Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Reviews a QIP draft or Slack Canvas proposal for approval readiness, missing alternatives, unclear consequences, architecture principle conflicts, and consultation coverage. Produces blockers, clarifying questions, suggested edits, and optional Slack-ready review comments. Use when reviewing a QIP or invoking /qv-qip-review.
.claude/skills/tetherto-qv-qip-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-13 | ✓→✗ | ▼ Worse | -6% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -9% | 0% |
Review a QIP for approval readiness without substituting for human approvers.
Use when:
/qv-qip-reviewDo NOT use for:
qv-qip-create)qv-qip-triage)Accept any of:
docs/architecture/PRINCIPLES.md before checking for principle conflictsdocs/architecture/ARCHITECTURE.md when the proposal touches runtime, package, plugin, registry, or deployment boundariesTemplate completeness
Architecture
Consultation coverage
Advice coverage is not a voting scheme.
Lead with findings ordered by approval risk:
markdown## Blockers - ... ## Clarifying questions - ... ## Suggested edits - ... ## Approval readiness Ready | Ready with minor edits | Not ready ## Slack comment <optional concise paste-ready comment if requested>
If there are no blockers, say so explicitly.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,572 | 4,319 | -68% | 1 | 1 | 0% | 1,993 | 1,280 | -36% | 0 | 0 | — |
case-02 | fail→fail | 22,297 | 13,318 | -40% | 1 | 1 | 0% | 3,504 | 2,647 | -24% | 0 | 0 | — |
case-03 | pass→pass | 9,764 | 4,518 | -54% | 1 | 1 | 0% | 1,430 | 1,308 | -9% | 0 | 0 | — |
case-04 | fail→fail | 8,673 | 4,422 | -49% | 1 | 1 | 0% | 1,293 | 1,368 | +6% | 0 | 0 | — |
case-05 | fail→fail | 2,312 | 6,227 | +169% | 1 | 1 | 0% | 352 | 1,674 | +376% | 0 | 0 | — |
case-06 | pass→pass | 8,098 | 3,337 | -59% | 1 | 1 | 0% | 1,112 | 1,079 | -3% | 0 | 0 | — |
case-07 | pass→pass | 13,050 | 4,937 | -62% | 1 | 1 | 0% | 1,874 | 1,427 | -24% | 0 | 0 | — |
case-08 | pass→pass | 11,694 | 7,259 | -38% | 1 | 1 | 0% | 1,881 | 1,761 | -6% | 0 | 0 | — |
case-09 | pass→pass | 11,024 | 5,138 | -53% | 1 | 1 | 0% | 1,838 | 1,410 | -23% | 0 | 0 | — |
case-10 | pass→pass | 14,246 | 4,544 | -68% | 1 | 1 | 0% | 2,127 | 1,438 | -32% | 0 | 0 | — |
case-11 | fail→fail | 2,204 | 4,781 | +117% | 1 | 1 | 0% | 343 | 1,410 | +311% | 0 | 0 | — |
case-12 | pass→pass | 7,120 | 5,256 | -26% | 1 | 1 | 0% | 1,156 | 1,365 | +18% | 0 | 0 | — |
case-13 | pass→fail | 8,117 | 3,547 | -56% | 1 | 1 | 0% | 1,164 | 1,091 | -6% | 0 | 0 | — |
case-14 | fail→pass | 15,461 | 9,608 | -38% | 1 | 1 | 0% | 2,422 | 2,160 | -11% | 0 | 0 | — |
case-15 | fail→pass | 16,111 | 1,739 | -89% | 1 | 1 | 0% | 982 | 902 | -8% | 0 | 0 | — |
case-16 | pass→pass | 11,602 | 7,460 | -36% | 1 | 1 | 0% | 1,786 | 1,543 | -14% | 0 | 0 | — |
case-17 | pass→pass | 8,977 | 2,914 | -68% | 1 | 1 | 0% | 1,434 | 1,059 | -26% | 0 | 0 | — |
case-18 | pass→pass | 9,799 | 5,464 | -44% | 1 | 1 | 0% | 1,488 | 1,459 | -2% | 0 | 0 | — |
case-19 | pass→pass | 5,627 | 2,168 | -61% | 1 | 1 | 0% | 932 | 913 | -2% | 0 | 0 | — |
case-20 | fail→pass | 8,112 | 2,755 | -66% | 1 | 1 | 0% | 1,232 | 1,019 | -17% | 0 | 0 | — |
case-21 | pass→pass | 8,076 | 1,613 | -80% | 1 | 1 | 0% | 1,238 | 850 | -31% | 0 | 0 | — |
case-22 | pass→pass | 6,881 | 1,788 | -74% | 1 | 1 | 0% | 965 | 883 | -8% | 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 0 percentage points is the difference between those two pass rates over the 21 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.