Redesign an existing website to make it better. Built on top of impeccable.
88%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Low
Low-risk findings worth noting
The class axis says what a feature is. It is read from evidence, never guessed from screenshots: a site's dynamic surface is invisible to a block-scoped, pixel-verified pipeline.
| class | meaning | signals the detector reads | usual disposition |
|---|---|---|---|
| L listing | a block that lists other pages | repeated same-site card groups in the content area, dates in cards | index-backed or static-snapshot (curated) — the choice stays human |
| S search | site search, autocomplete, results | form[role=search], q/s/query inputs, first-party XHR with `search | autocomplete |
| F form | any submission | <form> with or without action, form-less control groups (modern CMS front ends render forms without a form tag), token-like hidden names, form-vendor hosts | rebuild-native with a configurable endpoint; client-only when the form is pure compute |
| M modal / interactive | overlays, dialogs, tabs, accordions, chrome interactions | `modal | dialog |
| V media | players, embeds, maps | video-js, data-video-id/data-account, player hosts, iframe (an iframe without src is a runtime-injected embed) | embed-passthrough (the player URL is the content) |
| T tag / consent | analytics, tag managers, consent, RUM, pixels, review/chat widgets | vendor table by host; first-party collector subdomains are tags that need a CNAME; mount divs with vendor data-* attributes and zero script tags = tag-manager-injected | embed-passthrough, scaffolded disabled, owner-enabled |
| A API / personalisation / settings | first-party XHR that renders content, experimentation, CMS settings objects | `/api | /graphql |
| R relationship | many-to-many listings | card groups keyed by taxonomy | static-snapshot until modelled (listings.md Tier 3) |
| X auth / commerce | sign-in, account, cart, prices, session-bound features | `login | sign-in |
| I18N locale | language trees | link[rel=alternate][hreflang] (primary — sitemaps are not), language path prefixes | a tree, not a feature: every other class recurs inside it (locale-trees.md) |
| CR client-rendered | slots or whole pages filled after load | placeholder/skeleton/js-/data-endpoint nodes with content, DOM added after load, main empty at load | static-snapshot from the settled DOM; a blank capture is a hard content gap, never migrated silently |
| D sheet / data file | copy and settings from spreadsheets or JSON files | .json GETs that are not APIs, placeholders.json, fetchPlaceholders in a library | data-fed by syncing the files from the source origin |
scripts/dynamics-detect.mjs --from-state stardust/state.json probes one page per type plus the home page (or --urls). Per page it accepts consent, settles, scrolls, settles again — tags and lazy players fire late — and records the network log by host, first-party API paths with status, POST bodies, third-party XHR, scripts, forms and control groups, the trigger → dialog → content graph, media ids, iframes, mount divs, globals and settings keys, framework, auth/commerce/locale signals, client-rendered slots and listing candidates. Findings are deduped across pages and keyed class|feature.extract --dynamics records cheap per-page signals in the crawl (endpoints, forms, triggers, players); --reach stardust/current folds them into each finding as reach: pages/of. Features come from archetypes; reach comes from the roster — the two are never the same pass.chrome only; their evidence is the header contract (dynamics-plan --chrome), not the detector.scripts/vendors.json maps host/URL patterns to a class and a role and is the classification
engine. Grow it by adding rows, never by editing detector code. Unknown third-party hosts are never
dropped: they stay visible as unknown third-party host — inspect (class A when the host answered
XHR, T otherwise) until a row classifies them. Product names only — a customer's own hosts and
account ids never enter the table.
scripts/dynamics-plan.mjs --target-origin <host> requests every recorded first-party API path on
the target. A 4xx or network error marks the finding host-bound: the feature exists in code but
is dead off-origin. A pixel gate reports this as "band shorter"; the probe is what names it. Run it
whenever a target host exists, with the site auth header scoped to that origin.
aria-controls on nav dropdowns counts as a dialog trigger only when the target has role=dialog;
media-CDN edge hosts and geo lookups appear as unknown hosts until the table knows them; four
archetypes give features but not reach.
.tessl-plugin
skills
audit
reference
deploy
reference
scripts
test
diff
direct
dynamics
extract
migrate
prepare-migration
prototype
assets
motion
fixtures
reference
qa
replica
reference
scripts
test
reskin
rollout
spec
stardust
uplift
reference