Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write a clear pull-request description that gets reviewed fast and merged with confidence. Use when opening a PR, summarizing a change for review, or asked to write a PR/merge-request description. Produces a structured PR: what changed and why, how it was tested, risk and rollout, and a focused reviewer guide — so the reviewer understands intent before reading a single diff line.
.claude/skills/mohitagw15856-pr-description/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 19% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 68% | 0% |
A good PR description is a gift to the reviewer: it explains intent before they read the diff, so review is fast and confident. This skill turns a change into a structured PR write-up — what and why, how it was tested, the risk, and where to focus — the difference between a one-pass approval and three rounds of confused back-and-forth.
Ask for these only if they aren't already provided:
What & why — 2–4 sentences: the problem and what this change does about it. Link the issue (Closes #123).
Changes — the key changes as bullets (the substantive ones, not every file). Group if large.
How it was tested — tests added/updated, and the manual verification + edge cases checked. Be specific enough that the reviewer trusts it works.
Risk & rollout — blast radius, any migration/flag/config, backward-compatibility notes, and how to roll back if it goes wrong. Say "low risk, no migration" if so.
Reviewer guide — where to start, what to scrutinize, anything intentionally out of scope or deferred (with a follow-up note). Call out anything you're unsure about and want eyes on.
Screenshots / output (if UI or user-facing) — before/after.
Keep it proportional — a one-line fix gets a short description; a big change earns the full structure.
Code-review and PR best practices (explain intent, make review easy, surface risk) — modern engineering norms.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 17,886 | 14,058 | -21% | 1 | 1 | 0% | 2,404 | 2,354 | -2% | 0 | 0 | — |
case-02 | fail→fail | 14,740 | 11,874 | -19% | 1 | 1 | 0% | 1,391 | 1,973 | +42% | 0 | 0 | — |
case-03 | pass→pass | 12,536 | 8,453 | -33% | 1 | 1 | 0% | 1,495 | 1,992 | +33% | 0 | 0 | — |
case-04 | fail→fail | 11,393 | 10,435 | -8% | 1 | 1 | 0% | 1,058 | 1,614 | +53% | 0 | 0 | — |
case-05 | pass→fail | 15,888 | 29,994 | +89% | 1 | 1 | 0% | 1,852 | 5,446 | +194% | 0 | 0 | — |
case-06 | pass→pass | 9,626 | 20,162 | +109% | 1 | 1 | 0% | 1,633 | 3,666 | +124% | 0 | 0 | — |
case-07 | fail→pass | 9,509 | 9,572 | +1% | 1 | 1 | 0% | 800 | 1,341 | +68% | 0 | 0 | — |
case-08 | fail→pass | 10,837 | 13,196 | +22% | 1 | 1 | 0% | 1,150 | 1,882 | +64% | 0 | 0 | — |
case-09 | pass→pass | 13,154 | 11,889 | -10% | 1 | 1 | 0% | 1,419 | 1,841 | +30% | 0 | 0 | — |
case-10 | fail→pass | 15,164 | 14,803 | -2% | 1 | 1 | 0% | 1,755 | 2,545 | +45% | 0 | 0 | — |
case-11 | fail→pass | 16,892 | 16,886 | -0% | 1 | 1 | 0% | 1,981 | 2,362 | +19% | 0 | 0 | — |
case-12 | fail→pass | 8,300 | 15,294 | +84% | 1 | 1 | 0% | 1,438 | 2,415 | +68% | 0 | 0 | — |
case-13 | fail→pass | 14,818 | 13,789 | -7% | 1 | 1 | 0% | 1,368 | 2,170 | +59% | 0 | 0 | — |
case-14 | fail→pass | 12,621 | 8,275 | -34% | 1 | 1 | 0% | 1,324 | 1,773 | +34% | 0 | 0 | — |
case-15 | fail→pass | 9,713 | 15,301 | +58% | 1 | 1 | 0% | 1,352 | 2,057 | +52% | 0 | 0 | — |
case-16 | fail→pass | 12,806 | 16,685 | +30% | 1 | 1 | 0% | 1,677 | 2,193 | +31% | 0 | 0 | — |
case-17 | fail→pass | 14,584 | 11,208 | -23% | 1 | 1 | 0% | 1,349 | 1,773 | +31% | 0 | 0 | — |
case-18 | fail→pass | 13,222 | 13,986 | +6% | 1 | 1 | 0% | 1,424 | 1,956 | +37% | 0 | 0 | — |
case-19 | fail→fail | 8,960 | 13,585 | +52% | 1 | 1 | 0% | 1,305 | 2,099 | +61% | 0 | 0 | — |
case-20 | pass→pass | 12,272 | 11,654 | -5% | 1 | 1 | 0% | 1,295 | 1,937 | +50% | 0 | 0 | — |
case-21 | fail→pass | 11,821 | 16,993 | +44% | 1 | 1 | 0% | 1,123 | 2,338 | +108% | 0 | 0 | — |
case-22 | fail→pass | 8,423 | 11,889 | +41% | 1 | 1 | 0% | 1,231 | 1,722 | +40% | 0 | 0 | — |
case-23 | fail→pass | 14,890 | 15,458 | +4% | 1 | 1 | 0% | 1,587 | 2,288 | +44% | 0 | 0 | — |
case-24 | fail→pass | 11,935 | 14,689 | +23% | 1 | 1 | 0% | 1,214 | 2,052 | +69% | 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. 24 cases were attempted. The headline lift of +58 percentage points is the difference between those two pass rates over the 24 comparable cases. 1 case got worse with the skill loaded, and it is 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.