Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Creates a pull request following Storybook conventions. Use when creating PRs, opening pull requests, or submitting changes for review.
.claude/skills/storybookjs-pr/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 118% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 116% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 183% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 117% | 0% |
Creates a PR following Storybook conventions.
[Area]: [Description]
CSFFactories: Fix type exportNextjs-Vite: Add supportCLI: Fix automigrate issueAdd these labels to the PR:
Category (required, pick one):
bug - fixes incorrect behaviormaintenance - user-facing maintenancedependencies - upgrading/downgrading depsbuild - internal build/test updates (no changelog)cleanup - minor cleanup (no changelog)documentation - docs only (no changelog)feature request - new featureBREAKING CHANGE - breaks compatibilityother - doesn't fit aboveCI (required, pick one):
ci:normal - standard sandbox set; default for most code changesci:merged - merged sandbox setci:daily - daily sandbox set; use this when changes affect prerelease sandboxes or sandboxes pinned to a framework or React version other than latestci:docs - documentation-only changes (use with documentation category)Canary (optional):
ci:canary - publish pkg.pr.new canary packages for an in-repo PR; later pushes republish while the label remains. Does nothing on fork PRs. To canary a fork PR, a maintainer runs publish-canary.yml with the pr input; do not ask the fork author to publish. Do not add this label unless the user asks for a canary.QA (required, pick one):
Tells the release team whether manual QA is needed before the next minor release.
qa:needed — a human must manually verify this PR at release timeqa:skip — no per-PR manual QA needed at release timeHeuristics:
qa:neededqa:skipqa:neededqa:neededqa:neededqa:skipRead .github/PULL_REQUEST_TEMPLATE.md from the repository root.
Copy that template EXACTLY, including all HTML comments (<!-- ... -->). Fill in the relevant sections based on the changes, but keep all comments intact.
The Manual testing section is mandatory — never leave it empty. Write steps for a separate maintainer, not a log of how you tested.
Each step should be:
Verify your own steps first — run through them locally before opening the PR.
When useful, link to published Chromatic Storybooks (CI must finish first; links won't work immediately after opening the PR):
https://<branch>--635781f3500dd2c49e189caf.chromatic.com/?path=/story/<story_id>https://<branch>--630511d655df72125520f051.chromatic.com/?path=/story/<story_id>Replace <branch> with Chromatic's normalized slug (special chars → dashes, e.g. feature/foo → feature-foo) and <story_id> with the story path (e.g. example-button--primary).
Always create PRs in draft mode:
bashgh pr create --draft --title "<Area>: <Description>" --body "<FILLED_TEMPLATE>" --label "<category>,<ci>,<qa>"
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,636 | 6,499 | -44% | 1 | 1 | 0% | 1,732 | 1,298 | -25% | 0 | 0 | — |
case-02 | fail→fail | 13,300 | 7,914 | -40% | 1 | 1 | 0% | 1,783 | 1,559 | -13% | 0 | 0 | — |
case-03 | fail→fail | 10,785 | 13,277 | +23% | 1 | 1 | 0% | 1,512 | 1,319 | -13% | 0 | 0 | — |
case-04 | fail→fail | 4,043 | 6,671 | +65% | 1 | 1 | 0% | 479 | 1,220 | +155% | 0 | 0 | — |
case-05 | pass→pass | 7,324 | 4,937 | -33% | 1 | 1 | 0% | 1,039 | 1,699 | +64% | 0 | 0 | — |
case-06 | pass→pass | 5,251 | 3,481 | -34% | 1 | 1 | 0% | 696 | 1,384 | +99% | 0 | 0 | — |
case-07 | fail→pass | 8,133 | 17,102 | +110% | 1 | 1 | 0% | 1,251 | 2,726 | +118% | 0 | 0 | — |
case-08 | fail→pass | 7,157 | 10,013 | +40% | 1 | 1 | 0% | 1,143 | 2,465 | +116% | 0 | 0 | — |
case-09 | fail→pass | 10,318 | 9,858 | -4% | 1 | 1 | 0% | 1,716 | 2,715 | +58% | 0 | 0 | — |
case-10 | fail→pass | 9,483 | 27,675 | +192% | 1 | 1 | 0% | 1,450 | 4,106 | +183% | 0 | 0 | — |
case-11 | fail→pass | 11,694 | 26,472 | +126% | 1 | 1 | 0% | 1,543 | 3,352 | +117% | 0 | 0 | — |
case-12 | fail→pass | 10,759 | 11,497 | +7% | 1 | 1 | 0% | 1,532 | 2,997 | +96% | 0 | 0 | — |
case-13 | pass→pass | 8,869 | 10,937 | +23% | 1 | 1 | 0% | 1,362 | 2,601 | +91% | 0 | 0 | — |
case-14 | fail→pass | 18,530 | 14,715 | -21% | 1 | 1 | 0% | 1,404 | 3,090 | +120% | 0 | 0 | — |
case-15 | fail→pass | 26,144 | 6,085 | -77% | 1 | 1 | 0% | 872 | 1,939 | +122% | 0 | 0 | — |
case-16 | fail→pass | 14,103 | 3,778 | -73% | 1 | 1 | 0% | 1,879 | 1,474 | -22% | 0 | 0 | — |
case-17 | fail→pass | 11,993 | 9,331 | -22% | 1 | 1 | 0% | 1,788 | 2,657 | +49% | 0 | 0 | — |
case-18 | fail→pass | 8,016 | 17,174 | +114% | 1 | 1 | 0% | 1,263 | 3,444 | +173% | 0 | 0 | — |
case-19 | fail→pass | 13,650 | 18,087 | +33% | 1 | 1 | 0% | 2,269 | 3,703 | +63% | 0 | 0 | — |
case-20 | fail→pass | 9,015 | 12,039 | +34% | 1 | 1 | 0% | 1,361 | 3,057 | +125% | 0 | 0 | — |
case-21 | fail→pass | 13,555 | 6,223 | -54% | 1 | 1 | 0% | 2,141 | 1,800 | -16% | 0 | 0 | — |
case-22 | fail→pass | 11,319 | 7,447 | -34% | 1 | 1 | 0% | 1,252 | 2,241 | +79% | 0 | 0 | — |
case-23 | fail→pass | 18,120 | 3,408 | -81% | 1 | 1 | 0% | 1,481 | 1,522 | +3% | 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. 23 cases were attempted, and 18 counted toward the lift figure. The other 5 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 +70 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.