Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Untangle a document that exists in six copies across email, drives, and desktops — establish the canonical version defensibly, merge the divergent edits, and install the single-source rule that prevents the rematch. Use when asked which version is the real one, merge these document copies, we've been editing different files, or stop the version chaos on this doc. Produces the version census with the canonical verdict, the divergence merge plan, the announce-and-redirect step, and the single-sour
.claude/skills/mohitagw15856-version-chaos-untangler/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 76% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 8% | 0% |
Six copies of the contract: email attachments, two drives, someone's desktop, one renamed -final-JT-edits. Two have diverged — different edits on a shared ancestor — and the team is now editing in parallel universes. Untangling is forensic then structural: census every copy (where, modified-when, by-whom), establish the canonical defensibly (newest is not automatically rightest), merge what diverged, then kill the conditions — one home, links-not-attachments, and platform version history doing what filename suffixes failed to do.
Ask for these if not provided:
| Copy | Location | Modified | Editor | Signals | |---|---|---|---|---|
Canonical: which and the lineage why · divergence found: yes/no · if yes: the merge is the canonical]
Section-level reconciliation · judgment calls × confirming author · the audit-trail note for legal docs]
The one home (versioned platform) · pointers placed over old copies · the announcement line · links-only norm + the external copy-holder rule if applicable]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 51,932 | 22,388 | -57% | 1 | 1 | 0% | 6,416 | 4,219 | -34% | 0 | 0 | — |
case-02 | fail→pass | 65,153 | 27,362 | -58% | 1 | 1 | 0% | 7,661 | 4,528 | -41% | 0 | 0 | — |
case-03 | fail→fail | 62,377 | 26,627 | -57% | 1 | 1 | 0% | 7,500 | 4,650 | -38% | 0 | 0 | — |
case-04 | pass→fail | 12,106 | 18,012 | +49% | 1 | 1 | 0% | 1,808 | 3,335 | +84% | 0 | 0 | — |
case-05 | pass→pass | 27,306 | 23,214 | -15% | 1 | 1 | 0% | 2,882 | 3,722 | +29% | 0 | 0 | — |
case-06 | pass→pass | 19,124 | 19,215 | +0% | 1 | 1 | 0% | 2,384 | 3,488 | +46% | 0 | 0 | — |
case-07 | fail→pass | 18,566 | 32,068 | +73% | 1 | 1 | 0% | 2,295 | 3,344 | +46% | 0 | 0 | — |
case-08 | fail→pass | 17,974 | 27,779 | +55% | 1 | 1 | 0% | 2,383 | 4,195 | +76% | 0 | 0 | — |
case-09 | fail→pass | 21,827 | 24,653 | +13% | 1 | 1 | 0% | 2,254 | 3,691 | +64% | 0 | 0 | — |
case-10 | fail→pass | 30,704 | 20,569 | -33% | 1 | 1 | 0% | 3,369 | 3,646 | +8% | 0 | 0 | — |
case-11 | fail→pass | 18,186 | 16,557 | -9% | 1 | 1 | 0% | 1,459 | 2,590 | +78% | 0 | 0 | — |
case-12 | fail→pass | 26,274 | 19,473 | -26% | 1 | 1 | 0% | 2,756 | 2,890 | +5% | 0 | 0 | — |
case-13 | fail→pass | 19,488 | 17,695 | -9% | 1 | 1 | 0% | 2,240 | 3,205 | +43% | 0 | 0 | — |
case-14 | pass→pass | 10,567 | 12,696 | +20% | 1 | 1 | 0% | 1,618 | 2,158 | +33% | 0 | 0 | — |
case-15 | fail→pass | 20,410 | 24,114 | +18% | 1 | 1 | 0% | 2,518 | 4,027 | +60% | 0 | 0 | — |
case-16 | fail→pass | 24,076 | 26,149 | +9% | 1 | 1 | 0% | 3,089 | 4,508 | +46% | 0 | 0 | — |
case-17 | fail→pass | 16,113 | 16,149 | +0% | 1 | 1 | 0% | 1,903 | 3,604 | +89% | 0 | 0 | — |
case-18 | fail→pass | 28,977 | 23,671 | -18% | 1 | 1 | 0% | 3,930 | 3,814 | -3% | 0 | 0 | — |
case-19 | fail→pass | 14,517 | 22,355 | +54% | 1 | 1 | 0% | 1,379 | 3,388 | +146% | 0 | 0 | — |
case-20 | fail→pass | 15,470 | 14,330 | -7% | 1 | 1 | 0% | 1,485 | 2,326 | +57% | 0 | 0 | — |
case-21 | fail→fail | 22,052 | 20,129 | -9% | 1 | 1 | 0% | 2,572 | 4,418 | +72% | 0 | 0 | — |
case-22 | fail→pass | 13,307 | 27,285 | +105% | 1 | 1 | 0% | 1,255 | 4,307 | +243% | 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. The headline lift of +64 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.