Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Download a small corpus of open-access arXiv survey/review PDFs about agentic systems and extract text for style learning. **Trigger**: agent survey corpus, ref corpus, download surveys, 学习综述写法, 下载 survey. **Use when**: you want to study how real agent surveys structure sections (6–8 H2), size subsections, and write evidence-backed comparisons.
.claude/skills/willoscar-agent-survey-corpus/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-17 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -36% | 0% |
Goal: create a small, local reference library so you can learn from real agent surveys when refining:
This is intentionally not part of the pipeline; it is an optional, repo-level toolkit.
ref/agent-surveys/arxiv_ids.txtref/agent-surveys/pdfs/ref/agent-surveys/text/ref/agent-surveys/STYLE_REPORT.md (tracked; auto-generated summary)1) Edit ref/agent-surveys/arxiv_ids.txt (one arXiv id per line). 2) Run the downloader to fetch PDFs and extract the first N pages to text. 3) Skim the extracted text under ref/agent-surveys/text/:
uv run python .codex/skills/agent-survey-corpus/scripts/run.py --helpuv run python .codex/skills/agent-survey-corpus/scripts/run.py --workspace . --max-pages 20--workspace <dir> (use . to write into repo root)--inputs <semicolon-separated> (default: ref/agent-surveys/arxiv_ids.txt)--max-pages <N> (default: 20)--sleep <seconds> (default: 1.0)--overwrite (re-download + re-extract)ref/:uv run python .codex/skills/agent-survey-corpus/scripts/run.py --workspace . --max-pages 20uv run python .codex/skills/agent-survey-corpus/scripts/run.py --workspace /tmp/surveys --max-pages 30--sleep, or try fewer ids.--max-pages..gitignore (ref/**/pdfs/, ref/**/text/).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 3,041 | 7,019 | +131% | 1 | 1 | 0% | 415 | 829 | +100% | 0 | 0 | — |
case-02 | fail→fail | 9,033 | 5,120 | -43% | 1 | 1 | 0% | 1,807 | 861 | -52% | 0 | 0 | — |
case-03 | fail→fail | 3,232 | 2,457 | -24% | 1 | 1 | 0% | 418 | 849 | +103% | 0 | 0 | — |
case-04 | pass→pass | 9,441 | 3,056 | -68% | 1 | 1 | 0% | 1,830 | 1,201 | -34% | 0 | 0 | — |
case-17 | fail→pass | 9,300 | 2,265 | -76% | 1 | 1 | 0% | 1,684 | 960 | -43% | 0 | 0 | — |
case-05 | fail→pass | 10,172 | 1,879 | -82% | 1 | 1 | 0% | 1,690 | 935 | -45% | 0 | 0 | — |
case-06 | fail→pass | 8,452 | 2,403 | -72% | 1 | 1 | 0% | 1,720 | 1,074 | -38% | 0 | 0 | — |
case-07 | fail→pass | 5,360 | 2,299 | -57% | 1 | 1 | 0% | 1,035 | 1,026 | -1% | 0 | 0 | — |
case-08 | fail→pass | 8,819 | 2,323 | -74% | 1 | 1 | 0% | 1,560 | 996 | -36% | 0 | 0 | — |
case-09 | fail→pass | 9,698 | 1,496 | -85% | 1 | 1 | 0% | 1,687 | 791 | -53% | 0 | 0 | — |
case-10 | fail→pass | 6,698 | 1,596 | -76% | 1 | 1 | 0% | 1,116 | 803 | -28% | 0 | 0 | — |
case-11 | fail→pass | 9,598 | 1,632 | -83% | 1 | 1 | 0% | 1,399 | 818 | -42% | 0 | 0 | — |
case-12 | fail→pass | 6,543 | 1,673 | -74% | 1 | 1 | 0% | 1,208 | 917 | -24% | 0 | 0 | — |
case-13 | fail→pass | 3,172 | 1,686 | -47% | 1 | 1 | 0% | 499 | 853 | +71% | 0 | 0 | — |
case-14 | fail→pass | 9,058 | 1,762 | -81% | 1 | 1 | 0% | 1,436 | 856 | -40% | 0 | 0 | — |
case-15 | fail→pass | 9,603 | 3,255 | -66% | 1 | 1 | 0% | 1,867 | 1,112 | -40% | 0 | 0 | — |
case-16 | pass→pass | 6,746 | 4,118 | -39% | 1 | 1 | 0% | 1,048 | 1,230 | +17% | 0 | 0 | — |
case-18 | pass→pass | 5,120 | 3,149 | -38% | 1 | 1 | 0% | 980 | 1,198 | +22% | 0 | 0 | — |
case-19 | pass→pass | 8,750 | 2,831 | -68% | 1 | 1 | 0% | 1,482 | 1,036 | -30% | 0 | 0 | — |
case-20 | pass→pass | 2,825 | 3,125 | +11% | 1 | 1 | 0% | 549 | 1,108 | +102% | 0 | 0 | — |
case-21 | pass→pass | 3,215 | 3,639 | +13% | 1 | 1 | 0% | 684 | 1,332 | +95% | 0 | 0 | — |
case-22 | pass→pass | 2,437 | 2,293 | -6% | 1 | 1 | 0% | 452 | 939 | +108% | 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 20 counted toward the lift figure. The other 2 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 +55 percentage points is the difference between those two pass rates over the 20 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.