1. A Claude plugin, not a charting library
Installation is described through Claude Desktop / Cowork plugin flows. The repo contains .claude-plugin metadata and three skills under skills/, so the main product is agent behavior.
Link explainer · GitHub / Claude plugin
Chartwright for Tufte-style agent charts
A Claude plugin that packages Edward Tufte-inspired data-visualization review and rendering as three skills: route, assess, and render. The useful idea is not just “make charts prettier”; it is a workflow where agents critique chart integrity, pick a better genre, then emit inert SVG/HTML outputs with Tufte constraints baked in.
What it is
2. A Tufte ruleset encoded as skill memory
The skills point to a mirrored tufte-principles.md reference: nine scoring criteria, seven remedies, ten chart genres, chartjunk taxonomy, and named failure/exemplar catalogues.
3. A small executable renderer toolkit
Python scripts generate Tufte-style SVGs for time series, small multiples, quartile plots, and range-frame scatterplots, plus an HTML wrapper with vendored tufte-css.
Source map
| Path | Role | Why it matters |
|---|---|---|
.claude-plugin/plugin.json | Plugin metadata | Names the plugin tufte-vdqi, version 3.0.0, with data-visualization keywords. |
skills/orchestrate-tufte-vdqi/SKILL.md | Router skill | Decides whether a user needs assessment, rendering, or both. Useful because it prevents every chart request from becoming a generic “make a line chart” response. |
skills/assess-graphical-excellence/SKILL.md | Critique skill | Scores graphics, names chartjunk species, computes lie factor when relevant, and emits prioritized remedies with Tufte tags. |
skills/render-tufte-chart/SKILL.md | Output skill | Constrains generated charts: honest proportions, range-frame axes, direct labels, minimal ink, no active SVG content. |
skills/render-tufte-chart/scripts/*.py | Executable chart generators | Produces inert SVG/HTML artifacts rather than only verbal design advice. |
todos/*.md | Security/quality backlog | Shows active hardening work, especially around SVG/HTML wrapper safety. Treat as promising but still maturing. |
How the workflow is meant to run
STEP 01
User asks about a chart
“Is this misleading?”, “clean this up”, or “make a chart from this data.”
STEP 02
Router chooses path
orchestrate-tufte-vdqi routes to assessment, rendering, or assessment→rendering.
STEP 03
Assessment names faults
It checks lie factor, chartjunk, genre mismatch, data-ink ratio, typography, and uncertainty treatment.
STEP 04
Renderer builds artifact
Scripts emit SVG for Tufte-style line charts, small multiples, quartile plots, or range-frame scatterplots.
STEP 05
HTML wrapper packages it
wrap_html.py can create a shareable Tufte-css page and rejects active SVG constructs on CLI paths.
Fit for Ananth / Dab workflows
Best use: add as a reference pattern for Mission Control analytics, Daily Podcast Creator metrics, research summaries, and any Dab-generated “before/after” visual explanation where chart integrity matters more than dashboard gloss.
Mission Control analytics
Could help render calm, high-density cards for item throughput, link triage age, capture sources, or weekly focus drift without turning the Notice Board into a noisy dashboard.
Dab-generated reports
The critique skill is valuable as a preflight review: “does this chart lie?” before Dab publishes a graph into an artifact page.
Claude skill design
The repo is a good example of encoding specialized taste as skills plus runnable scripts. That pattern is directly relevant to Dab Improvements.
Maturity and caution flags
Fresh repo
Created in May 2026 and last pushed recently. Star count is healthy for a new niche plugin, but it is early-stage.
No release / no license metadata
GitHub reports no latest release and no license metadata. Before vendoring code into Dab systems, confirm the license from repo contents or ask the author.
Security backlog exists
The todos/ folder includes P1/P2 SVG/HTML hardening notes. That is a good sign of active review, but also means wrapper safety should be tested before trusting arbitrary input.
Important: do not feed untrusted SVG/HTML into this pipeline and assume it is fully sanitized. The project itself documents guard-bypass concerns around
build_html() and active SVG constructs. Use only the trusted local renderers or run a separate sanitizer/review pass before publishing generated HTML inside Mission Control.
Practical adoption path
- Read-only trial: use the repo as an idea/reference first; do not install as a required Claude plugin until license and security posture are clearer.
- Extract the workflow: for Dab, the most reusable piece is the three-stage pattern: route → critique → render, with acceptance criteria attached to each chart.
- Pilot on one internal artifact: pick a non-sensitive Mission Control metric page and generate one Tufte-style SVG manually, then inspect output and mobile fit.
- If useful, create a Dab charting skill: adapt the concepts into Dab’s own skill, with stricter local sanitization and no dependency on Claude Desktop Cowork.
Bottom line
Chartwright is worth keeping in Link Triage as a substantial Dab Improvement seed. The repo is most valuable as a design pattern for specialized agent skills with executable artifacts, and secondarily as a possible chart-rendering helper. It should not be treated as production-safe infrastructure without license clarification and a security review of the SVG/HTML wrapper path.