Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Update vault to the latest version of obsidian-claude-pkm. Creates backup, shows diffs, preserves your content. Use when a new version is available.
.claude/skills/ballred-upgrade/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -18% | 0% |
Updates your vault's system files to the latest version while preserving all your personal content.
/upgrade # Interactive upgrade with backup and diff review
/upgrade check # Preview what's changed without making modifications.claude/skills/*/SKILL.md — Skill definitions.claude/agents/*.md — Agent configurations.claude/hooks/*.sh — Automation scripts.claude/rules/*.md — Convention rules.claude/output-styles/*.md — Output style definitions.claude/settings.json — Permissions and hook configCLAUDE.md — Root context file (merged carefully)Templates/*.md — Note templatesDaily Notes/** — Your daily journal entriesGoals/** — Your goal files and reviewsProjects/** — Your project folders and CLAUDE.md filesArchives/** — Your archived contentInbox/** — Your captured itemsCLAUDE.local.md — Your personal overridesvault-config.json — Your preferencesRead the current version from CLAUDE.md (look for "System Version:" line).
Compare with the upstream repo. If using git:
bashgit fetch origin git log HEAD..origin/main --oneline
If not a git repo or no remote, inform user they need to download the latest vault-template/ manually.
Before any changes, create a timestamped backup:
bashBACKUP_DIR=".backup/upgrade-$(date +%Y%m%d-%H%M%S)" mkdir -p "$BACKUP_DIR" cp -r .claude "$BACKUP_DIR/" cp CLAUDE.md "$BACKUP_DIR/" cp -r Templates "$BACKUP_DIR/" 2>/dev/null
Tell the user: "Backup created at $BACKUP_DIR"
For each system file that differs from upstream:
For each confirmed file:
CLAUDE.md: merge carefully — preserve user customizations (Personal Mission, Current Focus sections) while updating system sections (Skills table, version number)settings.json: merge — add new permissions/hooks, preserve existing customizationschmod +x .claude/hooks/*.sh/onboard to reload context/upgrade check)When invoked with "check":
/upgrade to applymarkdown## Upgrade Check **Current version:** 3.0 (The Cascade) **Available version:** 3.1 ### Files to Update | File | Change Summary | |------|---------------| | `.claude/skills/daily/SKILL.md` | Added cascade context surfacing | | `.claude/agents/goal-aligner.md` | Added memory: project | | `.claude/hooks/session-init.sh` | Added priority surfacing | ### New Files - `.claude/skills/review/SKILL.md` — Smart review router - `.claude/hooks/skill-discovery.sh` — Auto-list skills ### No Changes Needed - `.claude/rules/` — Already up to date Run `/upgrade` to apply these updates (backup will be created first).
CLAUDE.local.md personal overrides preservedWorks with:
/onboard — Reload context after upgrade| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,446 | 5,202 | -4% | 1 | 1 | 0% | 956 | 1,399 | +46% | 0 | 0 | — |
case-02 | fail→fail | 9,473 | 2,822 | -70% | 1 | 1 | 0% | 1,739 | 1,421 | -18% | 0 | 0 | — |
case-03 | fail→fail | 5,309 | 5,644 | +6% | 1 | 1 | 0% | 353 | 1,513 | +329% | 0 | 0 | — |
case-04 | fail→pass | 12,093 | 2,689 | -78% | 1 | 1 | 0% | 2,017 | 1,581 | -22% | 0 | 0 | — |
case-05 | pass→pass | 9,678 | 5,212 | -46% | 1 | 1 | 0% | 1,731 | 2,110 | +22% | 0 | 0 | — |
case-06 | pass→pass | 12,612 | 5,145 | -59% | 1 | 1 | 0% | 2,034 | 1,972 | -3% | 0 | 0 | — |
case-07 | fail→pass | 8,364 | 3,115 | -63% | 1 | 1 | 0% | 1,434 | 1,621 | +13% | 0 | 0 | — |
case-08 | pass→pass | 2,276 | 1,084 | -52% | 1 | 1 | 0% | 411 | 1,239 | +201% | 0 | 0 | — |
case-09 | pass→pass | 6,125 | 3,222 | -47% | 1 | 1 | 0% | 1,194 | 1,697 | +42% | 0 | 0 | — |
case-10 | fail→pass | 12,685 | 1,804 | -86% | 1 | 1 | 0% | 2,065 | 1,415 | -31% | 0 | 0 | — |
case-11 | pass→pass | 8,708 | 2,175 | -75% | 1 | 1 | 0% | 1,389 | 1,456 | +5% | 0 | 0 | — |
case-12 | fail→pass | 8,589 | 1,171 | -86% | 1 | 1 | 0% | 1,377 | 1,295 | -6% | 0 | 0 | — |
case-13 | fail→pass | 9,303 | 1,513 | -84% | 1 | 1 | 0% | 1,628 | 1,341 | -18% | 0 | 0 | — |
case-14 | pass→pass | 8,352 | 5,477 | -34% | 1 | 1 | 0% | 1,499 | 2,099 | +40% | 0 | 0 | — |
case-15 | pass→pass | 8,631 | 2,094 | -76% | 1 | 1 | 0% | 1,325 | 1,431 | +8% | 0 | 0 | — |
case-16 | pass→pass | 8,903 | 3,819 | -57% | 1 | 1 | 0% | 1,474 | 1,768 | +20% | 0 | 0 | — |
case-17 | pass→pass | 10,067 | 1,751 | -83% | 1 | 1 | 0% | 1,689 | 1,410 | -17% | 0 | 0 | — |
case-18 | fail→pass | 10,913 | 2,359 | -78% | 1 | 1 | 0% | 1,783 | 1,513 | -15% | 0 | 0 | — |
case-19 | fail→pass | 12,146 | 3,409 | -72% | 1 | 1 | 0% | 2,021 | 1,673 | -17% | 0 | 0 | — |
case-20 | pass→pass | 9,318 | 4,284 | -54% | 1 | 1 | 0% | 1,556 | 1,841 | +18% | 0 | 0 | — |
case-21 | pass→pass | 10,093 | 4,022 | -60% | 1 | 1 | 0% | 1,789 | 1,846 | +3% | 0 | 0 | — |
case-22 | pass→pass | 6,906 | 3,143 | -54% | 1 | 1 | 0% | 1,225 | 1,666 | +36% | 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 20 counted toward the lift figure. The other 2 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 +32 percentage points is the difference between those two pass rates over the 20 comparable cases.
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.