Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Scaffold a new research project with standard directory structure, CLAUDE.md template, and documented README. Use this at the start of every new project to ensure consistent organization.
.claude/skills/brycewang-stanford-newproject/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -30% | 0% |
Create a new research project folder with Scott's standard structure. This skill is invoked at the start of every project.
[project-name]/
├── CLAUDE.md # Permanent research rules (copied from template)
├── README.md # Project-specific overview (auto-generated)
├── code/
│ ├── R/
│ ├── python/
│ └── stata/
├── data/
│ ├── raw/ # Original source data (never modify)
│ └── clean/ # Cleaned/merged datasets
├── output/
│ ├── tables/
│ └── figures/
├── documents/ # Outside PDFs, papers (use /split-pdf on these)
├── decks/ # Beamer presentations (rhetoric of decks)
├── notes/ # Scratch notes, random ideas, misc
└── progress_logs/ # Session continuity across Claude conversationsbash mkdir -p [project-name]/{code/{R,stata,python},data/{raw,clean},output/{figures,tables},documents,decks,notes,progress_logs}
~/mixtapetools/claude/CLAUDE.md:[Your Name] with ScottThe README must include this tree block:
markdown [project-name]/ ├── CLAUDE.md # Research rules & estimation philosophy (permanent) ├── README.md # This file — project-specific notes ├── code/ │ ├── R/ # R scripts │ ├── python/ # Python scripts │ └── stata/ # Stata do-files ├── data/ │ ├── raw/ # Original source data (never modify these) │ └── clean/ # Cleaned and merged datasets ├── output/ │ ├── tables/ # Generated tables (LaTeX, CSV) │ └── figures/ # Generated figures (PDF, PNG) ├── documents/ # Outside papers and PDFs (split with /split-pdf) ├── decks/ # Beamer presentations (rhetoric of decks philosophy) ├── notes/ # Scratch notes, ideas, miscellaneous └── progress_logs/ # Session logs for continuity across Claude conversations
progress_logs/YYYY-MM-DD_setup.md:ls, remind user to update CLAUDE.md.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,694 | 6,168 | -47% | 1 | 1 | 0% | 2,224 | 1,206 | -46% | 0 | 0 | — |
case-02 | fail→fail | 5,306 | 5,241 | -1% | 1 | 1 | 0% | 997 | 1,163 | +17% | 0 | 0 | — |
case-03 | fail→fail | 15,663 | 6,406 | -59% | 1 | 1 | 0% | 3,089 | 1,198 | -61% | 0 | 0 | — |
case-13 | fail→fail | 4,421 | 1,342 | -70% | 1 | 1 | 0% | 659 | 967 | +47% | 0 | 0 | — |
case-04 | pass→pass | 7,934 | 3,109 | -61% | 1 | 1 | 0% | 1,532 | 1,346 | -12% | 0 | 0 | — |
case-05 | fail→fail | 7,920 | 5,596 | -29% | 1 | 1 | 0% | 1,750 | 1,189 | -32% | 0 | 0 | — |
case-06 | fail→pass | 11,378 | 1,296 | -89% | 1 | 1 | 0% | 2,076 | 985 | -53% | 0 | 0 | — |
case-07 | pass→pass | 8,916 | 4,746 | -47% | 1 | 1 | 0% | 1,585 | 1,621 | +2% | 0 | 0 | — |
case-08 | fail→pass | 10,485 | 2,382 | -77% | 1 | 1 | 0% | 1,899 | 1,160 | -39% | 0 | 0 | — |
case-09 | fail→pass | 3,854 | 1,867 | -52% | 1 | 1 | 0% | 647 | 1,063 | +64% | 0 | 0 | — |
case-10 | fail→pass | 7,506 | 2,477 | -67% | 1 | 1 | 0% | 1,573 | 1,106 | -30% | 0 | 0 | — |
case-11 | fail→pass | 10,731 | 2,677 | -75% | 1 | 1 | 0% | 1,938 | 1,366 | -30% | 0 | 0 | — |
case-12 | fail→pass | 5,692 | 1,621 | -72% | 1 | 1 | 0% | 1,056 | 1,009 | -4% | 0 | 0 | — |
case-14 | fail→pass | 8,309 | 1,447 | -83% | 1 | 1 | 0% | 1,649 | 1,058 | -36% | 0 | 0 | — |
case-15 | pass→pass | 11,002 | 5,173 | -53% | 1 | 1 | 0% | 2,009 | 1,758 | -12% | 0 | 0 | — |
case-16 | fail→pass | 9,355 | 1,521 | -84% | 1 | 1 | 0% | 1,828 | 1,019 | -44% | 0 | 0 | — |
case-17 | fail→pass | 10,272 | 1,999 | -81% | 1 | 1 | 0% | 1,831 | 1,096 | -40% | 0 | 0 | — |
case-18 | fail→fail | 6,165 | 1,553 | -75% | 1 | 1 | 0% | 1,152 | 1,009 | -12% | 0 | 0 | — |
case-19 | fail→pass | 10,450 | 2,823 | -73% | 1 | 1 | 0% | 1,863 | 1,214 | -35% | 0 | 0 | — |
case-20 | pass→pass | 5,646 | 6,292 | +11% | 1 | 1 | 0% | 1,101 | 2,039 | +85% | 0 | 0 | — |
case-21 | pass→pass | 10,467 | 7,538 | -28% | 1 | 1 | 0% | 2,073 | 2,312 | +12% | 0 | 0 | — |
case-22 | pass→pass | 12,451 | 7,782 | -37% | 1 | 1 | 0% | 2,566 | 2,199 | -14% | 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 +45 percentage points is the difference between those two pass rates over the 18 comparable cases. 1 case got worse with the skill loaded, and it is 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.