Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Keep living documents trustworthy over time — the status header (draft/active/superseded) that tells readers what they're holding, the change-log-for-decisions inside the doc, the supersession chain that kills zombie versions, and the review-date heartbeat. Use when asked which version of this doc is current, our wiki is full of stale pages, set up doc lifecycle rules, or people keep following the old process doc. Produces the status-header standard, the in-doc change log, the supersession proto
.claude/skills/mohitagw15856-doc-versioning-discipline/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 227% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 28% | 0% |
| case-04 | ✓→✗ | ▼ Worse | 24% | 0% |
Documents don't announce their own death — the old process doc reads exactly as confidently as the new one, and readers follow whichever they found first. Trust in a doc system is a metadata problem: every living doc carries a status header (what am I holding — draft? active? superseded-by-X?), meaningful changes get logged inside the doc (decisions, not typo-fixes), superseded versions get killed properly (pointer left behind, per the version-chaos-untangler rule), and a review date gives every doc a heartbeat — because "current as of when?" is the question every reader silently asks.
Ask for these if not provided:
Status: active · Owner: [name] · Last reviewed: [date] · Supersedes: [link] / Superseded by: —. The reader learns in two seconds whether to trust, and the absence of the header becomes itself a signal ("no header = treat as unverified").Status: draft — do not follow at top, and drafts live out of the main navigation until active. Half of "the doc was wrong" incidents are "the doc was a draft that escaped."⚠ review overdue) — an honest "this may be stale" beats confident rot, and the quarterly sweep of overdue flags is the whole maintenance system.The four-line block · where it goes · the no-header-means-unverified norm]
Decision-grade entries, newest-first, one line + why · diffs stay in platform history]
The kill steps: pointer, redirects, same-sitting · the never-silent-delete and never-leave-intact rules]
Review cadence by doc class · the 5-minute review · the overdue flag · the quarterly sweep, owned by (role)]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→fail | 14,778 | 13,087 | -11% | 1 | 1 | 0% | 2,848 | 3,659 | +28% | 0 | 0 | — |
case-06 | pass→pass | 12,648 | 11,085 | -12% | 1 | 1 | 0% | 2,705 | 3,384 | +25% | 0 | 0 | — |
case-01 | fail→fail | 23,368 | 19,150 | -18% | 1 | 1 | 0% | 4,433 | 4,330 | -2% | 0 | 0 | — |
case-02 | fail→fail | 20,441 | 14,359 | -30% | 1 | 1 | 0% | 3,969 | 3,745 | -6% | 0 | 0 | — |
case-03 | fail→pass | 46,225 | 13,523 | -71% | 1 | 1 | 0% | 4,185 | 3,694 | -12% | 0 | 0 | — |
case-04 | pass→fail | 15,561 | 14,182 | -9% | 1 | 1 | 0% | 3,269 | 4,066 | +24% | 0 | 0 | — |
case-07 | pass→pass | 11,440 | 13,718 | +20% | 1 | 1 | 0% | 2,081 | 3,675 | +77% | 0 | 0 | — |
case-08 | pass→pass | 12,161 | 15,582 | +28% | 1 | 1 | 0% | 1,982 | 3,815 | +92% | 0 | 0 | — |
case-09 | pass→pass | 15,003 | 14,575 | -3% | 1 | 1 | 0% | 2,531 | 3,483 | +38% | 0 | 0 | — |
case-10 | pass→pass | 15,207 | 14,977 | -2% | 1 | 1 | 0% | 2,638 | 3,644 | +38% | 0 | 0 | — |
case-11 | fail→fail | 16,334 | 12,924 | -21% | 1 | 1 | 0% | 3,097 | 3,638 | +17% | 0 | 0 | — |
case-12 | pass→pass | 13,449 | 13,697 | +2% | 1 | 1 | 0% | 2,485 | 3,673 | +48% | 0 | 0 | — |
case-13 | fail→pass | 17,799 | 12,330 | -31% | 1 | 1 | 0% | 1,019 | 3,329 | +227% | 0 | 0 | — |
case-14 | pass→pass | 13,123 | 10,413 | -21% | 1 | 1 | 0% | 2,431 | 2,979 | +23% | 0 | 0 | — |
case-15 | fail→pass | 14,157 | 13,073 | -8% | 1 | 1 | 0% | 2,507 | 3,291 | +31% | 0 | 0 | — |
case-16 | pass→pass | 13,114 | 14,353 | +9% | 1 | 1 | 0% | 2,236 | 3,473 | +55% | 0 | 0 | — |
case-17 | pass→pass | 12,671 | 9,070 | -28% | 1 | 1 | 0% | 2,079 | 2,645 | +27% | 0 | 0 | — |
case-18 | pass→pass | 13,123 | 11,655 | -11% | 1 | 1 | 0% | 2,040 | 2,930 | +44% | 0 | 0 | — |
case-19 | pass→pass | 9,964 | 9,422 | -5% | 1 | 1 | 0% | 1,769 | 2,816 | +59% | 0 | 0 | — |
case-20 | pass→pass | 11,757 | 10,471 | -11% | 1 | 1 | 0% | 1,878 | 2,893 | +54% | 0 | 0 | — |
case-21 | pass→pass | 12,945 | 16,241 | +25% | 1 | 1 | 0% | 2,139 | 3,822 | +79% | 0 | 0 | — |
case-22 | pass→pass | 16,194 | 19,920 | +23% | 1 | 1 | 0% | 2,638 | 4,546 | +72% | 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 21 counted toward the lift figure. The other 1 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 +5 percentage points is the difference between those two pass rates over the 21 comparable cases. 2 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.