A wizard interface to create GitHub Agentic Workflows. Create ready-to-use GitHub Agentic Workflows in minutes — just answer a few questions and get a workflow file you can drop into your repo.
| Path | Description |
|---|---|
src/index.html |
The generator site markup |
src/styles/style.css |
Site styles |
src/js/ |
ES modules — main.js (entry), ui.js (DOM wiring), workflow.js (data-driven workflow/prompt rendering), highlight.js, next-steps.js, patterns.js, theme.js, plus the in-browser scenario assistant (slm.js, slm-runner.js, slm-cache.js, slm-ui.js) |
test/ |
Vitest unit tests for the pure generation logic |
patterns/ |
Pattern data powering the wizard — manifest.json, runtime generation data in workflow-generation.json, and one file per archetype under archetypes/ |
scripts/scan.sh |
Scanner that discovers agentic workflows across GitHub |
scripts/generate-patterns.py |
Deterministically generates patterns/ from scan results |
scripts/prompt-optimizer.mjs |
Local hourly prompt-optimization loop for the scenario assistant, driving MLC-LLM servers (small eval model, large optimizer model) |
scripts/fetch-vendor-assets.mjs |
Vendors third-party browser assets (Primer CSS, transformers.js, the onnxruntime-web wasm files) into vendor/ so the site serves them from its own origin instead of a CDN |
.github/skills/optimize-scenario-prompt/SKILL.md |
Skill for driving the prompt-optimizer harness from an agent CLI acting as the outer loop |
.github/workflows/update-patterns.yml |
Weekly scan and stable pattern-library PR automation |
.github/workflows/deep-research.md |
Read-only, manually dispatched analysis of committed pattern data |
.github/workflows/pattern-miner.md |
Daily mining of upstream github/gh-aw and githubnext/agentics workflow sources for new curated pattern entries |
The site is built with Vite and tested with Vitest.
npm install # install dependencies
npm run dev # start the dev server with hot reload
npm test # run the unit tests
npm run build # build the production site into dist/
npm run preview # serve the built site locallypatterns/ stays at the repository root (it is regenerated by the weekly scan) and is served in
development and copied into dist/ at build time.
The site hosts its third-party browser assets — Primer CSS and the scenario assistant runtime
(transformers.js plus the onnxruntime-web wasm files) — rather than loading them from a CDN, so it
keeps working on networks that block CDN hosts. npm run dev and npm run build download them into
the gitignored vendor/ directory (npm run vendor re-downloads them), and the build copies them
into dist/, which is published to GitHub Pages.
A weekly GitHub Actions workflow discovers active repos, refreshes run history, regenerates
patterns/, and creates or updates a stable pull request when the generated files change. The
deep research agent is manually dispatched and only analyzes
the committed pattern library; it never scans or writes repository content.
MIT