Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when preparing iPolloWork local changes for GitHub: stash or preserve local edits, sync the current personal branch with remote main, re-apply local work, diagnose conflicts, commit in English, push to the current remote branch, and open a PR to main without creating extra branches.
.claude/skills/devin-axis-github-sync-pr-flow/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 899% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 81% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 103% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 41% | 0% |
Use this workflow for iPolloWork developer submissions, personal-branch syncs, and PR requests. Keep the developer on their current personal branch unless the user explicitly asks for another branch.
main, committing, pushing, creating a PR, or calling GitHub APIs, get explicit user approval for that phase.main.origin/<current-branch>.origin/main.Choose the smallest workflow that matches the user's words.
origin/main into the current personal branch, then restore local work. Do not commit or push unless requested.main.main.Before any sync, commit, push, or PR action:
git status --short --branch, git branch -vv, and git remote -v.main, stop and ask for the intended personal branch..env files, local logs, build artifacts, or generated cache files unless explicitly intended.Use this when the user asks to pull, sync, update from main, or prepare their personal branch before submission.
main.origin/main.git stash push -u -m "<clear English message>".git stash fails, stop and report the error.git fetch origin --prune.git pull for this workflow because it hides whether the incoming changes came from main, the current branch upstream, or both.git merge origin/main unless the repository explicitly documents a rebase workflow.git stash pop for the saved stash.git status --short --branch.git diff --check.Use this when the user says "submit my code", "commit my code", or "push my code" without asking to merge to main.
git add -A when unrelated changes exist.git diff --check.git push -u origin <current-branch>.Use this when the user asks to open a PR or merge their code toward main.
main.Use this only when the user explicitly asks to merge into main.
main and the head branch is the user's personal branch.main from local unless the repository owner explicitly requires it.When conflicts occur, report:
git diff --ours or upstream context.git diff --theirs, stash context, or working-tree context.Do not resolve conflicts without user confirmation unless the resolution is mechanically obvious and explicitly allowed.
Use this only when the user explicitly asks for the full sequence from local changes to PR.
origin/main into the current branch.origin/<current-branch>.origin/main.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 16,696 | 16,722 | +0% | 1 | 1 | 0% | 358 | 2,258 | +531% | 0 | 0 | — |
case-02 | pass→pass | 14,028 | 9,003 | -36% | 1 | 1 | 0% | 260 | 2,319 | +792% | 0 | 0 | — |
case-03 | fail→pass | 16,790 | 15,081 | -10% | 1 | 1 | 0% | 349 | 3,486 | +899% | 0 | 0 | — |
case-04 | fail→pass | 15,047 | 11,936 | -21% | 1 | 1 | 0% | 1,640 | 2,961 | +81% | 0 | 0 | — |
case-05 | pass→pass | 13,091 | 9,020 | -31% | 1 | 1 | 0% | 1,261 | 2,527 | +100% | 0 | 0 | — |
case-06 | fail→fail | 10,230 | 16,955 | +66% | 1 | 1 | 0% | 1,018 | 4,008 | +294% | 0 | 0 | — |
case-07 | fail→pass | 30,368 | 9,821 | -68% | 1 | 1 | 0% | 1,515 | 2,609 | +72% | 0 | 0 | — |
case-08 | fail→pass | 12,216 | 3,612 | -70% | 1 | 1 | 0% | 1,163 | 2,363 | +103% | 0 | 0 | — |
case-09 | pass→pass | 13,224 | 10,151 | -23% | 1 | 1 | 0% | 1,401 | 2,647 | +89% | 0 | 0 | — |
case-10 | pass→pass | 10,807 | 51,371 | +375% | 1 | 1 | 0% | 935 | 2,325 | +149% | 0 | 0 | — |
case-11 | pass→pass | 5,075 | 9,342 | +84% | 1 | 1 | 0% | 962 | 2,561 | +166% | 0 | 0 | — |
case-12 | fail→fail | 5,643 | 9,722 | +72% | 1 | 1 | 0% | 710 | 2,484 | +250% | 0 | 0 | — |
case-21 | fail→pass | 10,611 | 4,447 | -58% | 1 | 1 | 0% | 1,797 | 2,527 | +41% | 0 | 0 | — |
case-13 | pass→pass | 11,171 | 9,174 | -18% | 1 | 1 | 0% | 1,106 | 2,548 | +130% | 0 | 0 | — |
case-14 | pass→pass | 14,870 | 10,296 | -31% | 1 | 1 | 0% | 1,752 | 2,810 | +60% | 0 | 0 | — |
case-15 | fail→pass | 8,841 | 10,412 | +18% | 1 | 1 | 0% | 1,551 | 2,825 | +82% | 0 | 0 | — |
case-16 | pass→pass | 16,940 | 10,120 | -40% | 1 | 1 | 0% | 2,179 | 2,631 | +21% | 0 | 0 | — |
case-22 | pass→pass | 10,471 | 10,252 | -2% | 1 | 1 | 0% | 1,812 | 3,478 | +92% | 0 | 0 | — |
case-17 | fail→pass | 8,983 | 8,308 | -8% | 1 | 1 | 0% | 1,477 | 2,335 | +58% | 0 | 0 | — |
case-18 | pass→pass | 9,480 | 5,673 | -40% | 1 | 1 | 0% | 1,605 | 2,774 | +73% | 0 | 0 | — |
case-19 | fail→fail | 7,578 | 6,131 | -19% | 1 | 1 | 0% | 1,252 | 2,772 | +121% | 0 | 0 | — |
case-20 | fail→pass | 11,300 | 2,635 | -77% | 1 | 1 | 0% | 1,924 | 2,251 | +17% | 0 | 0 | — |
case-23 | pass→pass | 13,194 | 9,765 | -26% | 1 | 1 | 0% | 2,554 | 3,633 | +42% | 0 | 0 | — |
case-24 | fail→fail | 9,719 | 10,190 | +5% | 1 | 1 | 0% | 1,802 | 3,872 | +115% | 0 | 0 | — |
case-25 | fail→pass | 8,835 | 4,199 | -52% | 1 | 1 | 0% | 1,532 | 2,621 | +71% | 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. 25 cases were attempted, and 23 counted toward the lift figure. The other 2 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 +36 percentage points is the difference between those two pass rates over the 23 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.