Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Identify survey/review papers in a retrieved set and extract taxonomy seeds into `outline/taxonomy.yml` (topics/subtopics/terminology). **Trigger**: survey seed harvest, taxonomy seeds, 从 survey 提 taxonomy, bootstrap taxonomy. **Use when**: retrieval/dedup 后想快速从已有 survey/review 论文中提取术语与主题结构，用于加速 `taxonomy-builder`。
.claude/skills/willoscar-survey-seed-harvest/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -8% | 0% |
Bootstrap taxonomy seeds from existing survey/review papers inside your retrieved set.
This is an accelerator for the early structure stage: it should make taxonomy-builder easier, not replace it.
papers/papers_dedup.jsonl (deduped paper metadata with titles/abstracts)outline/taxonomy.yml (seed taxonomy; expected to be refined)Uses: papers/papers_dedup.jsonl.
taxonomy-builder for domain-meaningful rewriting and scope alignment.outline/taxonomy.yml exists and is valid YAML.children used) and every node has a description.uv run python .codex/skills/survey-seed-harvest/scripts/run.py --helpuv run python .codex/skills/survey-seed-harvest/scripts/run.py --workspace <workspace>--top-k <n>: number of candidate terms to consider--min-freq <n>: minimum frequency thresholduv run python .codex/skills/survey-seed-harvest/scripts/run.py --workspace <workspace> --top-k 80 --min-freq 3taxonomy-builder.Fix:
Fix:
taxonomy-builder to rewrite under the actual scope.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 4,772 | 4,393 | -8% | 1 | 1 | 0% | 269 | 792 | +194% | 0 | 0 | — |
case-03 | fail→fail | 10,086 | 4,885 | -52% | 1 | 1 | 0% | 567 | 831 | +47% | 0 | 0 | — |
case-01 | fail→fail | 5,386 | 5,861 | +9% | 1 | 1 | 0% | 414 | 1,119 | +170% | 0 | 0 | — |
case-04 | fail→pass | 17,938 | 11,490 | -36% | 1 | 1 | 0% | 2,644 | 2,430 | -8% | 0 | 0 | — |
case-05 | pass→pass | 23,044 | 12,583 | -45% | 1 | 1 | 0% | 4,288 | 2,839 | -34% | 0 | 0 | — |
case-06 | pass→pass | 14,886 | 14,573 | -2% | 1 | 1 | 0% | 2,259 | 2,881 | +28% | 0 | 0 | — |
case-07 | pass→pass | 12,636 | 3,747 | -70% | 1 | 1 | 0% | 1,986 | 1,222 | -38% | 0 | 0 | — |
case-08 | pass→pass | 14,285 | 5,852 | -59% | 1 | 1 | 0% | 2,000 | 1,558 | -22% | 0 | 0 | — |
case-09 | fail→pass | 11,346 | 2,864 | -75% | 1 | 1 | 0% | 1,551 | 930 | -40% | 0 | 0 | — |
case-10 | pass→pass | 15,904 | 8,508 | -47% | 1 | 1 | 0% | 2,124 | 1,789 | -16% | 0 | 0 | — |
case-11 | pass→pass | 7,379 | 1,331 | -82% | 1 | 1 | 0% | 1,053 | 749 | -29% | 0 | 0 | — |
case-12 | fail→pass | 13,000 | 2,943 | -77% | 1 | 1 | 0% | 1,742 | 1,023 | -41% | 0 | 0 | — |
case-13 | fail→pass | 6,214 | 1,716 | -72% | 1 | 1 | 0% | 900 | 874 | -3% | 0 | 0 | — |
case-14 | pass→pass | 10,132 | 1,830 | -82% | 1 | 1 | 0% | 1,530 | 865 | -43% | 0 | 0 | — |
case-15 | pass→pass | 9,246 | 2,014 | -78% | 1 | 1 | 0% | 1,302 | 893 | -31% | 0 | 0 | — |
case-16 | pass→pass | 13,184 | 6,074 | -54% | 1 | 1 | 0% | 2,026 | 1,553 | -23% | 0 | 0 | — |
case-17 | fail→pass | 16,025 | 9,491 | -41% | 1 | 1 | 0% | 2,259 | 2,068 | -8% | 0 | 0 | — |
case-18 | fail→pass | 10,712 | 1,772 | -83% | 1 | 1 | 0% | 1,514 | 784 | -48% | 0 | 0 | — |
case-19 | fail→pass | 18,091 | 1,837 | -90% | 1 | 1 | 0% | 2,639 | 811 | -69% | 0 | 0 | — |
case-20 | pass→pass | 9,536 | 1,515 | -84% | 1 | 1 | 0% | 1,382 | 759 | -45% | 0 | 0 | — |
case-21 | fail→pass | 8,944 | 3,803 | -57% | 1 | 1 | 0% | 1,294 | 1,124 | -13% | 0 | 0 | — |
case-22 | fail→pass | 19,760 | 2,174 | -89% | 1 | 1 | 0% | 1,473 | 952 | -35% | 0 | 0 | — |
case-23 | pass→pass | 8,990 | 1,709 | -81% | 1 | 1 | 0% | 1,198 | 803 | -33% | 0 | 0 | — |
case-24 | pass→pass | 9,849 | 2,683 | -73% | 1 | 1 | 0% | 1,373 | 1,010 | -26% | 0 | 0 | — |
case-25 | pass→pass | 17,344 | 8,045 | -54% | 1 | 1 | 0% | 2,317 | 1,759 | -24% | 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. 25 cases were attempted, and 21 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 +36 percentage points is the difference between those two pass rates over the 21 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.