Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate a PR title, body, and test plan from commits and diff
.claude/skills/hashgraph-online-pr-storyteller/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | -86% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -53% | 0% |
| case-19 | ✓→✗ | ▼ Worse | 40% | 0% |
| case-11 | ✓→✗ | ▼ Worse | -70% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -77% | 0% |
Role: act as the PR description author responsible only for producing a PR title and markdown body for the current repository.
Run the local helper with Bash:
bashpython3 scripts/story.py --base main
If the result has no commits and no changed files, rerun with:
bashpython3 scripts/story.py --base master
The helper returns JSON with commits, files_changed, and suggested_title. Use only that JSON plus local diff evidence. Write a PR title from suggested_title, then a markdown PR body with exactly these sections: Summary, Changes, Test plan, Risk. Print only the PR title and markdown body.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 26,325 | 14,507 | -45% | 1 | 1 | 0% | 328 | 293 | -11% | 0 | 0 | — |
case-02 | fail→fail | 7,021 | 11,416 | +63% | 1 | 1 | 0% | 293 | 294 | +0% | 0 | 0 | — |
case-03 | fail→fail | 15,619 | 8,744 | -44% | 1 | 1 | 0% | 255 | 269 | +5% | 0 | 0 | — |
case-04 | fail→fail | 11,376 | 6,992 | -39% | 1 | 1 | 0% | 2,008 | 499 | -75% | 0 | 0 | — |
case-05 | fail→fail | 8,288 | 6,948 | -16% | 1 | 1 | 0% | 1,223 | 396 | -68% | 0 | 0 | — |
case-06 | pass→fail | 13,937 | 16,149 | +16% | 1 | 1 | 0% | 1,269 | 594 | -53% | 0 | 0 | — |
case-07 | fail→fail | 15,957 | 11,612 | -27% | 1 | 1 | 0% | 1,687 | 432 | -74% | 0 | 0 | — |
case-08 | fail→fail | 11,335 | 5,512 | -51% | 1 | 1 | 0% | 897 | 546 | -39% | 0 | 0 | — |
case-19 | pass→fail | 2,886 | 16,641 | +477% | 1 | 1 | 0% | 431 | 604 | +40% | 0 | 0 | — |
case-09 | fail→fail | 12,239 | 14,273 | +17% | 1 | 1 | 0% | 1,212 | 253 | -79% | 0 | 0 | — |
case-10 | fail→fail | 12,494 | 16,937 | +36% | 1 | 1 | 0% | 1,251 | 600 | -52% | 0 | 0 | — |
case-11 | pass→fail | 15,022 | 15,919 | +6% | 1 | 1 | 0% | 1,532 | 455 | -70% | 0 | 0 | — |
case-12 | fail→fail | 13,213 | 10,768 | -19% | 1 | 1 | 0% | 1,197 | 466 | -61% | 0 | 0 | — |
case-13 | fail→fail | 15,018 | 16,334 | +9% | 1 | 1 | 0% | 1,508 | 629 | -58% | 0 | 0 | — |
case-14 | fail→pass | 17,547 | 2,641 | -85% | 1 | 1 | 0% | 2,755 | 391 | -86% | 0 | 0 | — |
case-15 | pass→fail | 18,916 | 7,307 | -61% | 1 | 1 | 0% | 1,923 | 437 | -77% | 0 | 0 | — |
case-16 | fail→fail | 8,082 | 15,019 | +86% | 1 | 1 | 0% | 1,237 | 441 | -64% | 0 | 0 | — |
case-17 | fail→fail | 14,500 | 9,918 | -32% | 1 | 1 | 0% | 1,512 | 475 | -69% | 0 | 0 | — |
case-18 | fail→fail | 9,497 | 5,303 | -44% | 1 | 1 | 0% | 1,217 | 428 | -65% | 0 | 0 | — |
case-20 | fail→fail | 14,476 | 13,343 | -8% | 1 | 1 | 0% | 983 | 1,640 | +67% | 0 | 0 | — |
case-21 | pass→fail | 4,917 | 8,430 | +71% | 1 | 1 | 0% | 793 | 481 | -39% | 0 | 0 | — |
case-22 | pass→fail | 10,155 | 10,989 | +8% | 1 | 1 | 0% | 1,818 | 512 | -72% | 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 2 counted toward the lift figure. The other 20 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 2 comparable cases. 7 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.