memory/email/email-render-builder/

• Reads: the approved email creative (subject/preheader/body/CTA and its plain-text alternate) or raw HTML to QA; the destination URL; the mode (promo/cold/newsletter); target client list and any brand color/font/logo constraints; the message-match map from email-creative-builder when present.
• Writes: a user-facing HTML build (the rendered E/D unit) plus the render-QA report and a reusable handoff summary.
• Promotes: confirmed render blockers (a client that breaks the layout, an image-only block with no fallback, a dark-mode contrast failure) to

  memory/hot-cache.md

  and

  memory/open-loops.md

  ; propose durable build decisions (approved template skeleton, brand-safe dark-mode palette) as pending-decision items — never write

  decisions.md

  directly.
• Done when: the layout is a single-column responsive table that reflows on mobile, every color pair holds contrast in both light and dark mode, every image carries alt text and the email reads with images off, each CTA is a bulletproof (non-image) button, the client-render matrix names a pass/fail per target, and the plain-text alternate carries the same message and links as the HTML.
• Primary next skill: email-quality-auditor — score the built unit and run the SEND vetoes; or send-experiment-designer if the build feeds an A/B render test.

Emit the standard shape from skill-contract.md §Handoff Summary Format.

1. Confirm inputs — the approved creative (or raw HTML to QA), destination URL, mode, target client list, and brand color/font/logo constraints. If no copy and no HTML is supplied, there is nothing to build — see the Decision Gate / NEEDS_INPUT path.
2. Lay out the structure — a single-column, table-based skeleton with inline styles and a constrained content width (≈600px), from references/email-render-specs.md. Nested tables over floats/flex; no external stylesheet dependency. The layout carries the copy — it does not change a word of it.
3. Make it responsive — the single column reflows on narrow viewports; tap targets stay ≥44px; font-size stays legible without zoom on mobile. State whether the approach is fluid/hybrid or media-query-based and which clients honor it.
4. Run the dark-mode pass — check every foreground/background color pair for contrast under a dark-mode inversion; set explicit colors on text and containers so a client's forced inversion does not bury text or logos. Flag any pair that fails contrast in either mode. Per the SEND-E render lever, a body that only reads in light mode is a render defect.
5. Run the accessibility pass — semantic reading order, a meaningful

   alt

   on every image (empty

   alt=""

   only for true decoration), a language attribute, sufficient contrast, and a base font size that holds on mobile. Record each as pass/fail in the checklist from references/email-render-specs.md.
6. Specify image-off fallbacks — the email must carry its message with images blocked (many clients default to off). Every image gets alt text; no offer/claim/CTA lives only inside an image; background images have a solid fallback color; each CTA is a bulletproof (HTML/CSS, non-image) button so the click survives image-off. A hero-image-only build is a render defect, flag it.
7. Build the client-render matrix — for each target (Gmail app + web, Outlook desktop Word-engine + web, Apple Mail, iOS Mail, Android) record expected pass/fail and the specific breakage (Outlook

   mso

   conditionals, Gmail

   <style>

   stripping, unsupported CSS), labeling each row Measured (from a real seed/render test) or Estimated (from the support matrix). Use references/client-render-matrix.md.
8. Check plain-text parity — the

   text/plain

   alternate must carry the same core message, the same primary CTA, and the same destination URL as the HTML (deliverability + accessibility hygiene). If the creative shipped a plain-text alt, diff it against the HTML; if not, produce one. No image-only or HTML-only email.
9. Report defects, do not silently rewrite copy — if a render fix would require changing the words (e.g. a subject too long to render, a CTA label that will not fit a button), flag it and route back to email-creative-builder; do not edit the copy here.
10. De-slop any build notes — run humanizer-slop.md on the QA report before handoff.

• Stop and ask — no copy and no HTML supplied (nothing to build; return NEEDS_INPUT naming the missing creative or HTML); destination URL missing when the build must carry a CTA (message-match cannot be confirmed — name the missing URL). Present numbered options with their outcomes.
• Continue silently — target client list unspecified (default to the standard set: Gmail, Outlook, Apple Mail, iOS, Android, and note the assumption); brand palette unspecified (infer a neutral accessible palette and flag it); no seed/render test available (build to the support matrix and label every render row Estimated). Do not stop to ask fluid-hybrid vs media-query — pick the approach with wider client support for the target set and note it.

• Email Render Specs — the table-layout skeleton, responsive approach, dark-mode + accessibility checklists, and bulletproof-button + image-off fallback patterns
• Client Render Matrix — per-client support facts (Outlook Word engine, Gmail

  <style>

  stripping, dark-mode behavior) and the Measured/Estimated labeling rule
• SEND Benchmark — the framework; this skill produces the rendered E/D unit that email-quality-auditor scores and vetoes
• Humanizer Slop Check — pre-handoff pass that strips AI-slop phrasing from the QA report

• Primary: email-quality-auditor — score the built unit's SEND dimensions, enforce S1/S2/N1/D1, and compute the profile-weighted EQS. This skill scores nothing and runs no veto.
• If a render fix needs the copy changed (subject too long to render, CTA label overflows the button): email-creative-builder — revise the words, then return here to rebuild.
• If the build feeds a render/subject A/B test: send-experiment-designer — design the test across the built variants.
• If image-off or dark-mode breakage traces to a broken destination page (message-match fails post-click): landing-optimizer — fix the post-click page, then return.
• Global visited-set / max-depth (

  max-depth: 3

  ) termination contract from skill-contract.md applies; if the recommended next skill was already run this session, or routing is ambiguous, stop and report options instead of auto-following. Stop when the build passes the quality bar and is auditor-ready.