CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

adobe/stardust

Redesign an existing website to make it better. Built on top of impeccable.

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

parity-report.mdskills/dynamics/reference/

Dynamic parity report

Lives at stardust/dynamics/parity.json, written by the implement step, replayed by scripts/dynamics-check.mjs into stardust/qa/dynamics-report.md (+ .json), surfaced by the qa dynamics check and the rollout report. It sits next to the pixel gate and answers "what does the site do now?".

Schema

{
  "_provenance": { "writtenBy": "stardust:dynamics", "writtenAt": "<iso>" },
  "method": "one line: how features were fed / built",
  "features": [
    {
      "id": "contact-modal", "feature": "contact modal", "class": "M", "pages": 49,
      "disposition": "rebuild-native", "reproducibility": "self",
      "status": "done | interim | scaffolded-awaiting-owner | decided-out | pending <phase> | skipped-source-broken",
      "verifiedBy": "numbers: dialog 731px; heading; 22 fields; Escape closes",
      "owner": "optional: the exact decision the owner must take",
      "environmentLimit": "optional: what the test network cannot reach, egress region, why it is not a defect",
      "snapshot": "optional: date of the data snapshot the feature runs on",
      "checks": [ { "type": "click-dialog", "path": "/", "trigger": "a[href$='#modal']", "headingIncludes": "…", "minWidth": 700 } ]
    }
  ]
}

Check types (closed set, all replayable): fetch-json · dom-count · click-dialog · search-query · form-flow · video-plays · consent-gate · martech (from the martech contract, martech.md) · no-page-errors — fields in the script header. A feature with no checks is listed under "features without checks" with its status and owner; decided-out rows belong there.

search-query compares with what the SOURCE showed for the same term, read at detect time and recorded on the check: expectCount (the visible result count — a mismatch FAILS), expectTitles (the top ≤ 3 titles, compared as a set), expectIncludes (one known hit), optional titleSelector / countTolerance; two results sharing title + text are duplicates and fail. Example: { "type": "search-query", "path": "/search", "term": "bali", "resultSelector": ".search-results li", "expectCount": 3, "expectTitles": ["Bali Escape", "Bali by Bike", "Bali Food Trail"], "expectIncludes": "/tours/bali" }. A recorded hands-off run's check passed on presence while the typeahead returned 10 entries for a source answer of 3.

Rules

  1. Flows, not presence. A check passes when the user-visible flow completes. "Block rendered", "iframe present" or "video element present" is not a pass (video-plays asserts playback); an empty submission that "succeeded" was the first real defect a replay found.
  2. Numbers, not adjectives. verifiedBy carries counts and samples so a regression is diffed, not re-discovered.
  3. Secrets to the origin only. The site token rides a context.route filter on the origin host, never context-wide headers: a credentialed cross-origin XHR fails the vendor's CORS and the probe reports a vendor error real users never see. The replay records third-party request statuses next to every DOM assertion so a probe-induced failure is distinguishable from a vendor restriction.
  4. Environment limits are rows, not failures. Geo-fenced hand-off targets: fresh context, no referer, record the egress region, ask for verification from the right region.
  5. Decided-out is explicit. Reason + production statement, here and in the inconsistency register.
  6. Owner decisions by name. scaffolded-awaiting-owner lists the exact decision, so the report is honest about what runs.
  7. Re-run after every phase and before the final report. The check is read-only.

skills

README.md

tile.json