Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Score an existing reader-facing deliverable and emit a deterministic PASS/FAIL scorecard plus bounded repair plan; use after synthesis, not to rewrite content.
.claude/skills/willoscar-deliverable-selfloop/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 237% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -43% | 0% |
Runs the final quality gate for a reader-facing deliverable and always writes a PASS/FAIL report.
Primary input depends on the active pipeline contract:
output/DELIVERABLE_SELFLOOP_TODO.mdresearch-brief: output/BRIEF_SCORECARD.md and output/BRIEF_SCORECARD.jsonpaper-review: output/REVIEW_SCORECARD.md and output/REVIEW_SCORECARD.jsonidea-brainstorm: output/IDEA_SCORECARD.md and output/IDEA_SCORECARD.jsonevidence-review: output/EVIDENCE_SCORECARD.md and output/EVIDENCE_SCORECARD.jsonThe gate should dispatch by pipeline contract first:
quality_contract.deliverable_kindOnly fall back to legacy profile-name checks when contract metadata is missing.
scripts/run.py should:
It should not mutate the deliverable itself.
- Status: PASS or - Status: FAILresearch-brief additionally requires valid core-set pointers and the configured score thresholdpaper-review additionally requires the configured score threshold and all critical rubric dimensionsidea-brainstorm additionally requires traceable anchors, actionable lead directions, and the configured score thresholdevidence-review additionally requires clause-linked screening, complete extraction rows, synthesis pointers, and the configured score threshold| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-21 | pass→pass | 7,536 | 2,430 | -68% | 1 | 1 | 0% | 1,126 | 798 | -29% | 0 | 0 | — |
case-01 | fail→fail | 4,712 | 5,125 | +9% | 1 | 1 | 0% | 596 | 679 | +14% | 0 | 0 | — |
case-02 | fail→fail | 3,778 | 5,329 | +41% | 1 | 1 | 0% | 540 | 634 | +17% | 0 | 0 | — |
case-03 | fail→fail | 14,598 | 5,082 | -65% | 1 | 1 | 0% | 2,458 | 689 | -72% | 0 | 0 | — |
case-04 | pass→pass | 3,883 | 1,949 | -50% | 1 | 1 | 0% | 593 | 748 | +26% | 0 | 0 | — |
case-14 | fail→pass | 3,185 | 6,482 | +104% | 1 | 1 | 0% | 421 | 1,419 | +237% | 0 | 0 | — |
case-05 | fail→pass | 8,216 | 1,772 | -78% | 1 | 1 | 0% | 1,132 | 659 | -42% | 0 | 0 | — |
case-06 | pass→pass | 10,750 | 3,582 | -67% | 1 | 1 | 0% | 1,614 | 1,019 | -37% | 0 | 0 | — |
case-07 | fail→pass | 9,900 | 2,930 | -70% | 1 | 1 | 0% | 1,629 | 966 | -41% | 0 | 0 | — |
case-08 | fail→pass | 10,521 | 2,729 | -74% | 1 | 1 | 0% | 1,482 | 929 | -37% | 0 | 0 | — |
case-15 | fail→pass | 13,171 | 4,166 | -68% | 1 | 1 | 0% | 2,015 | 1,144 | -43% | 0 | 0 | — |
case-09 | fail→pass | 6,987 | 2,420 | -65% | 1 | 1 | 0% | 923 | 819 | -11% | 0 | 0 | — |
case-10 | pass→pass | 11,607 | 2,119 | -82% | 1 | 1 | 0% | 1,575 | 744 | -53% | 0 | 0 | — |
case-11 | fail→pass | 14,322 | 1,825 | -87% | 1 | 1 | 0% | 1,108 | 675 | -39% | 0 | 0 | — |
case-12 | fail→fail | 6,941 | 14,556 | +110% | 1 | 1 | 0% | 1,108 | 2,720 | +145% | 0 | 0 | — |
case-13 | fail→fail | 15,336 | 13,791 | -10% | 1 | 1 | 0% | 2,222 | 2,524 | +14% | 0 | 0 | — |
case-16 | fail→pass | 16,878 | 2,841 | -83% | 1 | 1 | 0% | 873 | 926 | +6% | 0 | 0 | — |
case-17 | fail→pass | 8,482 | 2,869 | -66% | 1 | 1 | 0% | 1,325 | 958 | -28% | 0 | 0 | — |
case-18 | fail→pass | 9,699 | 2,326 | -76% | 1 | 1 | 0% | 1,613 | 823 | -49% | 0 | 0 | — |
case-19 | pass→pass | 10,839 | 3,431 | -68% | 1 | 1 | 0% | 1,605 | 1,016 | -37% | 0 | 0 | — |
case-20 | fail→pass | 10,109 | 1,687 | -83% | 1 | 1 | 0% | 1,367 | 671 | -51% | 0 | 0 | — |
case-22 | fail→pass | 10,574 | 2,860 | -73% | 1 | 1 | 0% | 1,528 | 953 | -38% | 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 18 counted toward the lift figure. The other 4 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 +55 percentage points is the difference between those two pass rates over the 18 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.