eval answers "can my second brain actually find what I know?" It generates a gold Q&A set from your notes, then scores Recall@10 with a naive BM25 or LanceDB backend. The author's vault scored Recall@10 of 0.95.
- obsidian-orphan-killer answers "how do I fix orphan notes in Obsidian without hand-linking thousands of files?" It resolves plain-text entity mentions into wikilinks deterministically ($0), anchors remaining orphans via an embedding floor of 0.74 (concepts only), and can mint cluster hub notes (experimental, gated). 12 safety guards, frontmatter-only mutations, atomic writes, idempotent.
Checklist