Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when `evidence-review` has completed extraction and needs a bounded narrative synthesis. **Trigger**: synthesis, evidence synthesis, systematic review writing, 综合写作, SYNTHESIS.md. **Use when**: `evidence-review` 完成 screening+extraction（含 bias 评估）后进入写作阶段（C5）。 **Skip if**: 还没有 `papers/extraction_table.csv`（或 protocol/screening 尚未完成）。 **Network**: none. **Guardrail**: 以 extraction table 为证据底座；明确局限性与偏倚；不要在无数据支撑时扩写结论。
.claude/skills/willoscar-synthesis-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 109% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -52% | 0% |
Transforms the extraction table into the final evidence-review narrative.
papers/extraction_table.csvOptional:
output/PROTOCOL.mdDECISIONS.mdoutput/SYNTHESIS.mdThe synthesis must include stable sections:
## Included studies summary## Findings by theme## Risk of bias## Supported conclusions## Needs more evidencescripts/run.py should:
It should not invent findings not grounded in the extraction table.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 29,494 | 12,318 | -58% | 1 | 1 | 0% | 5,043 | 1,713 | -66% | 0 | 0 | — |
case-01 | fail→fail | 3,820 | 4,876 | +28% | 1 | 1 | 0% | 184 | 402 | +118% | 0 | 0 | — |
case-02 | fail→fail | 7,511 | 4,419 | -41% | 1 | 1 | 0% | 372 | 379 | +2% | 0 | 0 | — |
case-03 | fail→fail | 4,394 | 37,753 | +759% | 1 | 1 | 0% | 194 | 427 | +120% | 0 | 0 | — |
case-05 | fail→pass | 17,854 | 4,168 | -77% | 1 | 1 | 0% | 2,539 | 876 | -65% | 0 | 0 | — |
case-06 | fail→pass | 3,564 | 5,549 | +56% | 1 | 1 | 0% | 528 | 1,102 | +109% | 0 | 0 | — |
case-07 | pass→pass | 12,846 | 3,639 | -72% | 1 | 1 | 0% | 1,866 | 790 | -58% | 0 | 0 | — |
case-08 | fail→pass | 13,505 | 7,333 | -46% | 1 | 1 | 0% | 2,169 | 1,404 | -35% | 0 | 0 | — |
case-09 | fail→pass | 6,213 | 1,635 | -74% | 1 | 1 | 0% | 896 | 427 | -52% | 0 | 0 | — |
case-10 | fail→pass | 14,030 | 1,933 | -86% | 1 | 1 | 0% | 2,276 | 465 | -80% | 0 | 0 | — |
case-11 | fail→pass | 11,407 | 2,748 | -76% | 1 | 1 | 0% | 1,799 | 622 | -65% | 0 | 0 | — |
case-12 | fail→fail | 7,339 | 3,545 | -52% | 1 | 1 | 0% | 1,121 | 768 | -31% | 0 | 0 | — |
case-13 | fail→pass | 7,428 | 2,101 | -72% | 1 | 1 | 0% | 1,122 | 500 | -55% | 0 | 0 | — |
case-14 | fail→pass | 11,613 | 1,623 | -86% | 1 | 1 | 0% | 1,754 | 459 | -74% | 0 | 0 | — |
case-15 | fail→pass | 6,497 | 1,864 | -71% | 1 | 1 | 0% | 896 | 453 | -49% | 0 | 0 | — |
case-16 | pass→pass | 5,605 | 4,119 | -27% | 1 | 1 | 0% | 793 | 816 | +3% | 0 | 0 | — |
case-17 | pass→pass | 10,925 | 2,765 | -75% | 1 | 1 | 0% | 1,538 | 617 | -60% | 0 | 0 | — |
case-18 | pass→pass | 9,297 | 2,054 | -78% | 1 | 1 | 0% | 1,297 | 497 | -62% | 0 | 0 | — |
case-19 | pass→pass | 12,685 | 4,550 | -64% | 1 | 1 | 0% | 1,760 | 823 | -53% | 0 | 0 | — |
case-20 | pass→pass | 11,063 | 3,328 | -70% | 1 | 1 | 0% | 1,571 | 661 | -58% | 0 | 0 | — |
case-21 | fail→pass | 24,361 | 1,412 | -94% | 1 | 1 | 0% | 1,589 | 332 | -79% | 0 | 0 | — |
case-22 | pass→pass | 12,699 | 2,216 | -83% | 1 | 1 | 0% | 1,953 | 515 | -74% | 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 18 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 +50 percentage points is the difference between those two pass rates over the 18 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.