Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a `research-brief` workspace has a small paper set plus outline and needs a compact reader-facing briefing instead of a full survey. **Trigger**: snapshot, literature snapshot, 速览, 48h snapshot, one-page snapshot, SNAPSHOT.md. **Use when**: 你要在 `research-brief` 流程里 24-48h 内交付一个“可读的研究速览”（bullet-first，含关键引用），而不是完整 survey。
.claude/skills/willoscar-snapshot-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-23 | ✓→✗ | ▼ Worse | -58% | 0% |
Transforms a small core set plus a bullets-only outline into the final research-brief deliverable.
Required:
outline/outline.ymlpapers/core_set.csvOptional:
GOAL.mdDECISIONS.mdqueries.mdpapers/papers_dedup.jsonloutput/SNAPSHOT.mdscripts/run.py should stay a thin adapter over shared review tooling:
output/SNAPSHOT.mdDo not move topic-specific ranking rules or deep parsing logic into this script.
The output should:
output/SNAPSHOT.md exists## Scope, ## Key themes, ## What to read first, and ## Open problems / riskspapers/core_set.csvand covers at least two unique core-set papers across those sections
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,932 | 6,206 | +5% | 1 | 1 | 0% | 190 | 848 | +346% | 0 | 0 | — |
case-06 | pass→pass | 12,001 | 3,718 | -69% | 1 | 1 | 0% | 1,696 | 985 | -42% | 0 | 0 | — |
case-02 | fail→fail | 27,519 | 3,404 | -88% | 1 | 1 | 0% | 1,066 | 557 | -48% | 0 | 0 | — |
case-03 | fail→fail | 11,076 | 5,061 | -54% | 1 | 1 | 0% | 1,784 | 627 | -65% | 0 | 0 | — |
case-04 | pass→pass | 10,987 | 3,229 | -71% | 1 | 1 | 0% | 1,878 | 1,007 | -46% | 0 | 0 | — |
case-05 | fail→fail | 10,225 | 3,394 | -67% | 1 | 1 | 0% | 1,526 | 912 | -40% | 0 | 0 | — |
case-07 | fail→fail | 6,769 | 1,935 | -71% | 1 | 1 | 0% | 1,004 | 731 | -27% | 0 | 0 | — |
case-08 | fail→fail | 12,015 | 2,841 | -76% | 1 | 1 | 0% | 1,738 | 846 | -51% | 0 | 0 | — |
case-09 | fail→fail | 6,585 | 2,024 | -69% | 1 | 1 | 0% | 862 | 700 | -19% | 0 | 0 | — |
case-10 | fail→fail | 5,234 | 2,107 | -60% | 1 | 1 | 0% | 784 | 744 | -5% | 0 | 0 | — |
case-11 | fail→pass | 10,673 | 3,274 | -69% | 1 | 1 | 0% | 1,621 | 906 | -44% | 0 | 0 | — |
case-12 | fail→pass | 10,257 | 3,311 | -68% | 1 | 1 | 0% | 1,496 | 919 | -39% | 0 | 0 | — |
case-13 | pass→pass | 6,999 | 1,766 | -75% | 1 | 1 | 0% | 1,018 | 618 | -39% | 0 | 0 | — |
case-14 | fail→pass | 10,613 | 1,825 | -83% | 1 | 1 | 0% | 1,557 | 666 | -57% | 0 | 0 | — |
case-15 | pass→pass | 10,510 | 4,673 | -56% | 1 | 1 | 0% | 1,573 | 1,128 | -28% | 0 | 0 | — |
case-16 | fail→pass | 7,855 | 1,942 | -75% | 1 | 1 | 0% | 1,230 | 712 | -42% | 0 | 0 | — |
case-17 | fail→fail | 8,096 | 4,263 | -47% | 1 | 1 | 0% | 1,140 | 1,053 | -8% | 0 | 0 | — |
case-18 | fail→fail | 8,884 | 2,301 | -74% | 1 | 1 | 0% | 1,252 | 727 | -42% | 0 | 0 | — |
case-19 | fail→fail | 9,443 | 4,293 | -55% | 1 | 1 | 0% | 1,292 | 1,048 | -19% | 0 | 0 | — |
case-20 | pass→pass | 20,059 | 22,291 | +11% | 1 | 1 | 0% | 3,439 | 4,440 | +29% | 0 | 0 | — |
case-21 | pass→pass | 24,043 | 22,780 | -5% | 1 | 1 | 0% | 3,769 | 4,215 | +12% | 0 | 0 | — |
case-22 | fail→fail | 36,432 | 14,347 | -61% | 1 | 1 | 0% | 6,176 | 3,082 | -50% | 0 | 0 | — |
case-23 | pass→fail | 10,120 | 1,663 | -84% | 1 | 1 | 0% | 1,432 | 597 | -58% | 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 20 counted toward the lift figure. The other 3 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 +13 percentage points is the difference between those two pass rates over the 20 comparable cases. 3 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.