agentleFS
Sign inStart free
BrendenKennedyBrendenKennedy/claude-for-ai-platforms/.claude/skills/visualization/SKILL.md

visualization

Skill1 starsChanged 3 months agoMITView on GitHub

Charts that inform instead of mislead — for EDA, training diagnostics, and deliverables. Carries: chart-for-question mapping (distribution → histogram/ECDF; comparison → dot plot or bars-from-zero; relationship → scatter + trend; trend → line; almost never pie), the honesty rules (bar axes start at zero, shared axes across compared panels, error bars or intervals on every estimate, n stated on the figure), perceptual rules (viridis-family colormaps — never jet/rainbow; colorblind-safe categorical palettes; log scales labeled), figures-as-code (scripted + seeded, saved as tracker artifacts — no hand-tweaked screenshots), and the standard diagnostics set (learning curves, PR/calibration curves, confusion matrices, residuals, worst-case prediction grids). Load before making any figure. Triggers: plot, chart, figure, graph, visualize, matplotlib, seaborn, histogram, scatter, heatmap, colormap, colorblind, error bars, learning curve, plot the results, make a figure.

We split this skill into the rules it gives an agent

1 rule. 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every report-bound figure is produced by a script/function from tracked data (seeded when sampling), styled once via a shared mpl style, …

    line 50

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

visualization — figures that carry evidence

On-demand: load this before writing plotting code. Figures here are artifacts of record — they land in the tracker, reports (reporting), and error analyses (evaluation) — so they follow the same reproducibility rules as any result. Uncertainty shown on figures comes from statistics.

Chart follows question

QuestionReach forNot
How is X distributed?histogram (+ ECDF when comparing)a bar of the mean — it hides everything
Is A bigger than B?dot plot with intervals; bars from zerotruncated-axis bars
How do X and Y relate?scatter (+ trend, density for large n)dual-axis line charts
How did it change over training/time?line, one series per runstacked anything for >4 series
Where does the model fail?confusion matrix, PR curve, per-slice dot plot, worst-case image grida single aggregate number
Composition with >3 parts: sorted horizontal bars, not pie.

Honesty rules (the figure IS a claim)

  • Bars start at zero. Zoomed comparisons use dot plots with intervals instead — showing the interval is the honest way to magnify.
  • Compared panels share axes. Two panels with different y-scales is how "identical" reads as "improved."
  • Every point estimate carries its uncertainty — error bars / bands / n, from statistics (seed variance or bootstrap). A metric point without an interval is a vibe.
  • State n and the split on the figure (title or caption), and label log scales loudly.

Perception rules

  • Sequential data → viridis/magma family; never jet/rainbow (fake perceptual boundaries). Diverging data (± around zero) → a diverging map centered at zero.
  • Categorical → colorblind-safe palette, ≤6-8 distinguishable categories, then facet instead.
  • Dense scatters: alpha or hexbin — an opaque blob answers nothing.

Figures as code

Every report-bound figure is produced by a script/function from tracked data (seeded when sampling), styled once via a shared mpl style, saved as both PNG (viewing) and PDF/SVG (documents), and logged as a run artifact via the active tracker skill — so any figure can be regenerated when the data or model version bumps. A screenshot of a notebook cell is not a figure; it's a rumor.

Standard diagnostics per task

Training: train/val loss + LR schedule (training). Classification: confusion matrix, PR curve with the chosen operating point marked, calibration plot. Detection/segmentation: PR by class, prediction-vs-GT overlay grids, worst-K cases (evaluation's error-analysis ethos). Regression: residuals vs prediction, residuals vs key features. Forecasting: backtest fan across origins (timeseries). These are the figures a reviewer asks for — produce them before being asked.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free