dataviz
/dataviz
Projektuje i renderuje czytelne, dostępne wykresy do briefów analitycznych: wykresy SVG, dobór formy, palety kolorów i kontrola wizualna.
- Autor
- bmalthi ↗
- Repozytorium
- hal-light ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 05.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 10
- Skrypty
- 2
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 05.11.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
- 2 skrypty w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/render.py, scripts/validate_palette.py
Co robi i kiedy
Projektuje i renderuje czytelne, dostępne wykresy do briefów analitycznych: wykresy SVG, dobór formy, palety kolorów i kontrola wizualna.
Oryginalny opis (EN)
Design and render clear, accessible charts for analysis briefs. Use when a brief needs inline SVG charts, chart form selection, color palettes, chart anatomy, interaction affordances, or visual QA for a quantitative finding.
Jak działa
- What to read
- Validation
- Rendering the chart
- Done when
Podgląd SKILL.md
--- name: dataviz description: > Design and render clear, accessible charts for analysis briefs. Use when a brief needs inline SVG charts, chart form selection, color palettes, chart anatomy, interaction affordances, or visual QA for a quantitative finding. --- # dataviz Use this skill before authoring charts in `brief.html` (see the `analyse` skill). HAL briefs are static: use inline SVG, visible labels and an accompanying data table. Read the interaction guidance only for requested interactive outputs; a brief must not need JavaScript, hover or a browser session to reveal its values. Adapted from Anthropic's bundled dataviz material; see the repository `NOTICE`. The upstream material is excluded from HAL Light's Apache-2.0 license. ## What to read - `references/choosing-a-form.md` — choose whether the finding needs a chart, table, stat tile, or another form. - `references/palette.md` and `references/color-formula.md` — choose chart colors and validate them. - `references/marks-and-anatomy.md` — use the expected mark, axis, label, and annotation rules. - `references/anti-patterns.md` — check for common chart failures before shipping. - `references/components.md` and `references/interaction.md` — use when the chart needs table-view, tooltip, legend, or interaction affordances. ## Validation Run the palette validator when choosing or changing colors: ```bash uv run .claude/skills/dataviz/scripts/validate_palette.py "#2a78d6,#eb6834,#1baf7a,#eda100,#e87ba4,#008300,#6250d6,#e34948" --mode light ``` Validate the colors actually used, in display order. Use `--pairs all` whenever any two series can meet, including crossing lines, scatter plots, maps and small multiples. The first three default slots pass all-pairs checks in both modes; for more series, prefer facets or another form. Run each displayed scheme. Warnings require visible labels, distinct marks or the accompanying table as specified in the report; an exit code of zero alone does not finish the design. For a lone status/text color (not a palette), check WCAG text contrast instead: ```bash uv run .claude/skills/dataviz/scripts/validate_palette.py --contrast "#0b0b0b" "#fcfcfb" ``` For an additional text-legibility measure, use `--apca TEXT BACKGROUND`. Interpret its signed Lc value with font size and weight; it does not replace WCAG contrast checks when those criteria are required. ## Rendering the chart Pick the form from `choosing-a-form.md` and the colors from `palette.md` first, as above — before either has anything to do with `render.py`. Only once the form is already decided: if it's one of **line, bar, hbar, waterfall, or combo**, hand the coordinate math to `scripts/render.py` instead of hand-rolling the SVG path/viewBox arithmetic yourself. If the data's job calls for something else — a dumbbell/range, a slope chart, stacked-area, a timeline, a dot plot, or anything else the choosing table points to — write bespoke inline SVG by hand, following the same mark specs and palette roles. `render.py` is a shortcut for five recurring shapes, never a reason to bend a chart into one of them. `render.py` takes a compact JSON spec per chart (form, title/subtitle, series or rows, palette slot names, and any annotations) and emits the inline SVG string — title, gridlines, axis labels, direct end-labels, and the fixed chrome/ink tokens are all handled for you; the CSS custom properties it emits swap light/dark on `prefers-color-scheme` automatically. You still choose: the form, which palette slot each series/bar gets (a name like `"blue"` or `"muted"`, or a literal already-validated hex), which series to plot, and every title/subtitle/annotation string. ```bash uv run .claude/skills/dataviz/scripts/render.py <run>/scratch/chart_specs.json --out-dir <run>/scratch/charts ``` Charts default to dual-scheme (light/dark swap on `prefers-color-scheme`). When a chart is bound for a one-scheme page — the analyse skill's brief template is light-only — add `--single-mode light` so the SVG is pinned to that scheme and can't flip dark on a white page. Hand-written bespoke SVG for such a page should likewise skip the dark-mode block. Write your specs to `<run>/scratch/chart_specs.json` (an id-keyed dict of specs is easiest — one call renders every chart for the brief), then inline the resulting `<run>/scratch/charts/<id>.svg` files into `brief.html`. A malformed spec — an unknown `form`, a values array that doesn't match `x`/`categories` in length — fails loudly naming the problem, rather than emitting a broken chart. See the `REQUIRED_FIELDS` and the chart functions in `scripts/render.py` for each form's fields. ## Done when The chart form matches the data's job, the palette passes validation, labels and annotations state the takeaway directly, and the values are available in a table. Check the rendered layout when a preview is available: look for clipped labels, collisions and unreadable small text. If preview is unavailable, inspect the SVG and report that layout was not visually reviewed; do not turn browser setup into a prerequisite for analysis.
Pliki w paczce (10)
- SKILL.md 5 KB
- references/anti-patterns.md 6 KB
- references/choosing-a-form.md 5 KB
- references/color-formula.md 8 KB
- references/components.md 2 KB
- references/interaction.md 4 KB
- references/marks-and-anatomy.md 7 KB
- references/palette.md 9 KB
- scripts/render.py 43 KB
- scripts/validate_palette.py 20 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · bmalthi/hal-light | 100 | Apache-2.0 |
| xobotyi/cc-foundry | 100 | MIT |
| wilbeibi/claude-toolkit | 90 | MIT |
| largepuma/libre-artifacts | 90 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 9da2c98a83f0 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90charts
Tworzy czytelne i rzetelne wykresy narzędziami charts w formatach SVG, PNG i HTML i przerabia je bez zmiany danych. - 100boardplot-chart
Rysuje wykresy, tabele i mapy konsultingowe jako SVG lub PNG oraz buduje komponenty wykresów React w Boardplot. - 100datavis
Dostarcza zestaw do wizualizacji danych w D3.js, Chart.js i własnym SVG: interaktywne wykresy, palety kolorów, skale i kodowania, potoki danych z API oraz narracje oparte na danych. - 100data-charts
Analizuje dane tabelaryczne i rysuje wykresy słupkowe, liniowe, punktowe, kołowe i histogramy do PNG lub SVG z pandas i matplotlib. - 100inspiration
Przed pisaniem kodu wykresu, mapy czy dashboardu przeszukuje galerie opublikowanych wizualizacji danych (dataviz-inspiration.com, Dataviz Index) i podaje 3 do 5 referencji z formą, adnotacjami i paletą. - 100svg-dashboard
Generuje wizualizacje danych w SVG z raportów z analiz: panele sterowane manifestem lub swobodne wykresy z danych.