Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Executes all registered notebooks, strips noisy cell metadata, and syncs Jupytext pairs. Use when asked to re-run notebooks or refresh outputs.
.claude/skills/brycewang-stanford-execute/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -71% | 0% |
Execute all registered notebooks, strip noisy metadata, and sync Jupytext pairs.
_quarto.yml and extract all notebook paths from manuscript.notebooksbash uv run jupyter execute --inplace notebooks/<name>.ipynb Record execution time and success/failure for each notebook.
.ipynb file.Open each .ipynb as JSON and remove these keys from every cell's metadata object:
execution (timestamps added by jupyter execute)_sphinx_cell_id (MyST/Sphinx artifact)vscode (VS Code editor state)Save the cleaned JSON back to the file (preserve formatting with 1-space indent).
.md pairs:bash uv run jupytext --sync notebooks/<name>.md
_quarto.yml has no notebooks registered, report "No notebooks found in _quarto.yml" and stop.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 3,183 | 5,241 | +65% | 1 | 1 | 0% | 298 | 727 | +144% | 0 | 0 | — |
case-02 | fail→fail | 3,282 | 4,377 | +33% | 1 | 1 | 0% | 395 | 655 | +66% | 0 | 0 | — |
case-03 | fail→fail | 5,341 | 4,630 | -13% | 1 | 1 | 0% | 307 | 575 | +87% | 0 | 0 | — |
case-04 | fail→pass | 20,001 | 10,123 | -49% | 1 | 1 | 0% | 992 | 2,061 | +108% | 0 | 0 | — |
case-05 | pass→fail | 6,771 | 7,206 | +6% | 1 | 1 | 0% | 1,262 | 1,714 | +36% | 0 | 0 | — |
case-06 | fail→fail | 9,022 | 4,621 | -49% | 1 | 1 | 0% | 1,702 | 1,090 | -36% | 0 | 0 | — |
case-07 | pass→pass | 9,665 | 2,624 | -73% | 1 | 1 | 0% | 1,903 | 761 | -60% | 0 | 0 | — |
case-08 | fail→pass | 7,243 | 2,011 | -72% | 1 | 1 | 0% | 1,348 | 594 | -56% | 0 | 0 | — |
case-09 | pass→fail | 7,653 | 1,555 | -80% | 1 | 1 | 0% | 1,443 | 601 | -58% | 0 | 0 | — |
case-10 | fail→fail | 18,723 | 1,407 | -92% | 1 | 1 | 0% | 3,390 | 586 | -83% | 0 | 0 | — |
case-11 | pass→fail | 6,968 | 1,289 | -82% | 1 | 1 | 0% | 1,206 | 537 | -55% | 0 | 0 | — |
case-12 | pass→pass | 6,698 | 1,877 | -72% | 1 | 1 | 0% | 1,161 | 633 | -45% | 0 | 0 | — |
case-13 | fail→pass | 4,084 | 2,412 | -41% | 1 | 1 | 0% | 784 | 816 | +4% | 0 | 0 | — |
case-14 | fail→pass | 10,119 | 2,225 | -78% | 1 | 1 | 0% | 1,639 | 709 | -57% | 0 | 0 | — |
case-15 | fail→pass | 11,194 | 1,605 | -86% | 1 | 1 | 0% | 1,916 | 560 | -71% | 0 | 0 | — |
case-16 | fail→pass | 10,890 | 1,296 | -88% | 1 | 1 | 0% | 2,034 | 483 | -76% | 0 | 0 | — |
case-17 | fail→pass | 12,360 | 3,066 | -75% | 1 | 1 | 0% | 2,016 | 897 | -56% | 0 | 0 | — |
case-18 | fail→pass | 10,055 | 3,309 | -67% | 1 | 1 | 0% | 1,787 | 885 | -50% | 0 | 0 | — |
case-19 | pass→fail | 12,248 | 3,610 | -71% | 1 | 1 | 0% | 2,278 | 1,004 | -56% | 0 | 0 | — |
case-20 | pass→pass | 2,820 | 2,267 | -20% | 1 | 1 | 0% | 512 | 720 | +41% | 0 | 0 | — |
case-21 | pass→pass | 8,408 | 2,383 | -72% | 1 | 1 | 0% | 1,298 | 672 | -48% | 0 | 0 | — |
case-22 | fail→pass | 10,913 | 2,593 | -76% | 1 | 1 | 0% | 1,883 | 758 | -60% | 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 +23 percentage points is the difference between those two pass rates over the 18 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.