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 situation. A customer's EDS library (or any block set) implements the dynamic features
against relative API paths that exist only on the production host behind its CDN. On a migration
sandbox or preview origin those calls fail silently: empty dropdowns, empty bands, static
snapshots. Gates report "band shorter"; the truth is "feature dead off-origin". Name it with the
target-host probe (dynamics-plan.mjs --target-origin → host-bound), then pick a tier. The rule
throughout: feed the library, never fork it.
Libraries often read endpoints from a config document keyed by hostname. Add a set for the
migration host (POST admin.hlx.page/config/<org>/sites/<site>/public.json — GET, merge, POST the
whole object) pointing list endpoints at snapshot JSON while submit URLs stay on the source host
so real flows still complete. The production host keeps its default set.
scripts/snapshot-api.mjs records live responses from a same-origin browser context (cookies
and bot management pass; curl does not) into data/<feature>/*.json with _provenance.json.
Committed JSON is served under the same site authentication; nested JSON needs no sheet. Enumerate
parameter spaces from other snapshots. Snapshots age: keep provenance, expose data-snapshot
in the DOM, ship a sync step, show the date in the parity row.
A window.fetch wrapper installed when the hostname is a sandbox host: POST-with-term
autocomplete filtered from a local tree with live-like ranking; GET-with-params mapped to per-key
snapshot files; calls with no snapshot answered with the empty shape the block expects. Inert on
production.
Source row (API-rendered bands)When a band renders cards from an API per UI state, do not reverse-engineer the API: drive the live
UI through every state (select option → wait for that option's container → "show more" → scroll
lazy images into view) and record the rendered cards in the block's own vocabulary — every card
model the detector saw. Assert band count == block count per page before writing Source
rows (a mismatch feeds the wrong data silently). Block: Source | <url>#<bandIndex> → renders like
live, exposes data-snapshot, falls back to authored rows. Converter inserts Source only where the
snapshot has cards. Production path: the library's datasource or same-origin routing.
placeholders.json and /data/*.json drive copy and widgets. scripts/sync-sheets.mjs copies them
from the source origin into DA and publishes; rows equal on both sides.
Reused chrome links root-relative to the production host. Redirects sheet for aliases, the site query index as the "ours" oracle, a sandbox-only runtime rewrite of unserved same-origin links to the source host — then migrate the linked pages wave by wave.
Control counts, rendered link counts, a query with a known answer, option switch re-renders,
images loaded, console filtered to the library's API errors = 0. Numbers into the parity row;
replayed by dynamics-check.mjs. Hand-off targets may be geo-fenced: probe from the test network
with a fresh context and record a failure as environment-limit with the egress region, not as a
defect.
.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