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
spec-knowledge.mjs writes stardust/spec/knowledge/ from the run's raw material and judgement/; clients read it,
never crawl or judge. Rewritten whole on each run, rows in a stable order, names neutral. A patch only adds fields or
enum values, which clients ignore when unknown; removing or renaming one is a minor release.
url_id is the id of a urls.jsonl row.
| File | Rows and main fields |
|---|---|
site.json | site, origin, scope_path, reference_blocks {name, count}, rum {available, window, bundles}, sitemap_urls, inventory_total, inventory_source, sample_note (only when capped), fetch_sources {live, headed, archive}, evidence_note (only with archive captures), built_at, built_by, provenance, variant_cut, capture_format, i18n_notes[] |
urls.jsonl | id, url, path, section, depth, in_sitemap, status, final_url, final_status, outcome, template, variant_code, title, eds_path, needs_migration_redirect, pageviews_90d, rum_bundles, traffic_band, block_count, capture_key, main_chars, flag, blocked (why an error row was not read) |
page-blocks.jsonl | per page: url_id, url, blocks[] {pos, block, variant, kind (block, dynamic, widget, default, metadata, global), source[], path, nested_in, section, crop}; chrome globals first and last |
signals.jsonl | per live sitemap page: url_id, url, signals[] (hreflang, script:<host>, iframe:<host>, jsonld:<type>, and implementation.json#signals names) |
blocks.json | name, kind (`block |
variants.json | code, template_id, rank, label, core[], optional[], url_count, pageviews_90d, distinct_sets, rep_url, rep_capture_key |
templates.json | id, label, url_count, pageviews_90d, variant_count, top_variants_share, rep_url |
source-components.json | name, url_count, instance_count, maps_to {block: instances} |
redirects.jsonl | src, target, status, hops, kind (legacy: redirects today; migration: every changed delivered path, .html included), needs_entry (migration: 1 when the path changes beyond the .html, index-name and trailing-slash rule; null on legacy), target_status, external, in_sitemap, inbound_pages, rum_views, note |
broken.jsonl | url, status, source (sitemap / link / rum), kind, in_scope, inbound_pages, inbound_main, rum_views, rum_bundles, referrers, note |
bad-links.jsonl | per dead or redirected target: to_url, main[] and chrome[] (url ids of the pages linking to it) |
redirect-landings.json | path, views, bundles |
features.json | id, class, class_name, name, evidence, disposition, reproducibility, status, pattern, eds, decisions[], reach, sitewide, reach_pages, reach_templates, pageviews_90d, pages[] (empty when site-wide) |
martech.json | consent_summary, loading_order[] {stage, what}, vendors[], launch_rules[], datalayer[], evidence {launch, onetrust, consentMode} (spec-martech's files, unchanged) |
metadata.json | name, source, used_by[], coverage_pages, distinct_values, top_values |
query-indexes.json, helix-query.yaml | name, include_paths, exclude_paths, filter, properties, consumers, source, yaml; the yaml file joins them (dynamics skeleton) |
locales.json | tree, country, language, urls, rum_views_90d, sitemap, … |
site-config.json | key, now, eds, decision |
archetypes.json | cut, types[] {type, template, variant_codes[], rep_url, urls[], url_count}: one page type per variant covering cut of its template |
header.json | url, at, summary[]: the header's size per width (controls, states, links, depth), from S9's header exploration |
search-probes.json | term, expect_count, expect_titles[], expect_includes, feature, path, param |
open-questions.json | id, area, owner (stakeholder / implementer), blocking (1: confirm the default before launch; a different answer means rework), question, context, options[], default_assumption, default_option (index of the option the default is, or null), impact_rule, impact, link, features[], answer {answer, option, by, at}, effective |
findings.json | { title, text, numbers[], chapter, rank }, plain text, every number computed; chapter of the spec's story (start, pages, build, dynamic, risks, decide), rank 1 leads it |
Feature reach, question impact and {{…}} in findings use one rule format, evaluated by rules.mjs. A rule that
cannot be computed stops S10 with its text; there is no fallback to a typed number. Extend the format in rules.mjs
when a real run needs a figure it cannot express.
field=v[,v…] or field!=… on a row; * is a case-insensitive glob; null matches a missing value;
quote values with spaces or commas (template="page · news").urls …): live (outcome page), sitemap, nested, block:<b>[|<variant>][,…],
signal:<glob>[,…], verdict:<v> (a block on the page has that verdict), or any filter on a urls.jsonl row;
! negates; terms are ANDed. Example reach: live sitemap signal:script:*player*,iframe:*player*.urls <terms…> · count <set> [filters…] · sum <set>.<field> [filters…] ·
<set>[<key>].<field> · <set>[max:<field>].<field> · value:<n>. Sets: urls, page-blocks, blocks,
block-variants, variants, templates, features, redirects, broken, bad-links, vendors, launch-rules, datalayer,
metadata, query-indexes, locales, site-config, search-probes, open-questions, source-components.urls sitemap flag=empty · count redirects kind=legacy external=1 · blocks[cards].url_count..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