Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Audit your Obsidian vault in Claude Code — finds stale drafts, empty folders, duplicate filenames, and incomplete files. Saves a dated report.
.claude/skills/brianrwagner-vault-cleanup-auditor/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 109% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 166% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 183% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 58% | 0% |
Runs 4 targeted checks against your Obsidian vault and saves a prioritized audit report to vault-audit/YYYY-MM-DD-audit.md. Takes under 30 seconds. No APIs, no paid services.
Run it monthly or whenever the vault starts feeling messy.
Open Claude Code and say:
Run the Vault Cleanup Auditor skill against my vault at /path/to/vault.When this skill is invoked, follow these phases exactly.
Check whether the user has provided:
vault_path — absolute path to their Obsidian vault rootIf missing, ask:
What's the absolute path to your Obsidian vault? (e.g. /root/obsidian-vault)Do not proceed until confirmed.
Run all 4 checks. Capture raw output from each before writing the report.
Check 1 — Stale drafts (30+ days, unposted):
bashVAULT="VAULT_PATH_HERE" echo "=== CHECK 1: STALE DRAFTS ===" find "$VAULT/content/ready-to-post" -name "*.md" -mtime +30 2>/dev/null | while read f; do if grep -q '\*\*Posted:\*\* ❌' "$f" 2>/dev/null; then days=$(( ($(date +%s) - $(stat -c %Y "$f")) / 86400 )) echo "${days}d|${f##$VAULT/}" fi done echo "=== END CHECK 1 ==="
Check 2 — Incomplete files (no headings, 5+ lines):
bashVAULT="VAULT_PATH_HERE" echo "=== CHECK 2: INCOMPLETE FILES ===" find "$VAULT" -name "*.md" \ -not -path "*/.git/*" \ -not -path "*/.obsidian/*" \ -not -path "*/.openclaw/*" | while read f; do if ! grep -qE '^#{1,6} ' "$f" 2>/dev/null; then lines=$(wc -l < "$f") if [ "$lines" -gt 5 ]; then echo "${lines}lines|${f##$VAULT/}" fi fi done echo "=== END CHECK 2 ==="
Check 3 — Duplicate filenames:
bashVAULT="VAULT_PATH_HERE" echo "=== CHECK 3: DUPLICATE FILENAMES ===" find "$VAULT" -name "*.md" \ -not -path "*/.git/*" \ -not -path "*/.obsidian/*" \ -not -path "*/.openclaw/*" \ -printf "%f\n" | sort | uniq -d | while read name; do echo "DUPE:$name" find "$VAULT" -name "$name" \ -not -path "*/.git/*" \ -not -path "*/.obsidian/*" | while read f; do echo " PATH:${f##$VAULT/}" done done echo "=== END CHECK 3 ==="
Check 4 — Empty folders:
bashVAULT="VAULT_PATH_HERE" echo "=== CHECK 4: EMPTY FOLDERS ===" find "$VAULT" -type d -empty \ -not -path "*/.git/*" \ -not -path "*/.obsidian/*" \ -not -path "*/.openclaw/*" | while read d; do echo "${d##$VAULT/}/" done echo "=== END CHECK 4 ==="
3a. Create the report directory and file:
bashVAULT="VAULT_PATH_HERE" DATE=$(date +%Y-%m-%d) mkdir -p "$VAULT/vault-audit" REPORT="$VAULT/vault-audit/${DATE}-audit.md" echo "Report path: $REPORT"
3b. Write the report using this exact structure:
markdown# Vault Cleanup Audit — YYYY-MM-DD Generated: YYYY-MM-DD HH:MM **Summary:** 🔴 [N] stale | 🟡 [N] incomplete | 🟡 [N] dupes | 🟢 [N] empty --- ## 🔴 Stale Drafts (30+ days, unposted) - [Nd old] path/to/file.md [cap at 15 items — add "...and N more" if over] ## 🟡 Incomplete Files (no headings, 5+ lines) - path/to/file.md (N lines) [cap at 15] ## 🟡 Duplicate Filenames - **filename.md** - path/one/filename.md - path/two/filename.md [cap at 15 duplicate names] ## 🟢 Empty Folders - path/to/folder/ [cap at 15] --- **What to do:** - 🔴 Stale drafts: post them, delete them, or archive to `content/posted/` - 🟡 Incomplete files: finish or trash - 🟡 Duplicates: merge or rename the less important one - 🟢 Empty folders: safe to delete (or keep if you're about to use them)
3c. Print the summary to terminal:
✅ Audit saved to: vault-audit/YYYY-MM-DD-audit.md
Summary:
🔴 [N] stale drafts
🟡 [N] incomplete files
🟡 [N] duplicate filenames
🟢 [N] empty foldersBefore finalizing, check:
=== CHECK N === / === END CHECK N === marker. If a check produced no output, note "0 issues found" in the report — do not skip the section.vault-audit/YYYY-MM-DD-audit.md. If the write failed, report the error and provide the report content inline.Fix any failures before delivering the final output.
markdown# Vault Cleanup Audit — 2026-03-01 Generated: 2026-03-01 08:14 **Summary:** 🔴 3 stale | 🟡 2 incomplete | 🟡 1 dupe | 🟢 2 empty --- ## 🔴 Stale Drafts (30+ days, unposted) - [47d old] content/ready-to-post/linkedin/2026-01-12-ai-systems-post.md - [38d old] content/ready-to-post/twitter/2026-01-21-founder-ops.md - [31d old] content/ready-to-post/newsletter/2026-01-28-tools-roundup.md ## 🟡 Incomplete Files (no headings, 5+ lines) - bambf/research/untitled-notes.md (23 lines) - thinking/random-thoughts.md (11 lines) ## 🟡 Duplicate Filenames - **README.md** - bambf/brand/README.md - bambf/tracking/README.md ## 🟢 Empty Folders - content/queue/rejected/ - bambf/clients/archived/ --- **What to do:** - 🔴 Stale drafts: post them, delete them, or archive to `content/posted/` - 🟡 Incomplete files: finish or trash - 🟡 Duplicates: merge or rename the less important one - 🟢 Empty folders: safe to delete (or keep if you're about to use them)
find, grep, wc, stat, date, awkvault-audit/ inside the vault (auto-created)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 14,673 | 11,226 | -23% | 1 | 1 | 0% | 2,545 | 3,960 | +56% | 0 | 0 | — |
case-05 | pass→pass | 12,516 | 25,895 | +107% | 1 | 1 | 0% | 2,385 | 7,014 | +194% | 0 | 0 | — |
case-03 | fail→fail | 5,013 | 13,251 | +164% | 1 | 1 | 0% | 262 | 2,810 | +973% | 0 | 0 | — |
case-08 | pass→fail | 6,106 | 14,460 | +137% | 1 | 1 | 0% | 1,163 | 2,981 | +156% | 0 | 0 | — |
case-13 | pass→pass | 10,591 | 11,264 | +6% | 1 | 1 | 0% | 1,909 | 4,277 | +124% | 0 | 0 | — |
case-06 | pass→pass | 11,513 | 9,546 | -17% | 1 | 1 | 0% | 2,115 | 3,696 | +75% | 0 | 0 | — |
case-07 | fail→pass | 14,791 | 1,813 | -88% | 1 | 1 | 0% | 2,584 | 2,202 | -15% | 0 | 0 | — |
case-01 | fail→fail | 20,823 | 13,315 | -36% | 1 | 1 | 0% | 3,701 | 3,381 | -9% | 0 | 0 | — |
case-02 | fail→fail | 7,172 | 15,357 | +114% | 1 | 1 | 0% | 689 | 3,008 | +337% | 0 | 0 | — |
case-09 | fail→fail | 11,086 | 17,066 | +54% | 1 | 1 | 0% | 1,985 | 3,021 | +52% | 0 | 0 | — |
case-10 | pass→fail | 19,282 | 9,251 | -52% | 1 | 1 | 0% | 2,980 | 3,228 | +8% | 0 | 0 | — |
case-11 | fail→pass | 8,912 | 11,060 | +24% | 1 | 1 | 0% | 1,481 | 3,096 | +109% | 0 | 0 | — |
case-12 | pass→pass | 9,013 | 5,448 | -40% | 1 | 1 | 0% | 1,567 | 2,926 | +87% | 0 | 0 | — |
case-14 | fail→pass | 15,575 | 2,541 | -84% | 1 | 1 | 0% | 915 | 2,436 | +166% | 0 | 0 | — |
case-15 | fail→pass | 5,190 | 2,680 | -48% | 1 | 1 | 0% | 882 | 2,494 | +183% | 0 | 0 | — |
case-16 | fail→pass | 18,525 | 2,000 | -89% | 1 | 1 | 0% | 1,473 | 2,330 | +58% | 0 | 0 | — |
case-17 | fail→pass | 11,327 | 3,052 | -73% | 1 | 1 | 0% | 1,635 | 2,441 | +49% | 0 | 0 | — |
case-18 | fail→pass | 12,054 | 3,563 | -70% | 1 | 1 | 0% | 1,984 | 2,600 | +31% | 0 | 0 | — |
case-19 | fail→pass | 2,677 | 7,705 | +188% | 1 | 1 | 0% | 367 | 2,469 | +573% | 0 | 0 | — |
case-20 | fail→pass | 7,418 | 4,694 | -37% | 1 | 1 | 0% | 1,131 | 2,834 | +151% | 0 | 0 | — |
case-21 | fail→pass | 9,538 | 4,241 | -56% | 1 | 1 | 0% | 1,599 | 2,714 | +70% | 0 | 0 | — |
case-22 | fail→pass | 9,270 | 2,598 | -72% | 1 | 1 | 0% | 1,405 | 2,439 | +74% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 16 counted toward the lift figure. The other 6 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 16 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.