Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Scan merged PRs and commits since a given reference, extract titles, labels, types, and signals. Produces structured input for release notes drafting.
.claude/skills/cobusgreyling-changelog-scan/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 8 |
| gemini-3.1-pro-preview | 100% | 1 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -38% | 0% |
markdown### PR #1234 — feat(auth): add magic link login (merged 2026-06-08) - Type: feature - Labels: enhancement - Breaking: no - Security: no - Linked: #1220 - Summary (one sentence from PR or commit): Users can now log in via emailed magic links. - Files touched (high level): auth/, emails/
Rules for what to include:
! conventional commit.markdown## Scan Summary - Total items: N - Features: N - Fixes: N - Breaking: N (list them) - Security: N (list them) - Recommended next action for loop: draft-release-notes | human review needed first | too many items — split window
Be precise and cite sources (PR numbers / shas). Do not invent details.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-21 | pass→pass | 5,480 | 5,781 | +5% | 1 | 1 | 0% | 997 | 1,411 | +42% | 0 | 0 | — |
case-01 | fail→fail | 9,601 | 9,254 | -4% | 1 | 1 | 0% | 2,304 | 2,659 | +15% | 0 | 0 | — |
case-02 | fail→pass | 15,580 | 8,183 | -47% | 1 | 1 | 0% | 3,049 | 2,418 | -21% | 0 | 0 | — |
case-03 | fail→pass | 12,132 | 8,068 | -33% | 1 | 1 | 0% | 2,446 | 2,138 | -13% | 0 | 0 | — |
case-04 | fail→pass | 9,844 | 4,814 | -51% | 1 | 1 | 0% | 1,766 | 1,232 | -30% | 0 | 0 | — |
case-05 | fail→pass | 11,989 | 4,417 | -63% | 1 | 1 | 0% | 2,170 | 1,169 | -46% | 0 | 0 | — |
case-06 | fail→fail | 3,736 | 3,539 | -5% | 1 | 1 | 0% | 609 | 1,069 | +76% | 0 | 0 | — |
case-07 | fail→pass | 11,988 | 4,368 | -64% | 1 | 1 | 0% | 2,013 | 1,245 | -38% | 0 | 0 | — |
case-08 | fail→pass | 11,933 | 6,466 | -46% | 1 | 1 | 0% | 1,871 | 962 | -49% | 0 | 0 | — |
case-09 | fail→pass | 6,883 | 2,973 | -57% | 1 | 1 | 0% | 1,237 | 950 | -23% | 0 | 0 | — |
case-10 | pass→pass | 12,728 | 4,120 | -68% | 1 | 1 | 0% | 2,186 | 1,170 | -46% | 0 | 0 | — |
case-11 | fail→pass | 3,918 | 4,083 | +4% | 1 | 1 | 0% | 724 | 1,182 | +63% | 0 | 0 | — |
case-22 | pass→pass | 3,757 | 1,975 | -47% | 1 | 1 | 0% | 722 | 734 | +2% | 0 | 0 | — |
case-12 | fail→pass | 13,597 | 6,238 | -54% | 1 | 1 | 0% | 2,496 | 1,567 | -37% | 0 | 0 | — |
case-13 | fail→pass | 5,207 | 4,786 | -8% | 1 | 1 | 0% | 1,028 | 1,459 | +42% | 0 | 0 | — |
case-14 | fail→pass | 6,885 | 6,290 | -9% | 1 | 1 | 0% | 1,117 | 1,682 | +51% | 0 | 0 | — |
case-15 | fail→pass | 12,840 | 3,420 | -73% | 1 | 1 | 0% | 2,150 | 934 | -57% | 0 | 0 | — |
case-16 | fail→pass | 5,151 | 2,235 | -57% | 1 | 1 | 0% | 965 | 783 | -19% | 0 | 0 | — |
case-17 | fail→pass | 8,052 | 2,672 | -67% | 1 | 1 | 0% | 1,458 | 827 | -43% | 0 | 0 | — |
case-18 | fail→pass | 6,776 | 2,779 | -59% | 1 | 1 | 0% | 1,142 | 835 | -27% | 0 | 0 | — |
case-19 | fail→pass | 4,616 | 13,131 | +184% | 1 | 1 | 0% | 939 | 1,551 | +65% | 0 | 0 | — |
case-20 | pass→pass | 14,146 | 9,779 | -31% | 1 | 1 | 0% | 2,274 | 2,094 | -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. The headline lift of +73 percentage points is the difference between those two pass rates over the 22 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.