Synthetic Web Team
A portable agent team that reproduces Stanford Web Services practice inside whichever AI coding tool you already use.
The badge reads its own score from badge.json, which the site deploys — no gist, no secret, no service beyond shields.io. It will show inaccessible until the first deploy publishes the file.
Open Claude Code, Cursor, VS Code, Codex, or Antigravity in a Stanford project and get the working knowledge of a full SWS web team: strategy, information architecture, content design, UX, front end, accessibility, discoverability, and delivery. The output is a Stanford site that is compliant by default and that you can actually maintain.
Everything here is Markdown, YAML, and a small CLI. No application code, no starter templates, no design system fork.
Status
Early. The plan is complete and reviewed; the implementation is partway through.
| |
|---|
| Project plan | Done. PROJECT-PLAN.md |
| Scope | Static sites, content in the repo, no CMS. standards/scope.md |
| Standards (L0) | Complete for v1. 8 policy files, 7 patterns, prior art, 3 recipes, 4 hosting profiles, the footer fragment, reference versions |
astro-static recipe | Written, and executed end to end against a real build |
astro-ssr recipe | Written, extends astro-static. Not executed, and no SWS Astro project runs SSR, so its hosting half is reasoned rather than copied |
next-ssr recipe | Written, extends astro-static. Renamed from next-netlify on 2026-09-03. Not yet executed end to end |
| Hosting profiles | New. standards/hosting/: a capability contract plus GitHub Pages, Netlify, and Vercel |
| Shared skills | 8 of 8 done, including sws-github (provisions a repo and Pages deploy through gh) and sws-preflight (is this machine equipped?) |
| Role skills | 8 of 8 built, 11 of 11 stubs. 31 skills total, all validated |
| Documentation site | Working. site/, built by our own recipe, scores 100/100 with zero failures |
sws CLI | Working. preflight reports whether the machine has the tools to build at all. doctor and check run 13 check modules against 68 criteria. sws a11y runs axe and an interactive-state audit — hover and focus measured with a real mouse and a real Tab key — and sws perf measures a byte budget, all in real Chromium |
| Report delivery | Working. PR comment and a persistent "Site health" issue, both updated in place. Score trend, sparkline, HTML artifact, README badge |
| Install wizard | Working, agent-first, two-part. user scope installs the skills into your tools once per machine and is removable; project scope merges the contract into AGENTS.md and puts the standards and the per-site record under .sws/. Non-interactive by default off a TTY, --json result with machine-readable next steps, --answers input, idempotent re-runs that preserve project state |
| Copy-a-prompt install | Working. Five prompts on the site — install into your editor, start a new site, add to an existing project, review, update — each carrying an npx route and a git clone fallback |
| Publishable packages | Working, not yet published. Two packages — @su-sws/synthetic-web-team and @su-sws/synthetic-web-team-mcp — verified by installing the tarballs into a clean project with no repository present |
| Updates | Working. Re-install is the update: project state preserved, local edits reported as conflicts rather than overwritten, stale files reported not deleted, staleness nag in sws doctor |
| Recipe canary | Deliberately deferred, 2026-09-01. Not in production, so nobody is exposed to upstream drift yet. Revisit before the first pilot |
| Standards freshness CI | Planned, and the priority ahead of the canary. Keeps policy, prior art, and sourced facts from going stale |
| MCP server | Working. packages/mcp/. 5 tools, 26 resources, verified over the real stdio protocol |
End to end, verified: synthetic-web-team installs, an agent follows the recipe, then npm run build && sws a11y && sws perf && sws check reports 100/100 with zero automated failures — 52 criteria passing, none to fix, 8 unchecked. axe runs in real Chromium against every built route and finds 0 violations at WCAG 2.1 AA; the byte budget passes at 154 KB of 800. Every one of the 8 unchecked items is genuinely unautomatable — the manual WCAG checklist, ODA review, subdomain approval, MFA attestation, the DRA question — and each says so with a reason rather than being quietly dropped.
A green axe run is a floor, not a conformance claim. It covers roughly 30 percent of accessibility issues per ODA guidance, and this project says so in the report itself.
Install
Easiest: paste a prompt into your agent
Copy the install prompt from the site
and paste it into Claude Code, Cursor, VS Code Copilot, Codex, Zed — anything that
can run a command. It tells the agent how to install, what to read first, and
to ask you for the owner names and emails rather than inventing them, because
MinWeb requires real ones.
There are five, following the two-step shape: install into your editor,
then either start a new site or add to an existing project, plus review
an existing site without changing it, and update. Each names both the npx
route and a git clone fallback, so it stays correct even where the registry is
unreachable.
Or run it yourself and answer the questions
It installs in two parts. Once per machine, into whichever AI tools you use:
npx @su-sws/synthetic-web-team install
This is the beginning of the README. Read the rest at the original. Read the full README ↗
