vault-hygiene-checkerlisted

# Vault Hygiene Checker ## Why this skill exists This is the **checker** in Osmani's maker/checker loop pattern, applied to the user's Zettelkasten instead of to code. The agent that *writes* permanent notes and MOC links is "too nice grading its own homework" — it claims a note links to 5 atoms without verifying those atoms exist, links resolve, or the note isn't a near-duplicate. Your own vault history has the cautionary tale (the "integration suite was always red — review-theater lesson": 11 PRs merged green that were actually broken). Same failure mode, different domain. This skill runs a **separate, skeptical pass** whose job is to *refute* the synthesis, then report findings to the user for approval. you stay the decision-maker — findings come to your, fixes happen only on your yes. It also pairs with the Karpathy "Goal-Driven Execution" principle in your CLAUDE.md: a synthesis pass should have a verifiable stop condition ("every new permanent note links to ≥2 existing atoms, is in exactly one MOC, quotes ≥1 verbatim source"), and this checker is what verifies it. ## Two halves: mechanical (automated) + judgment (read) **Mechanical checks** — objective, zero judgment, run via the script. These can run unattended: ```bash python3 ~/.claude/skills/vault-hygiene-checker/check_vault.py --since YYYY-MM-DD # or: --days N | --files "Note A" "Note B" | --scope "Resources/Permanent Notes" # add --fix-reciprocal to WRITE the missing note→MOC backlinks (see check 6