Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use for technical communication - GitHub/GitLab tickets, PR/MR descriptions, issue comments, code review comments, commit messages. Direct, brief style with no AI-speak. NOT for README.md, public docs, or blog posts.
.claude/skills/umputun-writing-style/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 66% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 107% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 116% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 29% | 0% |
Before applying this guide, check if the user already has their own writing-style rules:
writing-style skill defined in their own skills directoryIf user-defined writing rules exist: defer to those rules entirely. Do not apply this guide. Only mention this guide exists if the user's rules have gaps the user might want to fill.
If no user-defined rules exist: apply this guide as the default.
USE THIS STYLE FOR:
For these exceptions, use proper English with complete sentences, proper capitalization, no abbreviations, and professional tone.
Format:
[brief problem statement]
[what was changed/fixed]AI-generated text has recognizable patterns. Avoid these to sound natural:
Filler phrases (delete entirely):
Overused AI words (use simpler alternatives):
Abstract nouns (convert to verbs):
Hedging phrases (be direct instead):
Excessive transitions (use sparingly):
Meta-commentary (delete):
Don't use:
like thislanguage- or 1.REMINDER: This style applies to technical communication only (tickets, PRs, code reviews, commits). Use proper English for README.md, public docs, and blog posts.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 6,807 | 3,452 | -49% | 1 | 1 | 0% | 1,118 | 1,824 | +63% | 0 | 0 | — |
case-02 | fail→pass | 7,507 | 4,113 | -45% | 1 | 1 | 0% | 1,191 | 1,976 | +66% | 0 | 0 | — |
case-03 | fail→pass | 10,002 | 5,407 | -46% | 1 | 1 | 0% | 1,462 | 2,117 | +45% | 0 | 0 | — |
case-04 | pass→pass | 11,010 | 4,886 | -56% | 1 | 1 | 0% | 1,905 | 1,912 | +0% | 0 | 0 | — |
case-05 | pass→pass | 13,626 | 11,409 | -16% | 1 | 1 | 0% | 2,295 | 3,159 | +38% | 0 | 0 | — |
case-06 | pass→pass | 18,588 | 9,062 | -51% | 1 | 1 | 0% | 2,935 | 2,664 | -9% | 0 | 0 | — |
case-07 | pass→pass | 5,250 | 4,188 | -20% | 1 | 1 | 0% | 865 | 1,965 | +127% | 0 | 0 | — |
case-08 | pass→pass | 7,249 | 4,257 | -41% | 1 | 1 | 0% | 1,226 | 2,019 | +65% | 0 | 0 | — |
case-09 | pass→pass | 7,221 | 3,724 | -48% | 1 | 1 | 0% | 1,084 | 1,828 | +69% | 0 | 0 | — |
case-10 | pass→pass | 3,374 | 2,149 | -36% | 1 | 1 | 0% | 472 | 1,575 | +234% | 0 | 0 | — |
case-11 | fail→pass | 5,254 | 3,138 | -40% | 1 | 1 | 0% | 811 | 1,677 | +107% | 0 | 0 | — |
case-12 | pass→pass | 6,582 | 2,873 | -56% | 1 | 1 | 0% | 1,021 | 1,669 | +63% | 0 | 0 | — |
case-13 | fail→pass | 5,826 | 3,303 | -43% | 1 | 1 | 0% | 877 | 1,897 | +116% | 0 | 0 | — |
case-14 | pass→pass | 6,667 | 2,515 | -62% | 1 | 1 | 0% | 1,019 | 1,612 | +58% | 0 | 0 | — |
case-15 | fail→pass | 8,659 | 4,531 | -48% | 1 | 1 | 0% | 1,477 | 1,908 | +29% | 0 | 0 | — |
case-16 | pass→pass | 12,001 | 5,436 | -55% | 1 | 1 | 0% | 1,720 | 2,067 | +20% | 0 | 0 | — |
case-17 | pass→pass | 4,980 | 4,173 | -16% | 1 | 1 | 0% | 893 | 1,978 | +122% | 0 | 0 | — |
case-18 | pass→pass | 9,810 | 4,599 | -53% | 1 | 1 | 0% | 1,485 | 2,026 | +36% | 0 | 0 | — |
case-19 | pass→pass | 7,540 | 4,058 | -46% | 1 | 1 | 0% | 1,166 | 1,904 | +63% | 0 | 0 | — |
case-20 | fail→pass | 9,300 | 5,226 | -44% | 1 | 1 | 0% | 1,391 | 2,019 | +45% | 0 | 0 | — |
case-21 | pass→pass | 3,909 | 3,774 | -3% | 1 | 1 | 0% | 615 | 1,883 | +206% | 0 | 0 | — |
case-22 | pass→pass | 8,786 | 5,191 | -41% | 1 | 1 | 0% | 1,410 | 2,104 | +49% | 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 +27 percentage points is the difference between those two pass rates over the 22 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.