Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Publish intentional local changes by confirming scope, creating a branch when needed, staging selected files, committing, pushing, and creating a Draft Pull Request through the connected GitHub service.
.claude/skills/devin-axis-github-publish-changes/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 277% | 0% |
| case-05 | ✓→✗ | ▼ Worse | -25% | 0% |
| case-08 | ✓→✗ | ▼ Worse | -43% | 0% |
| case-12 | ✓→✗ | ▼ Worse | 208% | 0% |
| case-14 | ✓→✗ | ▼ Worse | -9% | 0% |
Use this skill only when the user explicitly wants local changes committed, pushed, or opened as a Pull Request.
git status -sb, the complete diff, the current branch, and remotes.git add -A without confirmed scope.agent/<short-description> branch. Otherwise keep the intended feature branch.owner, repo, remote default branch, and pushed head branch.create-pull-request through the GitHub service. Keep draft true unless the user explicitly requests a ready-for-review PR.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,582 | 16,468 | +21% | 1 | 1 | 0% | 1,429 | 796 | -44% | 0 | 0 | — |
case-02 | fail→fail | 22,044 | 15,350 | -30% | 1 | 1 | 0% | 1,349 | 552 | -59% | 0 | 0 | — |
case-03 | fail→fail | 12,153 | 16,431 | +35% | 1 | 1 | 0% | 1,259 | 544 | -57% | 0 | 0 | — |
case-04 | fail→pass | 17,825 | 21,343 | +20% | 1 | 1 | 0% | 870 | 3,279 | +277% | 0 | 0 | — |
case-05 | pass→fail | 9,747 | 16,222 | +66% | 1 | 1 | 0% | 869 | 651 | -25% | 0 | 0 | — |
case-06 | fail→fail | 10,382 | 16,133 | +55% | 1 | 1 | 0% | 889 | 618 | -30% | 0 | 0 | — |
case-07 | fail→fail | 9,225 | 18,038 | +96% | 1 | 1 | 0% | 710 | 633 | -11% | 0 | 0 | — |
case-08 | pass→fail | 12,602 | 16,647 | +32% | 1 | 1 | 0% | 1,363 | 781 | -43% | 0 | 0 | — |
case-09 | fail→fail | 12,840 | 25,355 | +97% | 1 | 1 | 0% | 1,410 | 2,298 | +63% | 0 | 0 | — |
case-10 | fail→fail | 11,633 | 16,357 | +41% | 1 | 1 | 0% | 1,047 | 791 | -24% | 0 | 0 | — |
case-11 | fail→fail | 9,041 | 17,176 | +90% | 1 | 1 | 0% | 482 | 741 | +54% | 0 | 0 | — |
case-12 | pass→fail | 12,178 | 25,171 | +107% | 1 | 1 | 0% | 1,221 | 3,757 | +208% | 0 | 0 | — |
case-13 | fail→fail | 10,894 | 16,448 | +51% | 1 | 1 | 0% | 1,012 | 667 | -34% | 0 | 0 | — |
case-14 | pass→fail | 11,635 | 24,192 | +108% | 1 | 1 | 0% | 1,005 | 919 | -9% | 0 | 0 | — |
case-15 | pass→fail | 11,146 | 16,639 | +49% | 1 | 1 | 0% | 1,053 | 743 | -29% | 0 | 0 | — |
case-16 | pass→fail | 10,032 | 14,477 | +44% | 1 | 1 | 0% | 896 | 635 | -29% | 0 | 0 | — |
case-17 | fail→fail | 12,201 | 26,043 | +113% | 1 | 1 | 0% | 1,168 | 3,738 | +220% | 0 | 0 | — |
case-18 | fail→fail | 10,045 | 16,420 | +63% | 1 | 1 | 0% | 781 | 637 | -18% | 0 | 0 | — |
case-19 | pass→pass | 12,184 | 10,059 | -17% | 1 | 1 | 0% | 1,200 | 1,109 | -8% | 0 | 0 | — |
case-20 | pass→pass | 11,510 | 8,209 | -29% | 1 | 1 | 0% | 1,092 | 804 | -26% | 0 | 0 | — |
case-21 | pass→pass | 9,263 | 7,745 | -16% | 1 | 1 | 0% | 642 | 728 | +13% | 0 | 0 | — |
case-22 | pass→pass | 9,490 | 6,936 | -27% | 1 | 1 | 0% | 816 | 634 | -22% | 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 7 counted toward the lift figure. The other 15 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 -23 percentage points is the difference between those two pass rates over the 7 comparable cases. 6 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.