Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write a PR description grounded in the REAL diff — read the branch's actual changes via the GitHub connector, not a template the user fills in. Use when asked to write my PR description, describe this pull request, draft the PR body from my branch, or document these changes in Cowork. Reads the commits and diff via the GitHub connector, derives what changed and why from the code itself, and produces a PR-description artifact (summary, changes, testing, risk) ready to paste — matching the repo's
.claude/skills/mohitagw15856-pr-description-live/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 41% | 0% |
A good PR description is written from the diff, not from memory — memory forgets the file you touched at 2am. In Claude Cowork this skill reads the actual changes on the branch and writes the description grounded in them, so the reviewer gets an accurate map of what moved and why.
Ask for these if not provided:
.github/pull_request_template.md; if present, fill its sections; else use the default structure above.Guardrails: describe only what the diff actually contains — never list a change that isn't there; surface diff-vs-intent mismatches instead of hiding them; don't push/update the PR without explicit approval; if the connector is unauthorised, work from a pasted diff and say the branch couldn't be read.
A PR Description (or the repo template, filled):
what & why — 2–3 lines
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 20,443 | 22,158 | +8% | 1 | 1 | 0% | 2,478 | 3,250 | +31% | 0 | 0 | — |
case-02 | fail→pass | 21,331 | 17,498 | -18% | 1 | 1 | 0% | 2,094 | 2,709 | +29% | 0 | 0 | — |
case-03 | fail→fail | 27,272 | 20,244 | -26% | 1 | 1 | 0% | 3,355 | 2,587 | -23% | 0 | 0 | — |
case-04 | fail→fail | 15,610 | 18,724 | +20% | 1 | 1 | 0% | 894 | 2,033 | +127% | 0 | 0 | — |
case-05 | pass→pass | 13,994 | 14,399 | +3% | 1 | 1 | 0% | 1,488 | 2,102 | +41% | 0 | 0 | — |
case-06 | fail→fail | 17,958 | 16,023 | -11% | 1 | 1 | 0% | 1,671 | 1,747 | +5% | 0 | 0 | — |
case-07 | pass→pass | 20,433 | 22,935 | +12% | 1 | 1 | 0% | 1,487 | 2,173 | +46% | 0 | 0 | — |
case-08 | pass→pass | 24,151 | 15,202 | -37% | 1 | 1 | 0% | 2,663 | 2,495 | -6% | 0 | 0 | — |
case-09 | pass→pass | 10,828 | 15,273 | +41% | 1 | 1 | 0% | 1,691 | 2,209 | +31% | 0 | 0 | — |
case-10 | fail→fail | 14,776 | 11,201 | -24% | 1 | 1 | 0% | 1,009 | 1,686 | +67% | 0 | 0 | — |
case-16 | pass→pass | 22,772 | 21,650 | -5% | 1 | 1 | 0% | 2,527 | 3,236 | +28% | 0 | 0 | — |
case-11 | pass→pass | 11,793 | 22,908 | +94% | 1 | 1 | 0% | 878 | 1,525 | +74% | 0 | 0 | — |
case-12 | fail→pass | 16,582 | 12,849 | -23% | 1 | 1 | 0% | 1,676 | 2,161 | +29% | 0 | 0 | — |
case-13 | fail→fail | 12,143 | 3,433 | -72% | 1 | 1 | 0% | 740 | 1,411 | +91% | 0 | 0 | — |
case-14 | fail→fail | 10,542 | 17,457 | +66% | 1 | 1 | 0% | 879 | 2,527 | +187% | 0 | 0 | — |
case-15 | pass→pass | 23,456 | 9,725 | -59% | 1 | 1 | 0% | 1,603 | 2,364 | +47% | 0 | 0 | — |
case-17 | fail→pass | 26,201 | 8,794 | -66% | 1 | 1 | 0% | 1,921 | 1,521 | -21% | 0 | 0 | — |
case-18 | pass→pass | 14,582 | 11,617 | -20% | 1 | 1 | 0% | 1,367 | 1,895 | +39% | 0 | 0 | — |
case-19 | pass→pass | 18,400 | 16,645 | -10% | 1 | 1 | 0% | 2,052 | 2,239 | +9% | 0 | 0 | — |
case-20 | pass→pass | 13,972 | 12,435 | -11% | 1 | 1 | 0% | 1,372 | 1,827 | +33% | 0 | 0 | — |
case-21 | fail→pass | 16,085 | 11,073 | -31% | 1 | 1 | 0% | 1,441 | 1,655 | +15% | 0 | 0 | — |
case-22 | pass→pass | 12,996 | 8,939 | -31% | 1 | 1 | 0% | 1,322 | 2,113 | +60% | 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 +18 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 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.