Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Research, outline, draft, cite, and revise evidence-based articles, briefs, and reports.
.claude/skills/kunagent-content-research-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-13 | ✓→✗ | ▼ Worse | 244% | 0% |
| case-21 | ✓→✗ | ▼ Worse | 17% | 0% |
Copyright (c) 2026 KunAgent. Licensed under the MIT License.
Research, outline, draft, cite, and revise evidence-based articles, briefs, and reports.
| Tool or skill | Use | |---|---| | web_fetch | Fetch public source text. | | browser_use | Inspect interactive or multi-page public sources. | | render_chart | Show structured evidence when a chart is materially clearer. | | read | Read authorized local source files. |
Lead with the outcome, name the evidence used for verification, and disclose any real limitation that remains.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 36,460 | 15,559 | -57% | 1 | 1 | 0% | 5,499 | 1,340 | -76% | 0 | 0 | — |
case-02 | fail→fail | 50,108 | 17,319 | -65% | 1 | 1 | 0% | 8,242 | 1,410 | -83% | 0 | 0 | — |
case-03 | fail→fail | 45,188 | 28,462 | -37% | 1 | 1 | 0% | 7,396 | 5,255 | -29% | 0 | 0 | — |
case-04 | fail→fail | 3,927 | 29,738 | +657% | 1 | 1 | 0% | 584 | 4,099 | +602% | 0 | 0 | — |
case-05 | fail→pass | 10,654 | 2,754 | -74% | 1 | 1 | 0% | 1,610 | 655 | -59% | 0 | 0 | — |
case-06 | fail→fail | 16,075 | 18,032 | +12% | 1 | 1 | 0% | 2,386 | 2,927 | +23% | 0 | 0 | — |
case-07 | fail→pass | 8,718 | 3,120 | -64% | 1 | 1 | 0% | 1,342 | 777 | -42% | 0 | 0 | — |
case-08 | fail→fail | 16,125 | 18,577 | +15% | 1 | 1 | 0% | 2,421 | 3,202 | +32% | 0 | 0 | — |
case-09 | pass→pass | 18,216 | 12,231 | -33% | 1 | 1 | 0% | 2,689 | 2,161 | -20% | 0 | 0 | — |
case-10 | fail→fail | 8,827 | 37,374 | +323% | 1 | 1 | 0% | 995 | 2,476 | +149% | 0 | 0 | — |
case-11 | pass→pass | 4,047 | 3,402 | -16% | 1 | 1 | 0% | 446 | 927 | +108% | 0 | 0 | — |
case-12 | pass→pass | 14,172 | 19,868 | +40% | 1 | 1 | 0% | 2,471 | 3,401 | +38% | 0 | 0 | — |
case-13 | pass→fail | 17,878 | 97,828 | +447% | 1 | 1 | 0% | 2,673 | 9,206 | +244% | 0 | 0 | — |
case-14 | fail→fail | 23,810 | 34,085 | +43% | 1 | 1 | 0% | 3,342 | 4,538 | +36% | 0 | 0 | — |
case-15 | pass→pass | 17,355 | 15,985 | -8% | 1 | 1 | 0% | 2,096 | 2,064 | -2% | 0 | 0 | — |
case-16 | pass→pass | 9,063 | 10,142 | +12% | 1 | 1 | 0% | 1,360 | 1,870 | +38% | 0 | 0 | — |
case-17 | pass→pass | 10,138 | 10,624 | +5% | 1 | 1 | 0% | 1,657 | 2,018 | +22% | 0 | 0 | — |
case-18 | fail→pass | 17,583 | 20,524 | +17% | 1 | 1 | 0% | 2,919 | 3,694 | +27% | 0 | 0 | — |
case-19 | pass→pass | 10,199 | 6,961 | -32% | 1 | 1 | 0% | 1,474 | 1,632 | +11% | 0 | 0 | — |
case-20 | pass→pass | 12,040 | 10,850 | -10% | 1 | 1 | 0% | 1,567 | 1,647 | +5% | 0 | 0 | — |
case-21 | pass→fail | 20,485 | 20,912 | +2% | 1 | 1 | 0% | 3,052 | 3,562 | +17% | 0 | 0 | — |
case-22 | fail→fail | 8,363 | 37,398 | +347% | 1 | 1 | 0% | 1,409 | 5,967 | +323% | 0 | 0 | — |
case-23 | fail→fail | 21,796 | 21,937 | +1% | 1 | 1 | 0% | 2,351 | 3,420 | +45% | 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 19 counted toward the lift figure. The other 4 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 +4 percentage points is the difference between those two pass rates over the 19 comparable cases. 4 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.