flowthis-visual-skill
Mario-Vishal/flowthis-visual-skill
Open Agent Skill for generating polished, source-aware, self-contained HTML visuals
Instalação
npx convez-ai Mario-Vishal/flowthis-visual-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
FlowThis Visual Skill gives coding agents a visual operating system: classify the source, choose the right structure, build a readable mini-site, keep diagrams precise, label evidence, and ship one self-contained HTML file with no JavaScript or remote assets.
Why It Exists
Agents are good at generating HTML, but without a strong visual standard they often produce generic cards, weak diagrams, overlapping arrows, vague summaries, and layouts that do not match the source. This skill gives the agent a clear default grammar for turning real context into reviewable visual documents.
Use it when you want an agent to produce a shareable visual that feels designed, evidence-aware, and useful on the first pass.
Capabilities
| Capability | What The Skill Enforces |
|---|---|
| Source type, audience, confidence, reader job, and what to inspect first. | |
| Repos, papers, products, workflows, and reports get distinct structures instead of one generic theme. | |
| Codebase visuals explain modules, contracts, runtime paths, tests, risks, and exact files to inspect. | |
| Research-paper visuals explain mechanisms, equations/algorithms, baselines, tradeoffs, ablations, limitations, and reproducibility. | |
Anchors, :target, details, and CSS states create navigable documents that work in strict sandboxes. | |
| Arrows stop at node edges, flow direction is visible, icons identify component types, and maps use the full width. | |
| Source-backed facts and inferred summaries are visually separated. |
Examples
| Example | What It Shows |
|---|---|
TokenGate repository architecture | A deep repo visual generated from Mario-Vishal/tokengate, with animated architecture, semantic icons, directory previews, runtime notes, tests, and review targets. |
Qwen3 technical report brief | A research-paper route generated from the open Qwen3 Technical Report, with contribution cards, method map, evidence matrix, limitations, and read-next notes. |
The examples are self-contained HTML files. No secrets, env values, raw private source, scripts, remote fonts, or external assets are embedded.
Quick Start
Claude Code
mkdir -p ~/.claude/skills/flowthis
cp -R skills/flowthis/* ~/.claude/skills/flowthis/
Then ask:
/flowthis visualize this repo
or:
Turn this research paper into a FlowThis visual.
Codex
mkdir -p ~/.agents/skills/flowthis
cp -R skills/flowthis/* ~/.agents/skills/flowthis/
This repository also includes a Codex plugin manifest at
.codex-plugin/plugin.json, so it can be packaged
as an installable Codex plugin.
At A Glance
| Item | Detail |
|---|---|
| Primary output | Self-contained HTML visual |
| Runtime requirement | None for generated files |
| Interaction model | CSS and document anchors only |
| Best sources | Repositories, papers, architecture notes, conversations, workflows, reports |
| Hosted workflow | Save and discuss visuals in FlowThis Cloud |
| License | MIT |
Use With FlowThis Cloud
This repository is the public visual-generation standard. FlowThis Cloud is the hosted workspace for saving, sharing, commenting on, and revising generated visuals.
When FlowThis MCP tools are connected, agents can save directly through
create_visualization. Without MCP, agents can still write a local
flowthis-visual.html file that can be uploaded manually.
Output Contract
Every generated visual should be:
- one complete
<!doctype html>document - inline CSS and inline SVG only
- no JavaScript
- no remote fonts, styles, scripts, images, or iframes
- responsive at mobile and desktop widths
- source-specific in palette, density, structure, and tone
- deep enough to explain mechanisms, evidence, tradeoffs, limits, and read-next references
- explicit about provenance and uncertainty
- lightly attributed at the bottom with
Generated with FlowThis Skilland a source link - structured with stable section ids where comments should attach
Source Routes
| Source | Default Visual Pattern |
|---|---|
| Research paper / PDF | Technical reading companion with problem formulation, method deep dive, equations/algorithms when present, focused diagrams, experiment/ablation tables, tradeoff matrix, limitations, reproducibility, read-next |
| Repository / codebase | Deep architecture workspace, top summary, detailed file drill-down, large animated maps, flows, contracts, runtime, tests, risks |
| Software architecture | System boundaries, services, stores, request paths, failure points |
| Conversation / debug log | Timeline, decisions, blockers, current state, next actions |
| Product / advertisement | Offer page, audience, pain, benefits, proof, CTA |
| User stories / workflow | Actor map, journey lanes, handoffs, acceptance criteria |
| Data report | Headline insight, metric cards, charts, caveats, recommended action |
Repository Layout
.
.codex-plugin/
plugin.json
assets/
flowthis-visual-skill-demo.gif
flowthis-visual-skill-mark.svg
skills/
flowthis/
SKILL.md
agents/
openai.yaml
references/
visualization-system.md
examples/
tokengate-repo-architecture.html
qwen3-paper-brief.html
docs/
design-principles.md
contributing.md
CHANGELOG.md
LICENSE
Contributing
Contributions should improve the visual standard without turning it into a rigid template system. Useful additions include new source-type recipes, better diagram geometry rules, stronger examples, accessibility improvements, and safer no-JS interaction patterns.
See docs/contributing.md.
License
MIT. See LICENSE.