IconFlow
Getting started · source install
IconFlow turns product intent into an editable SVG, then makes the design survive native pixels, platform crops, and a source-bound review before any asset ships.
Help people compare research without losing the thread.
The operating model
The agent researches the product, explores distinct concepts, and authors semantic SVG. IconFlow provides the playbook, deterministic renderer, quality gate, target transforms, and learning loop. It is not a prompt-to-image generator.
Choose your path
Use the agent path when you want help with product thinking and SVG craft. Use the CLI path when you already own the design decisions. Both end at the same review contract.
Install the open-format IconFlow skill. A compatible agent can discover the workflow when you ask for an app icon, favicon, logo mark, or tray icon.
Read the design playbook, author the SVG finalists, and use the same commands the agent runs. No hidden service sits behind the CLI.
iconflow.tomlUsing another AI agent? Any agent that can read repository instructions and run shell commands can use IconFlow through AGENTS.md plus the CLI. Automatic skill discovery depends on the client.
Five-minute engine proof
This builds IconFlow’s checked-in reviewed family into a disposable work directory. It proves your environment—not that a distinctive new identity can be designed in five minutes.
Windows · PowerShell
git clone https://github.com/snowyukitty/ai-iconflow.git
cd ai-iconflow
python -m venv .venv
powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\setup.ps1
.\.venv\Scripts\python.exe -m iconflow doctor
.\.venv\Scripts\python.exe -m iconflow ship `
--config brand\iconflow.toml `
--review brand\master-review.json `
--out work\quick-start\icon-outmacOS / Linux · shell
git clone https://github.com/snowyukitty/ai-iconflow.git
cd ai-iconflow
python3 -m venv .venv
sh ./scripts/setup.sh
./.venv/bin/python -m iconflow doctor
./.venv/bin/python -m iconflow ship \
--config brand/iconflow.toml \
--review brand/master-review.json \
--out work/quick-start/icon-outdoctor reports the renderer ready. ship verifies the checked-in source and review contract, reruns QA, and writes web, PWA, Tauri desktop, Electron, and tray assets under work/quick-start/icon-out/.
Start with intent
A noun-only prompt produces decoration. Include the product’s job, audience, personality, palette, clichés to avoid, and exact output surfaces.
Agent prompt
Use the IconFlow skill to design, review, and ship a distinctive icon family for this project.
Product intent: [what the product enables]
User job: [the progress a person wants]
Essence: [one word]
Personality: [two or three traits]
Existing palette: [tokens or colors]
Clichés to avoid: [obvious symbols]
Targets: [web, pwa, tauri, electron, tray]
Keep the editable SVG and iconflow.toml in this project. Show me the concept bake-off and native-size review before shipping. Do not stop before recording the case.Universal fallback
This works with an agent that can read files and execute the local Python CLI, even when its client does not support Agent Skills.
Read [ICONFLOW_REPO]/AGENTS.md completely.
Use [ICONFLOW_REPO]/.venv/.../python -m iconflow ...
Keep your shell in this consuming project.
Follow concept → compare → check → review → ship → case.The complete loop
Run these from the consuming project. Use the IconFlow virtual-environment Python by absolute path so your config and final sources stay with the product they belong to.
Record intent before drawing. The config also fixes targets, colors, paths, and tray behavior.
python -m iconflow init --out iconflow.toml \
--name "My App" --essence proof \
--targets web,tauri,electron,trayCreate at least four concepts through different lenses. Technique presets are scaffolds, never finished identities.
python -m iconflow styles
python -m iconflow new flat-geometric \
--out work/my-app/a.svgBake two or three finalists together. Read the sheet at 128px, 16px, and in silhouette.
python -m iconflow compare \
work/my-app/a.svg work/my-app/b.svg \
work/my-app/c.svg --out work/my-app/bake.pngFix every warning. Then inspect native pixels, alpha, adaptive crops, desktop corners, and tray contexts.
python -m iconflow check master.svg
python -m iconflow review --config iconflow.toml \
--out work/my-app/review.png \
--html work/my-app/review.htmlExport the Review Lab receipt only after every axis reaches 4/5. Source or target changes make that receipt stale.
python -m iconflow ship \
--config iconflow.toml \
--review master-review.jsonRecord what failed first, what changed, and one reusable rule. An unrecorded icon teaches nothing.
python -m iconflow case new ...
python -m iconflow case lint
python -m iconflow case statsExact output families
Targets are explicit. IconFlow does not claim “all platforms,” and a full app card does not silently become a useful menu-bar silhouette.
Favicon SVG and ICO, Apple touch icon, 192px and 512px PWA assets, plus a maskable icon and manifest snippet.
--targets web,pwaPNG sizes, ICO, ICNS, and the conventional src-tauri/icons layout for Tauri v2 desktop applications.
--targets tauriCross-platform PNG, ICO, and ICNS assets for Electron and electron-builder packaging workflows.
--targets electronColor tray PNGs, macOS template assets, and a TypeScript helper. A dedicated mark-only tray.svg is strongly preferred.
--targets trayExample build
The editable master.svg remains the source of truth beside the project. Generated files can always be rebuilt.
icon-out/
├── favicon.svg
├── favicon.ico
├── icon-192.png
├── icon-512.png
├── icon-512-maskable.png
├── icons/
│ ├── icon.ico
│ └── icon.icns
├── build/
│ ├── icon.png
│ ├── icon.ico
│ └── icon.icns
└── tray/
├── tray.png
├── trayTemplate.png
└── trayIcon.tsKnown boundaries
pip install ai-iconflow until an official release exists.Ready for your project
Run the engine proof once. Then open your product repository, give the agent a concrete brief, and keep every design decision where future maintainers can inspect it.