npm - @typeroll/mcp-server - Versions diffs - 0.33.2 → 0.35.0 - Mend

@typeroll/mcp-server 0.33.2 → 0.35.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (11) hide show

package/dist/bundled-content.js CHANGED Viewed

@@ -10,7 +10,7 @@ export const BUNDLED_SKILLS = {
     "tr-content-write": "---\nname: tr-content-write\ndescription: Use when the user asks to write, draft, or rewrite a page on a Typeroll site. Loads the site's design conventions before writing so the new content matches the existing voice and style.\n---\n\n# Write a page that fits the site\n\n> **The buffer model (draft writes).** Every content write in this recipe\n> (pages, blocks, partials, collection items) lands in an unsaved per-doc\n> DRAFT — deploys and plain previews only see SAVED content. For recipe-style\n> build work, pass `save: true` on write calls (the work is pre-approved by\n> the task itself), or run `commit_working_copy` per doc before any\n> `trigger_deploy`. Preview your drafts with `include_working_copy: true`.\n\n\nThe default failure mode for an AI writing a page is \"good generic\nHTML in the wrong voice.\" This skill makes the discovery step\nnon-optional.\n\n## Recipe\n\n### 1. Always discover first\n\n```\nget_site                          # site name (use it in copy)\nread_site_settings                # tagline, contact info, brand colors\nread_partial partial_id=\"header\"  # what other pages exist in the nav\nlist_pages limit=5\nbatch_read_pages page_ids=[<2-3 representative pages>]\n```\n\nRead the actual HTML of an existing page. Note:\n- Heading structure (single `<h1>` per page? subtitle pattern?)\n- Whether the site uses CSS variables (`var(--color-primary)`) or\n  hardcoded values\n- Tone (sober, playful, technical, marketing-y)\n- Length conventions (do existing pages run 200 words or 2000?)\n- Whether internal links use absolute or relative URLs\n\n### 2. Ask for the brief\n\nIf the user hasn't told you, ask:\n\n- **Topic + purpose**: what's the page for, who's it for?\n- **Key points**: must-include facts, calls to action\n- **Target length**: short landing vs. long-form\n- **Audience**: anything specific (existing customers, agencies,\n  developers)\n- **Reference page**: is there an existing page to match in tone or\n  structure?\n\n### 3. Draft\n\nWrite in semantic HTML, matching the site's conventions you observed\nin step 1:\n\n- Use `<section>`, `<article>`, `<h1>`/`<h2>`, `<p>`, `<ul>` — avoid\n  div soup.\n- Match the existing site's class naming or CSS variable usage. Don't\n  introduce a new design system mid-page.\n- Insert images via `<img src=\"https://cdn...\" alt=\"...\">` — use\n  `list_media` to find existing images first; only generate new ones\n  if necessary (see `tr-images` skill).\n- Default status: `draft`. Don't auto-publish unless the user said so.\n\n### 4. Create or update\n\n```\n# New page:\ncreate_page title=\"...\" slug=\"...\" html_content=\"<full body>\"\n            status=\"draft\" kind=\"page\"\n            seo_title=\"...\" seo_description=\"...\"\n\n# Or update an existing one:\nupdate_page page_id=<id> patch={ html_content: \"...\" }\n```\n\nFor an existing page, `read_page` first and preserve the existing\nstructure — replace one section at a time rather than rewriting the\nwhole body, unless the user explicitly asked for a full redo.\n\n### 5. Preview + iterate\n\n```\nget_preview_link page_id=<id>\n```\n\nShow the URL to the user. Iterate on feedback. Common rounds:\nshortening, adding a CTA, tweaking SEO description.\n\n### 6. Status change is the user's call\n\nDon't `update_page status:\"published\"` without an explicit \"looks\ngood, publish it\" from the user. Same for `trigger_deploy`.\n\n## SEO conventions worth knowing\n\n- **`kind: \"article\"`** for blog posts and news. Switches to\n  `og:type=article` + emits Article JSON-LD. Set `author` too — empty\n  author = no Person schema = no author rich-result eligibility.\n- **SEO title** target 50-60 chars. Past 60 Google truncates.\n- **Meta description** target 150-160 chars. Don't write fluff to\n  fill it; Google rewrites descriptions when they go off-topic.\n- **OG image** per page matters for shareable content. For articles\n  especially.\n\n## Pitfalls\n\n- Reading 0 pages and just inventing a design is the most common\n  failure. Always sample at least one existing page first.\n- Skipping the brief and producing 1000 words of plausible filler when\n  the user wanted a 200-word landing. Ask up front.\n- Auto-publishing. Don't.\n",
     "tr-design-review": "---\nname: tr-design-review\ndescription: Use to review a deployed/previewed Typeroll page like a designer — a MEASURED multi-dimension pass (responsive, a11y, functional, content, SEO, performance) that emits a per-dimension scorecard and an explicit OK verdict. Run it before telling the user a design is approved; it's the \"how\" for tr-redesign-branch's approval round.\n---\n\n# Review a design — measured, not glanced\n\nA design review is a MEASUREMENT, not a look. The failure mode is reporting\n\"looks good\" off a couple of screenshots — which silently misses overflow at\nuntested widths, sub-AA contrast, broken/blank images, and small touch targets.\nThis skill is the deterministic routine: per dimension, a check you RUN (a\nbrowser-eval snippet or a curl), and a scorecard you fill with PASS / FAIL /\nUNTESTED. Never report \"approved\" off a partial pass — list what you didn't test\nas caveats.\n\n`tr-redesign-branch` step 6 lists the dimensions (the \"what\"). This is the \"how\".\n\n## Cardinal rule: a screenshot is evidence, not proof\n\n**Full-page screenshots lie about lazy-loaded images.** A page with\n`loading=\"lazy\"` images below the fold will screenshot with BLANK boxes where\nthose images sit — they hadn't entered the viewport when the capture fired. If\nyou trust that, you will report a non-existent \"empty illustration box\" gap.\n(This has happened — on a real review, across three variants at once.)\n\nSo, always:\n\n- **Before any full-page capture**, scroll the whole page to trigger lazy loads\n  and let it settle (snippet in §5), THEN screenshot.\n- **Verify every suspected blank/broken image via the DOM** (`naturalWidth` after\n  scroll), never from the screenshot. A real broken image has `complete === true\n  && naturalWidth === 0`; a lazy one that just hasn't loaded has `complete ===\n  false` — scroll it into view and re-check before calling it broken.\n\n## Setup\n\n1. Get a URL for the version under review. While iterating, use the DB-live\n   `get_preview_link` (mint once, reuse — defaults to a 24h TTL) — it renders\n   from the DB with no build, so fixes show on reload without re-deploying, and\n   it's the loop for the review-fix-recheck cycle. For a FINAL bit-for-bit check\n   of the compiled output before merge, deploy once (`trigger_deploy\n   version=\"<branch>\"` → poll `get_deploy_status` → use the immutable\n   `deploy_url`, a Cloudflare Pages hash URL). Review the SAME url end to end.\n2. The snippets below run in a browser tool's \"evaluate JavaScript\" (Playwright /\n   chrome-devtools / puppeteer MCP). **One origin per eval:** the iframe trick\n   needs same-origin, so run each variant's snippet on its own page (different\n   `*.pages.dev` hashes are cross-origin → `contentDocument` is null).\n3. If you run several variants with one shared browser profile, do them\n   SEQUENTIALLY — parallel browser agents on one profile contaminate each other's\n   tabs/screenshots.\n\n## The dimensions — run each, record the result\n\n### 1. Responsive — width ladder 390/768/1024/1440/1920, zero overflow\n\nMeasure horizontal overflow at every width in ONE eval using same-origin iframes\n(each iframe is its own layout viewport, so `@media` fires correctly — no 15\nresizes):\n\n```js\nasync () => {\n  const url = location.href, out = [];\n  for (const w of [390,768,1024,1440,1920]) {\n    const f = document.createElement('iframe');\n    f.style.cssText = `width:${w}px;height:2400px;border:0;position:fixed;left:-99999px;top:0`;\n    document.body.appendChild(f);\n    await new Promise(r => { f.onload = r; f.src = url; });\n    await new Promise(r => setTimeout(r, 700));\n    const d = f.contentDocument, culprits = [];\n    for (const el of d.body.querySelectorAll('*')) {\n      const r = el.getBoundingClientRect();\n      if (r.right > w + 1 && r.width <= w + 40 && r.width > 4)\n        culprits.push(el.tagName.toLowerCase() + '.' + (el.className||'').toString().slice(0,40));\n    }\n    out.push({ w, hOverflow: d.documentElement.scrollWidth - w, n: culprits.length, sample: [...new Set(culprits)].slice(0,6) });\n    f.remove();\n  }\n  return out;\n}\n```\n\nPASS = `hOverflow <= 0` at every width. A decorative element flagged while\n`hOverflow` is 0 is clipped by an `overflow:hidden` parent (no scrollbar) — a\nnon-issue. Then eyeball one tablet (768) capture for stacking — but capture\nAFTER the scroll-settle in §5.\n\n### 2. Accessibility — compute contrast, don't eyeball\n\n```js\n() => {\n  const L = c => { const a = c.map(v => (v/=255, v<=.03928?v/12.92:((v+.055)/1.055)**2.4)); return .2126*a[0]+.7152*a[1]+.0722*a[2]; };\n  const P = c => { const m = c.match(/rgba?\\(([^)]+)\\)/); if(!m) return null; const p = m[1].split(',').map(parseFloat); return {rgb:[p[0],p[1],p[2]], a:p[3]??1}; };\n  const R = (f,b) => { const x=L(f),y=L(b),h=Math.max(x,y),l=Math.min(x,y); return (h+.05)/(l+.05); };\n  const bg = el => { let e=el; while(e){ const s=getComputedStyle(e); if(s.backgroundImage!=='none') return {img:1}; const c=P(s.backgroundColor); if(c&&c.a>.5) return {rgb:c.rgb}; e=e.parentElement; } return {rgb:[255,255,255]}; };\n  const bad=[], seen=new Set();\n  for (const el of document.body.querySelectorAll('*')) {\n    const t=[...el.childNodes].filter(n=>n.nodeType===3&&n.textContent.trim()).map(n=>n.textContent.trim()).join(' ');\n    if(!t) continue;\n    const r=el.getBoundingClientRect(); if(r.width<2||r.height<2) continue;\n    const s=getComputedStyle(el); if(s.visibility==='hidden'||s.display==='none'||+s.opacity<.1) continue;\n    const fg=P(s.color); if(!fg) continue;\n    const b=bg(el); if(b.img) continue; // can't compute over an image — eyeball hero text separately\n    const cr=R(fg.rgb,b.rgb), fs=parseFloat(s.fontSize), fw=+s.fontWeight||400;\n    const need = (fs>=24||(fs>=18.66&&fw>=700)) ? 3 : 4.5;\n    if (cr<need) { const k=t.slice(0,30)+cr.toFixed(2); if(seen.has(k))continue; seen.add(k);\n      bad.push({txt:t.slice(0,45), ratio:+cr.toFixed(2), need, fs:Math.round(fs), fw, color:s.color, bg:'rgb('+b.rgb.join(',')+')'}); }\n  }\n  return { failures: bad.length, items: bad.slice(0,15) };\n}\n```\n\nPASS = 0 failures (AA: body ≥4.5:1, large/UI ≥3:1). Fix a failure by deepening\nthe offending colour token. Text over an image background is skipped — eyeball\nthose (hero overlays) for legibility separately.\n\nStructure + alt + landmarks, same eval session:\n\n```js\n() => {\n  const h=[...document.querySelectorAll('h1,h2,h3,h4')].map(e=>+e.tagName[1]);\n  const skips=h.map((v,i)=>i&&v-h[i-1]>1?`${h[i-1]}->${v}`:0).filter(Boolean);\n  const imgs=[...document.querySelectorAll('img')];\n  const inputs=[...document.querySelectorAll('input:not([type=hidden]),textarea,select')];\n  const labelFor=new Set([...document.querySelectorAll('label[for]')].map(l=>l.getAttribute('for')));\n  return {\n    h1: h.filter(x=>x===1).length, levelSkips: skips,\n    imgsMissingAlt: imgs.filter(i=>i.getAttribute('alt')===null).length,\n    landmarks: ['header','nav','main','footer'].filter(t=>document.querySelector(t)),\n    unlabeledInputs: inputs.filter(i=>!(i.id&&labelFor.has(i.id))&&!i.getAttribute('aria-label')).map(i=>i.name||i.id),\n  };\n}\n```\n\nPASS = exactly one `h1`, `levelSkips` empty, `imgsMissingAlt` 0, all four\nlandmarks present, `unlabeledInputs` empty. (Decorative images SHOULD have\n`alt=\"\"` — that's not \"missing\".)\n\nTouch targets — interactive elements ≥44px at mobile. Run in a 390px iframe;\nEXCLUDE `aria-hidden` (the form honeypot is a visible-sized but hidden input —\ncounting it is a false positive) and inline text links inside `p`/`li`:\n\n```js\nasync () => {\n  const f=document.createElement('iframe');\n  f.style.cssText='width:390px;height:2400px;border:0;position:fixed;left:-99999px;top:0';\n  document.body.appendChild(f);\n  await new Promise(r=>{ f.onload=r; setTimeout(r,3000); f.src=location.href; });\n  await new Promise(r=>setTimeout(r,700));\n  const d=f.contentDocument, small=[];\n  if(d) for (const el of d.querySelectorAll('a,button,input:not([type=hidden]),textarea,select,[role=button]')) {\n    const r=el.getBoundingClientRect(); if(r.width<2||r.height<2) continue;\n    const s=getComputedStyle(el); if(s.display==='none'||s.visibility==='hidden'||+s.opacity<.1) continue;\n    if(el.getAttribute('aria-hidden')==='true') continue;\n    if(el.tagName==='A'&&el.closest('p,li')) continue;\n    if(r.height<44||r.width<44) small.push({tag:el.tagName.toLowerCase(), txt:(el.innerText||el.value||el.getAttribute('aria-label')||'').trim().slice(0,24), w:Math.round(r.width), h:Math.round(r.height)});\n  }\n  f.remove(); return { undersized: small };\n}\n```\n\nAlso confirm `:focus-visible` and `prefers-reduced-motion` exist (grep the page\nHTML: `grep -c 'focus-visible' page.html`, `grep -c 'prefers-reduced-motion'`).\nNote honestly: presence in CSS ≠ verified per-element — tab through live if you\nclaim keyboard focus works.\n\n### 3. Functional — console, links, form\n\n- **Console:** read the browser tool's console messages after load. PASS = 0\n  errors/warnings.\n- **Links:** PASS = no `href=\"#\"`/empty; every in-page `#anchor` has a matching\n  `id`.\n- **Form (markup — does NOT prove a live submit):** curl the page and verify the\n  `<form>` `action` is the real submit endpoint, the hidden `_token` is\n  non-empty, the honeypot is present + `aria-hidden`, required fields have\n  `required`, the email field is `type=\"email\"`. State explicitly that you did\n  NOT submit (a live POST creates a real submission) unless you actually did.\n\n### 4. Content — verbatim, no placeholders\n\n`grep -Ei 'lorem|ipsum|\\{\\{|placeholder|TODO|FIXME' page.html` → 0. Copy matches\nthe live page (the source of truth) verbatim.\n\n### 5. Broken / blank images (the anti-lazy-load check — run THIS before trusting any screenshot)\n\n```js\nasync () => {\n  const H=document.body.scrollHeight;\n  for(let y=0;y<=H;y+=400){ window.scrollTo(0,y); await new Promise(r=>setTimeout(r,120)); }\n  window.scrollTo(0,0); await new Promise(r=>setTimeout(r,1500));\n  const imgs=[...document.querySelectorAll('img')];\n  const empty=[]; // genuinely empty boxes: large, no text/img/svg/bg-image\n  for (const el of document.querySelectorAll('div,section,figure')) {\n    const r=el.getBoundingClientRect(); if(r.width<160||r.height<140) continue;\n    if((el.innerText||'').trim()||el.querySelector('img,svg,picture,canvas,video')) continue;\n    if(getComputedStyle(el).backgroundImage!=='none') continue;\n    empty.push({cls:(el.className||'').toString().slice(0,36), w:Math.round(r.width), h:Math.round(r.height)});\n  }\n  return {\n    broken: imgs.filter(i=>i.complete&&i.naturalWidth===0).map(i=>i.src.slice(-45)), // real failures\n    stillLoading: imgs.filter(i=>!i.complete).map(i=>i.src.slice(-45)),               // lazy, scroll first\n    emptyBoxes: empty.slice(0,8),                                                     // true placeholders\n  };\n}\n```\n\nPASS = `broken` empty, `emptyBoxes` empty. A non-empty `emptyBoxes` is a genuine\nunfilled illustration slot (fill it — pages shouldn't be text deserts). NOW\ncapture screenshots (the page is scrolled-and-settled, images loaded).\n\n### 5b. Clipped artwork — the logo (and any brand image) cut off by its own frame\n\nThe single most-repeated visual bug: the header logo rendered with its top/edges\nsliced. It produces ZERO page overflow (§1 misses it), the image isn't broken\n(§5 misses it), and at full-page screenshot scale a few clipped pixels are easy\nto glance past. So MEASURE it: does the artwork's rendered content touch the edge\nof its own box on any side? Content flush against the frame (gap ≈ 0) = clipped\nor about-to-clip. Don't just check the logo — check it, then trust the number.\n\nFor a raster/`<img>` logo, draw it to a same-origin canvas and scan the border\nrows/cols for opaque pixels (cross-origin taints the canvas — fetch the asset to\na localhost file first, as in §setup, or measure on the asset directly):\n\n```js\nasync (url) => { // url = the logo's currentSrc, served same-origin\n  const img = new Image(); await new Promise((r,e)=>{img.onload=r;img.onerror=e;img.src=url;});\n  const h = 64, w = Math.round(h*img.naturalWidth/img.naturalHeight);\n  const c = document.createElement('canvas'); c.width=w; c.height=h;\n  const x = c.getContext('2d'); x.drawImage(img,0,0,w,h);\n  const d = x.getImageData(0,0,w,h).data, op=(px)=>d[px*4+3]>20;\n  let top=h,bot=0,left=w,right=0;\n  for(let y=0;y<h;y++)for(let xx=0;xx<w;xx++)if(op(y*w+xx)){top=Math.min(top,y);bot=Math.max(bot,y);left=Math.min(left,xx);right=Math.max(right,xx);}\n  return { topGap:top, bottomGap:h-1-bot, leftGap:left, rightGap:right }; // any 0 → flush/clipped\n}\n```\n\nPASS = every gap ≥ ~2% of the dimension. A `0` on any side means the artwork (or\nits stroke) sits on the frame — for an SVG that's a viewBox trimmed flush to the\nart (look for `-trim`/`-tight` in the filename); the fix is to re-export the SVG\nwith viewBox padding (e.g. widen `viewBox` by ~8% each side) so the stroke never\ntouches the edge. Verify the fix by re-running this with the patched asset. (A\nheavy `stroke-width` + `paint-order=\"stroke\"` outline makes a flush viewBox clip\nvisibly — and small header renders make it worse, so also check the logo isn't\nshrunk below ~64–72px in the header.)\n\nAlso confirm no ANCESTOR clips the logo: walk the logo's parents for\n`overflow:hidden|clip` combined with a fixed height or negative/overlap margin —\nand always judge the logo from a screenshot of the header REGION in context,\nnever the logo element in isolation (an element screenshot re-renders the full\nart and hides the clip).\n\n**The inverse bug — an image FLOATING inside its frame (don't blame the file).**\nA full-bleed illustration that renders with a margin of empty frame around it\nusually isn't a bad asset — it's CSS. In blocks-mode the site-template's global\n`:where(.page-content) img{ margin:1rem 0 }` (and a default `border-radius`)\nleaks onto any `<img>` you didn't reset, so a framed hero/figure gets a 1rem gap\ninside its frame and looks like it \"floats\". Before re-cropping or regenerating,\n**open the actual image file** (`curl` the `.avif`/`.png`) — if the motif fills\nthe file edge-to-edge, the float is CSS: set `margin:0` (and `border-radius:0`)\non the framed `<img>` (e.g. `.your-frame img{margin:0}` or a blanket\n`.your-scope img{margin:0}`). Measure it: the `<img>`'s `getBoundingClientRect`\nshould equal its frame's inner box (no gap). Only when the *file itself* has\nbuilt-in background margin (motif ≪ frame) is cropping the right fix.\n\n### 6. Findable (SEO/meta) — curl, fast\n\n`<title>` (≤60 chars) + meta description present + sensible; `og:title/description/image`;\n`canonical`; `favicon` + `apple-touch-icon`; `<html lang>`; branches must be\n`noindex`. One curl + greps covers it.\n\n### 7. Fast (performance)\n\nPASS signals: no render-blocking JS you didn't add; responsive variants\n(`srcset` + AVIF/WebP) so a 1024px asset isn't shipped to a 380px slot;\n`width`/`height` or aspect-ratio set (no layout shift); **below-fold images\n`loading=\"lazy\"`, above-fold `eager`**. Flag a section that eager-loads every\nimage, or a multi-hundred-KB original served when a small AVIF variant exists.\n\n### 8. Cross-browser\n\nThe same CSS renders differently per engine. Re-check in another engine if you\ncan. If only Chromium is available, say so as UNTESTED and statically flag risky\nprops: `backdrop-filter` without fallback, `-webkit-`-only masks, `100vh` on\nmobile (prefer `100svh`), `position:sticky` inside `overflow`.\n\n## Deliver a scorecard + an explicit verdict\n\nReport a table — one row per dimension, value PASS / FAIL(detail) / UNTESTED —\nthen a one-line verdict. Rules:\n\n- \"Approved\" requires PASS on responsive, a11y, functional, content, SEO,\n  performance. Untested dimensions (commonly cross-browser, live form submit) are\n  listed as CAVEATS, not silently dropped — an OK with caveats is honest; an\n  unqualified \"approved\" off a partial pass is not.\n- Brand FIT (palette/voice matching `brand.md`) is a direction judgment, not a\n  pass/fail defect — call it out separately so the user decides direction.\n- If you fixed anything mid-review, just reload the DB-live preview and re-run\n  the affected dimension before signing off — no re-deploy needed (deploy only\n  for the final compiled-output check, if any).\n\nSee `tr-redesign-branch` for the surrounding branch → preview → approve → merge\nflow; this skill is its measured approval round.\n",
     "tr-directory": "---\nname: tr-directory\ndescription: Use when the user wants to build a directory site or import a structured dataset (restaurants, products, events, agencies, etc.) where each item should have its own URL. Covers collection schema creation, per-item URLs via route_template, listing page, deploy.\n---\n\n# Build a directory site from external data\n\n> **The buffer model (draft writes).** Every content write in this recipe\n> (pages, blocks, partials, collection items) lands in an unsaved per-doc\n> DRAFT — deploys and plain previews only see SAVED content. For recipe-style\n> build work, pass `save: true` on write calls (the work is pre-approved by\n> the task itself), or run `commit_working_copy` per doc before any\n> `trigger_deploy`. Preview your drafts with `include_working_copy: true`.\n\n\nTyperoll collections support per-item URLs: every published item\nin a collection with a `route_template` materialises as its own static\npage at build time. This is the right pattern when you have hundreds\nof similar entities (restaurants, products, listings, profiles).\n\n## Big-picture flow\n\n1. **Data source** → 2. **Collection schema** → 3. **Items** → 4. **Listing page**\n→ 5. **Preview** → 6. **Deploy**\n\nYou drive everything from Claude Code locally — the scrape, the data\nshaping, the writes. The MCP just receives the final shape.\n\n## Recipe\n\n### 1. Get the data\n\nWhatever source the user has — scraped CSV, public API, vendor feed,\nmanual research, another LLM's output. Normalise to a flat shape:\none object per item with stable, kebab-case field names.\n\n```jsonc\n[\n  {\n    \"title\": \"Joe's Pizza\",\n    \"slug\": \"joes-pizza\",\n    \"address\": \"123 Main St, Anytown\",\n    \"phone\": \"+1-555-0100\",\n    \"cuisine\": \"italian\",\n    \"rating\": 4.5,\n    \"image\": \"https://...\",        // optional: a hosted image URL\n    \"excerpt\": \"Family-run since 1987...\",\n    \"body\": \"<p>Long-form description with HTML.</p>\"\n  }\n]\n```\n\nThe `slug` field is what populates `route_template`. Make it\nkebab-case, unique within the dataset. If the source doesn't have one,\nderive from `title`: lowercase, replace non-alphanumeric with `-`,\ncollapse consecutive dashes.\n\n### 2. Decide the URL structure with the user\n\nCommon patterns:\n\n- `/restaurants/{slug}` (default — simple, predictable)\n- `/r/{slug}` (compact)\n- `/{cuisine}/{slug}` (categorised)\n- `/dir/{slug}` (short prefix to avoid collisions with page slugs)\n\nPick one before creating the collection — changing `route_template`\nlater renames every URL and requires redirect rules.\n\n### 3. Create the collection schema\n\n```\ncreate_collection\n  name=\"restaurants\"\n  label_singular=\"Restaurant\"\n  label_plural=\"Restaurants\"\n  icon=\"🍕\"\n  fields=[\n    {\"name\":\"title\",\"label\":\"Name\",\"type\":\"text\",\"required\":true},\n    {\"name\":\"slug\",\"label\":\"Slug\",\"type\":\"text\",\"required\":true},\n    {\"name\":\"address\",\"label\":\"Address\",\"type\":\"text\"},\n    {\"name\":\"phone\",\"label\":\"Phone\",\"type\":\"text\"},\n    {\"name\":\"cuisine\",\"label\":\"Cuisine\",\"type\":\"text\"},\n    {\"name\":\"rating\",\"label\":\"Rating\",\"type\":\"number\"},\n    {\"name\":\"image\",\"label\":\"Image URL\",\"type\":\"text\"},\n    {\"name\":\"excerpt\",\"label\":\"Excerpt\",\"type\":\"textarea\"},\n    {\"name\":\"body\",\"label\":\"Body\",\"type\":\"richtext\"}\n  ]\n  slug_field=\"slug\"\n  sort_field=\"title\"\n  sort_dir=\"asc\"\n  route_template=\"/restaurants/{slug}\"\n  item_template_html=\"<article class=\\\"directory-item\\\">\n    <header>\n      <h1>{{title}}</h1>\n      {{cuisine}} · ⭐ {{rating}}\n    </header>\n    <img src=\\\"{{image}}\\\" alt=\\\"{{title}}\\\" />\n    <address>{{address}} · <a href=\\\"tel:{{phone}}\\\">{{phone}}</a></address>\n    <section class=\\\"description\\\">{{{body}}}</section>\n  </article>\"\n```\n\nThe `item_template_html` is what renders for each item. `{{field}}`\nHTML-escapes; `{{{field}}}` leaves raw (use for richtext bodies that\nintentionally carry HTML).\n\n### 4. Bulk-import items\n\nLoop over your data array. For each item:\n\n```\ncreate_collection_item\n  collection=\"restaurants\"\n  fields={ title:\"Joe's Pizza\", slug:\"joes-pizza\", ... }\n  status=\"published\"\n```\n\nFor larger datasets (1000+), batch outside the MCP — spawn 5 parallel\n`create_collection_item` calls at a time, watch the 60-writes/min rate\nlimit (you'll hit it on big imports, the API returns 429 with\n`Retry-After`).\n\nSet `status: \"draft\"` for items the user still needs to review; only\npublished items get static pages.\n\n### 5. Build a listing page\n\nItems have per-item URLs but not a default index. Create one with a\nmarker pair that `regenerate_collection_listing` will keep up to date:\n\n```\ncreate_page\n  title=\"Restaurants\"\n  slug=\"restaurants\"\n  status=\"published\"\n  html_content=\"<h1>All restaurants</h1>\n    <!-- typeroll:listing:restaurants -->\n    <!-- /typeroll:listing:restaurants -->\n  \"\n```\n\nThen populate (and refresh whenever items change) with one call:\n\n```\nregenerate_collection_listing\n  collection=\"restaurants\"\n  page_id=\"restaurants\"\n  item_template=\"<article class=\\\"directory-card\\\">\n    <h2><a href=\\\"{{url}}\\\">{{title}}</a></h2>\n    <p>{{cuisine}} · ⭐ {{rating}}</p>\n    <p>{{address}}</p>\n  </article>\"\n  wrap_open=\"<div class=\\\"directory-grid\\\">\"\n  wrap_close=\"</div>\"\n```\n\n`{{field}}` substitutes HTML-escaped, `{{{field}}}` raw (for richtext\nfields), `{{url}}` resolves through the collection's `route_template`.\nThe tool replaces only what's between the marker pair — anything before\nor after the markers stays put.\n\nWhen the customer adds a new restaurant later, the agent re-runs the\nsame `regenerate_collection_listing` call and the index updates. No\ndiff-the-HTML-by-hand, no stale listings.\n\n(When the block editor lands, you'll be able to drop in a \"collection\nlisting\" block instead of hand-writing this. For now, raw HTML.)\n\n### 6. Preview an item\n\n```\nget_preview_link collection_name=\"restaurants\" item_id=\"<id>\"\n```\n\nReturns a URL the user can open. Internal links inside the preview\nstay inside the preview surface, so navigating to another item works.\n\n### 7. Deploy\n\n```\ntrigger_deploy\nget_deploy_status job_id=<id>\n```\n\nEach published item gets its own URL in the static build, with\n`sitemap.xml` automatically including them all.\n\n## Patterns worth knowing\n\n### Conditional rendering without Mustache conditionals\n\n`item_template_html` substitution is plain `{{field}}` / `{{{field}}}` — no loops, no `{{#if}}` blocks. To hide a section/element when a field is empty, use a data-attribute that resolves to either the empty string or a non-empty value, plus a CSS selector:\n\n```html\n<aside class=\"podcast-guest\" data-empty-if-blank=\"{{guest_name}}\">\n  <h2>Om gästen</h2>\n  <p>{{guest_bio}}</p>\n</aside>\n```\n```css\n.podcast-detail [data-empty-if-blank=\"\"] { display: none !important; }\n```\n\nWhen `guest_name` is empty, the attribute becomes `data-empty-if-blank=\"\"` and the CSS matches and hides the block. When non-empty, the rule misses and the block renders. Works for optional images, optional audio, optional sub-sections — any \"show only if this field has a value\" need.\n\n### Pre-render list-typed source data into a richtext field\n\nFor nested arrays in the source (e.g. `chapters: [{time, title}, …]`), pre-render to HTML during import and store in a dedicated `*_html` richtext field. The template just splats `{{{chapters_html}}}`. Three concrete recipes worth applying:\n\n1. **Podcast detail page:** `chapters_html`, `guest_links_html`.\n2. **Restaurant directory:** `hours_html` table.\n3. **Product directory:** `variants_html` grid.\n\nSee `tr-collection-template` for full code examples.\n\n### Batch-import via subagent for large datasets\n\n22+ `create_collection_item` calls bloat the main agent's context with response payloads. Spawn a `general-purpose` subagent with the manifest path and a tight contract (\"report ok/fail per item, end with a one-line summary\"). The sub returns one line per item instead of a JSON blob per item back into the main turn.\n\n## Pitfalls\n\n- **Slugs must be unique within the collection** — duplicates cause\n  build failures (two pages claiming the same URL). De-dupe before\n  importing.\n- **Don't reuse `slug` across collections without thinking.**\n  `/restaurants/joes` and `/products/joes` are fine; just avoid\n  `/joes` for both (collection items vs. pages don't collide because\n  pages always win, but two collections sharing a `slug_field=slug`\n  with the same `route_template` is a foot-gun).\n- **Required fields.** `route_template=\"/restaurants/{slug}\"` will\n  silently skip items where `slug` is missing. Check\n  `list_collection_items` after import — if you imported 500 and the\n  listing only shows 480, look at the dropped 20's source data.\n- **Template too clever.** Substitution is plain `{{field}}` — no\n  loops, no conditionals. If your design needs more, prefer flat\n  fields (`star_html`, `rating_label`) prebuilt in the data step.\n  See the \"Patterns worth knowing\" section above.\n- **Field type changes drift data.** Adding a new field after import\n  is fine; renaming one orphans the old data on every item. Plan the\n  schema before import.\n- **Never derive display labels from slugs.** Slugs are ASCII-folded\n  for URL-safety. Computing the visible label as\n  `slug.replace(\"-\", \" \").capitalize()` produces **wrong words** in\n  languages with diacritics: `innehall` → `Innehall` (should be\n  `Innehåll`), `affarssystem` → `Affärssystem`. Always carry the\n  real title from the source and look it up by slug:\n\n  ```python\n  slug_to_title = {t[\"slug\"]: t[\"title\"] for t in topic_manifest}\n  label = slug_to_title.get(slug, slug.replace(\"-\", \" \").title())  # fallback only\n  ```\n\n  Caught during a real migration where 20 of 22 podcast detail pages\n  ended up with `Innehall` / `Prissattning` / `Affarssystem` in their\n  topic chips.\n- **Numeric `sort_field` sorts numerically** as of the 2026-05 fix —\n  episode 9 ranks below episode 23 under desc sort. Earlier versions\n  compared as strings (9 > 23), so if you're working against an older\n  portal deploy add a `sort_key` text field with zero-padded values\n  (`f\"{episode:03d}\"`) and set `sort_field: \"sort_key\"` instead.\n\n## Mixing scraped + generated content\n\nThe whole point of the local-agent model: you can blend sources.\n\n- Scrape addresses + phone from a yellow-pages site.\n- Generate excerpt + body from a local Claude pass over the raw\n  scraped HTML.\n- Generate hero images per item via `tr-images`.\n- All three merged into one `create_collection_item` per record.\n\nKeep a local manifest (`./directory-state.json`) of what's been\nimported so a partial run is resumable. The MCP doesn't track that\nstate — your local script does.\n",
-    "tr-forms": "---\nname: tr-forms\ndescription: Use when the user wants to add a contact form, booking form, or any web form to their Typeroll site. Triggers on \"lägg till formulär\", \"contact form\", \"add a form\", \"let visitors message us\", \"booking form\", \"formulär\", or similar. Covers both creating the form definition and embedding the HTML widget on a page.\n---\n\n# Add a form to a Typeroll site\n\nTyperoll forms are server-backed: submissions go to\n`/api/forms/submit` (HMAC-signed, rate-limited, honeypot-protected)\nand are stored in Firestore. The user sees them in the portal's\nSubmissions inbox. No third-party service needed.\n\n## Preconditions\n\n- Site exists and MCP is configured.\n- You know what fields the form needs.\n\n## Forms 2.0 — steps mode (template_capabilities_version ≥ 0.18.0)\n\nPrefer this over the hand-built core/html embed when the portal supports\nit. A form's `steps[]` are Block[] trees of `form/*` field blocks mixed\nwith content blocks; place `{ type: 'core/form', data: { form_id } }` on\nthe page and the build renders everything (token, honeypot, runtime,\nproof-of-work) — you never hand-write the <form> markup.\n\n```\nupdate_form form_id=ansokan patch={ steps: [\n  { id: 'steg1', title: 'Om företaget', blocks: [\n    { type: 'form/text',  data: { name: 'foretag', label: 'Företag', required: true } },\n    { type: 'form/email', data: { name: 'epost', label: 'E-post', required: true } } ] },\n  { id: 'steg2', title: 'Detaljer', blocks: [\n    { type: 'form/textarea', data: { name: 'meddelande', label: 'Meddelande' } },\n    { type: 'form/consent',  data: { name: 'gdpr', text: '<p>Jag godkänner …</p>' } } ] } ] }\nadd_block target={kind:'page', id:'kontakt'} block={ type: 'core/form', data: { form_id: 'ansokan' } }\n```\n\nValidation derives from the field blocks (required/pattern/min/max) —\nthere is no separate field list to maintain. Partial submissions persist\nper step (TTL `partial_ttl_days`, default 30). Form-scoped CSS goes in\n`styles`; field look is themable via `--form-field-*` tokens in the\nsite's custom_css. On older portals, use the legacy recipe below.\n\n**Simple forms don't need hand-built steps** (template_capabilities_version\n≥ 0.29.0, `forms_steps_only` capability): steps are the ONLY stored model,\nand a plain `create_form` with `fields[]` is authoring sugar — the server\nconverts the list to a single static step (read_form shows the resulting\nsteps). Every form renders through the `core/form` block (styled inputs,\nclient validation, token + honeypot all handled). So for any single-step\nform: `create_form` with `fields`, then\n`add_block { type: 'core/form', data: { form_id } }`. Reach for the\nraw-HTML embed (below) only on older portals or HTML-mode pages — and note\nit supports single-step forms only.\n\n## Recipe\n\n### 1. Create the form definition\n\n```\ncreate_form {\n  \"id\": \"kontakt\",\n  \"name\": \"Kontaktformulär\",\n  \"fields\": [\n    {\"name\":\"name\",    \"label\":\"Namn\",          \"type\":\"text\",     \"required\":true},\n    {\"name\":\"email\",   \"label\":\"E-post\",         \"type\":\"email\",    \"required\":true},\n    {\"name\":\"phone\",   \"label\":\"Telefon\",        \"type\":\"text\"},\n    {\"name\":\"message\", \"label\":\"Meddelande\",     \"type\":\"textarea\", \"required\":true},\n    {\"name\":\"subject\", \"label\":\"Ämne\",           \"type\":\"select\",\n     \"options\":[\"Prisförfrågan\",\"Samarbete\",\"Övrigt\"]}\n  ],\n  \"success_message\": \"Tack! Vi återkommer inom 24 timmar.\"\n}\n```\n\nThere is no `recipient_email` field — submissions land in the portal's\nSubmissions inbox (`/app/sites/{siteId}/forms/{formId}/submissions`).\nEmail notification is a form *action* (not yet wired platform-side);\ntell the customer to check the inbox.\n\n**Field types:** `text`, `email`, `tel`, `url`, `number`, `textarea`,\n`select`, `radio`, `checkbox`.\n\n**Field name rules:** Lowercase ASCII only: `[a-z][a-z0-9_-]*`.\n- `besokt` not `besökt` (`ö→o`)\n- `foretag` not `företag` (`ö→o`, `ä→a`)\n- `meddelande` not `Meddelande` (the `name` must be lowercase; `label` can be anything)\n\n### 2. Get the signed embed token\n\nThe submit endpoint only accepts requests carrying a platform-signed\nHMAC token. `create_form` returns it directly; you can also fetch it any\ntime with:\n\n```\nread_form form_id=\"kontakt\"\n```\n\nThe response includes `submit_token` (put it in the hidden `_token`\ninput) and `submit_url` (use it as the form's `action`). The token is\nstable — it only stops working if the platform rotates its signing\nsecret — so baking it into static page HTML is correct. If\n`submit_token` comes back `null`, the server has no signing secret\nconfigured (dev setups); the form cannot accept submissions until that's\nfixed — tell the user instead of embedding a broken form.\n\n### 3. Embed the form on a page\n\nA plain HTML form POST is the default and needs **no JavaScript**: the\nendpoint answers a normal form-encoded POST with a small confirmation\npage (the form's `success_message` + a link back to the page the\nvisitor came from). Validation errors get the same treatment. The HTML:\n\n```html\n<section class=\"contact-section\">\n  <div class=\"contact-container\">\n    <h2>Kontakta oss</h2>\n    <p>Fyll i formuläret så återkommer vi inom 24 timmar.</p>\n\n    <form class=\"contact-form\"\n          action=\"SUBMIT_URL\"\n          method=\"POST\">\n      <!-- The signed token is the only hidden field the platform needs;\n           it encodes org + site + form identity. -->\n      <input type=\"hidden\" name=\"_token\"    value=\"SIGNED_TOKEN\">\n      <!-- Honeypot — must stay empty, bots fill it -->\n      <input type=\"text\"   name=\"_hp\"       style=\"display:none\" tabindex=\"-1\" autocomplete=\"off\">\n\n      <div class=\"form-group\">\n        <label for=\"name\">Namn *</label>\n        <input type=\"text\" id=\"name\" name=\"name\" required>\n      </div>\n\n      <div class=\"form-group\">\n        <label for=\"email\">E-post *</label>\n        <input type=\"email\" id=\"email\" name=\"email\" required>\n      </div>\n\n      <div class=\"form-group\">\n        <label for=\"message\">Meddelande *</label>\n        <textarea id=\"message\" name=\"message\" rows=\"5\" required></textarea>\n      </div>\n\n      <div class=\"form-group\">\n        <label for=\"subject\">Ämne</label>\n        <select id=\"subject\" name=\"subject\">\n          <option value=\"Prisförfrågan\">Prisförfrågan</option>\n          <option value=\"Samarbete\">Samarbete</option>\n          <option value=\"Övrigt\">Övrigt</option>\n        </select>\n      </div>\n\n      <button type=\"submit\" class=\"btn-primary\">Skicka meddelande</button>\n    </form>\n  </div>\n</section>\n\n<style>\n.contact-section{padding:4rem 2rem}\n.contact-container{max-width:600px;margin:0 auto}\n.form-group{margin-bottom:1.5rem}\n.form-group label{display:block;font-weight:600;margin-bottom:0.4rem;font-size:0.9rem}\n.form-group input,.form-group textarea,.form-group select{\n  width:100%;padding:0.75rem 1rem;border:1px solid var(--color-surface);\n  border-radius:0.375rem;font-family:inherit;font-size:1rem;\n  background:var(--color-surface);color:var(--color-text)\n}\n.form-group textarea{resize:vertical}\n.btn-primary{\n  background:var(--color-primary);color:#fff;border:none;\n  padding:0.875rem 2rem;border-radius:0.375rem;font-size:1rem;\n  font-weight:600;cursor:pointer;width:100%\n}\n.btn-primary:hover{opacity:0.9}\n</style>\n```\n\n**Replace:**\n- `SUBMIT_URL` → `submit_url` from `read_form` / `create_form`\n- `SIGNED_TOKEN` → `submit_token` from the same response\n\n### 4. Inline feedback (optional — requires user action)\n\n**The page sanitizer strips inline `<script>` from page and partial\nHTML**, so you cannot ship fetch-based submit handling yourself — a\n`<script>` you put in `html_content` is silently removed. The plain\nPOST above is the reliable path; use it.\n\nIf the customer wants inline feedback without a page reload, the script\nmust go into the site's `scripts_body_end` setting, which only a human\ncan edit (Settings → Custom code — it's deliberately excluded from the\nAI tool surface). Hand them this snippet for that box; it intercepts the\nform and POSTs the JSON shape `{ token, data }`:\n\n```html\n<script>\n(function(){\n  const form = document.querySelector('.contact-form');\n  if(!form) return;\n  form.addEventListener('submit', async (e) => {\n    e.preventDefault();\n    const btn = form.querySelector('[type=submit]');\n    btn.disabled = true;\n    btn.textContent = 'Skickar…';\n    // The endpoint's fetch contract is JSON: { token, data }.\n    const fd = new FormData(form);\n    const token = fd.get('_token');\n    const data = {};\n    fd.forEach((v, k) => { if (k !== '_token') data[k] = v; });\n    try {\n      const res = await fetch(form.action, {\n        method: 'POST',\n        headers: { 'Content-Type': 'application/json' },\n        body: JSON.stringify({ token, data }),\n      });\n      const json = await res.json();\n      if (res.ok && json.success) {\n        form.innerHTML = '<p class=\"form-success\">' + (json.message || 'Tack!') + '</p>';\n      } else {\n        btn.disabled = false;\n        btn.textContent = 'Skicka meddelande';\n        alert('Något gick fel: ' + ((json.errors && json.errors.join(', ')) || json.error || 'okänt fel'));\n      }\n    } catch {\n      btn.disabled = false;\n      btn.textContent = 'Skicka meddelande';\n      alert('Nätverksfel — försök igen.');\n    }\n  });\n})();\n</script>\n```\n\n### 5. Update the contact page with the form HTML\n\n```\nupdate_page page_id=\"kontakt\" patch={\n  \"html_content\": \"<full page HTML including the form section>\"\n}\n```\n\n### 6. Verify\n\n```\nread_form form_id=\"kontakt\"\nlist_forms\n```\n\nConfirm the form appears and fields match what you embedded.\n\n### 7. Deploy\n\n```\ntrigger_deploy\nget_deploy_status job_id=<id>\n```\n\nAfter deploy, test by submitting the live form. Submissions appear in the\nportal at `/app/sites/{siteId}/forms/kontakt/submissions`.\n\n## Common form patterns\n\n### Booking / appointment request\n```json\n{\"fields\": [\n  {\"name\":\"name\",  \"type\":\"text\",  \"label\":\"Namn\",   \"required\":true},\n  {\"name\":\"email\", \"type\":\"email\", \"label\":\"E-post\",  \"required\":true},\n  {\"name\":\"date\",  \"type\":\"text\",  \"label\":\"Önskat datum (YYYY-MM-DD)\"},\n  {\"name\":\"time\",  \"type\":\"select\",\"label\":\"Tid\",    \"options\":[\"09:00\",\"10:00\",\"11:00\",\"14:00\",\"15:00\"]},\n  {\"name\":\"notes\", \"type\":\"textarea\",\"label\":\"Kommentar\"}\n]}\n```\n\n### Newsletter signup (minimal)\n```json\n{\"fields\": [\n  {\"name\":\"email\", \"type\":\"email\", \"label\":\"E-postadress\", \"required\":true}\n]}\n```\n\n### Job application\n```json\n{\"fields\": [\n  {\"name\":\"name\",       \"type\":\"text\",  \"label\":\"Namn\",      \"required\":true},\n  {\"name\":\"email\",      \"type\":\"email\", \"label\":\"E-post\",     \"required\":true},\n  {\"name\":\"role\",       \"type\":\"select\",\"label\":\"Roll\",       \"options\":[\"Designer\",\"Projektledare\",\"Övrigt\"]},\n  {\"name\":\"experience\", \"type\":\"textarea\",\"label\":\"Berätta om dig själv\"},\n  {\"name\":\"portfolio\",  \"type\":\"url\",   \"label\":\"Portfolio-URL\"}\n]}\n```\n\n## Pitfalls\n\n- **Tokens are stable, not expiring.** The `submit_token` stays valid\n  until the platform rotates its signing secret (rare, operator-driven).\n  Bake it into the static HTML; no refresh logic needed. If submissions\n  suddenly 403 after working, re-fetch via `read_form` and republish.\n- **Inline `<script>` does not survive.** The page sanitizer strips it\n  from `html_content` — never rely on client JS you embed yourself. The\n  plain form POST works without it (section 3).\n- **Field names must be lowercase ASCII** (`[a-z][a-z0-9_-]*`):\n  `foretag` not `företag`, `amne` not `ämne`. Labels can be anything.\n- **Don't use the same form_id on two different forms.** IDs must be\n  unique per site — use descriptive names: `kontakt`, `boka`, `nyhetsbrev`.\n- **Honeypot must be invisible.** `_hp` field must have `display:none`.\n  If it's visible and a real user fills it, their submission is rejected.\n- **Email notifications are admin-only — not settable via MCP.** Submissions\n  are stored and visible in the portal's Forms → Submissions inbox. A site\n  admin can also configure post-submission emails (admin notification +\n  autoresponder) under **Forms → <form> → Email**, after setting up an email\n  connector under **Settings → Email & notifications**. These carry recipient\n  addresses + templates over submission data, so they're deliberately off the\n  agent surface (`create_form`/`update_form` ignore `actions`). Point the\n  customer at those screens; you can't set them up for them.\n",
+    "tr-forms": "---\nname: tr-forms\ndescription: Use when the user wants to add a contact form, newsletter signup, booking form, or other web form to a Typeroll site.\n---\n\n# Add a form to a Typeroll site\n\nTyperoll forms are server-backed. The platform renders the form shell, accepts\nsigned submissions, validates declared fields, stores submissions, and can run\nadmin-configured email or webhook actions.\n\n## Choose the placement\n\n- Block-mode page: add `{ type: 'core/form', data: { form_id } }`.\n- HTML-mode page: add `<x-form id=\"form-id\" />` to `html_content`.\n\nBoth are authoring references to the same renderer. Preview and static\ngeneration expand them to the complete form HTML, signed token, honeypot,\ninitial step state, styles, and shared runtime. Never hand-write the `<form>`\nshell or paste a token into page HTML.\n\n## Create a simple form\n\n`fields[]` is authoring sugar for one static step:\n\n```json\n{\n  \"id\": \"newsletter\",\n  \"name\": \"Newsletter\",\n  \"fields\": [\n    {\"name\":\"email\", \"type\":\"email\", \"label\":\"E-postadress\", \"required\":true}\n  ],\n  \"submit_text\": \"Prenumerera\",\n  \"success_message\": \"Tack! Du är anmäld.\"\n}\n```\n\nField names must match `[a-z][a-z0-9_-]*`. Labels may be localized. Supported\nsimple types include `text`, `email`, `tel`, `url`, `number`, `textarea`,\n`select`, `radio`, `checkbox`, `hidden`, and `gdpr_consent`.\n\n## Place the form\n\nOn a block-mode page:\n\n```\nadd_block target={kind:'page', id:'start'}\n  block={type:'core/form', data:{form_id:'newsletter'}}\n```\n\nOn an HTML-mode page:\n\n```html\n<section class=\"newsletter-signup\">\n  <h2>Få våra nyheter</h2>\n  <x-form id=\"newsletter\" />\n</section>\n```\n\n`<x-form>` is not a browser-side shortcode. It is replaced during server\npreview/build, so it also supports multi-step forms and never needs inline JS.\n\n## Multi-step form\n\nFor a funnel, write `steps[]` directly. Each step is a block tree containing\n`form/*` field blocks and optional content blocks:\n\n```\nupdate_form form_id=ansokan patch={steps:[\n  {id:'company', title:'Företag', blocks:[\n    {type:'form/text', data:{name:'company', label:'Företag', required:true}},\n    {type:'form/email', data:{name:'email', label:'E-post', required:true}}\n  ]},\n  {id:'details', title:'Detaljer', blocks:[\n    {type:'form/textarea', data:{name:'message', label:'Meddelande'}},\n    {type:'form/consent', data:{name:'consent', text:'<p>Jag godkänner …</p>'}}\n  ]}\n]}\n```\n\nSubmissions accumulate in one partial record and become complete on the final\nstep. Abandoned partials use `partial_ttl_days` (default 30).\n\n## Storage and integrations\n\nCompleted submissions appear in Forms → Submissions. Admins can configure\nactions in the form editor:\n\n- Email notification or autoresponder through the site's email connector.\n- Generic webhook to a public HTTPS endpoint. The admin chooses an explicit\n  field allowlist and signing secret. Typeroll signs the exact request body in\n  `X-Typeroll-Signature`, sends an idempotency key, retries transient failures,\n  and stores delivery status.\n\nActions are deliberately excluded from MCP/API-key writes and reads because\nthey can exfiltrate submitted data. Direct the user to the portal form editor\nto configure them.\n\n## Verify\n\n1. `read_form form_id=\"newsletter\"` and confirm the steps/fields.\n2. Preview the page and confirm the authoring reference has expanded to a form\n   with `data-tr-form-el`, a signed token, and the platform runtime.\n3. Submit a test entry and confirm it appears in Forms → Submissions.\n4. If a webhook is configured, confirm its delivery status and the receiving\n   system's idempotency key before deploying.\n\n## Common patterns\n\nNewsletter:\n\n```json\n{\"fields\":[{\"name\":\"email\",\"type\":\"email\",\"label\":\"E-postadress\",\"required\":true}]}\n```\n\nContact:\n\n```json\n{\"fields\":[\n  {\"name\":\"name\",\"type\":\"text\",\"label\":\"Namn\",\"required\":true},\n  {\"name\":\"email\",\"type\":\"email\",\"label\":\"E-post\",\"required\":true},\n  {\"name\":\"message\",\"type\":\"textarea\",\"label\":\"Meddelande\",\"required\":true}\n]}\n```\n\nBooking request:\n\n```json\n{\"fields\":[\n  {\"name\":\"name\",\"type\":\"text\",\"label\":\"Namn\",\"required\":true},\n  {\"name\":\"email\",\"type\":\"email\",\"label\":\"E-post\",\"required\":true},\n  {\"name\":\"time\",\"type\":\"select\",\"label\":\"Tid\",\"options\":[\"09:00\",\"10:00\",\"14:00\"]},\n  {\"name\":\"notes\",\"type\":\"textarea\",\"label\":\"Kommentar\"}\n]}\n```\n\n## Pitfalls\n\n- Do not hand-write a form, token, honeypot, or submit script.\n- Do not put a raw `<script>` in page HTML; the sanitizer removes it.\n- Do not expose action configuration through agent surfaces.\n- Do not send every submitted field to a webhook by default; choose the\n  smallest allowlist the external register needs.\n- `submit_token` is stable until the platform rotates its form-signing secret;\n  a rebuild refreshes it after rotation.\n",
     "tr-header-footer": "---\nname: tr-header-footer\ndescription: Vetted, robust header and footer presets to drop into the header/footer partials. Use when building or restyling a site's site-wide header or footer — start from a preset and restyle it instead of hand-rolling layout + overflow (the usual source of clipped logos and broken mobile menus).\n---\n\n# Header & footer presets\n\n> **The buffer model (draft writes).** Every content write in this recipe\n> (pages, blocks, partials, collection items) lands in an unsaved per-doc\n> DRAFT — deploys and plain previews only see SAVED content. For recipe-style\n> build work, pass `save: true` on write calls (the work is pre-approved by\n> the task itself), or run `commit_working_copy` per doc before any\n> `trigger_deploy`. Preview your drafts with `include_working_copy: true`.\n\n\nHeaders and footers are the two partials every page shows, and hand-rolling them\nis where logos get clipped and mobile menus break. **Start from a preset below,\nfill the placeholders, restyle with the site's colours — don't build the layout\nfrom scratch.** Each preset is deliberately robust; the \"why\" notes call out the\ntraps it avoids.\n\n## How to use\n\n1. `read_site_settings` — grab `logo`, `site_name`, `tagline`, `contact.email`,\n   and the colour palette.\n2. `read_partial partial_id=\"header\"` (and `footer`) — see what's already there;\n   don't blow away a working one without reason.\n3. Pick a preset, replace every `{{PLACEHOLDER}}`, adjust colours to the palette\n   (the presets already read `--color-*` / `--font-heading` with fallbacks).\n4. `update_partial partial_id=\"header\" patch={ html_content: \"…\" } version=\"…\"`.\n5. **Preview and self-review in context** (see `tr-redesign-branch` step 6):\n   the logo must be FULLY VISIBLE (not clipped), legible against its background,\n   and the mobile layout must work at 390px. Screenshot the header region in\n   context — never the logo element in isolation (that hides clipping).\n\nPlaceholders: `{{SITE_NAME}}`, `{{LOGO_URL}}`, `{{TAGLINE}}`, `{{EMAIL}}`, `{{YEAR}}`.\n\n---\n\n## Header A — Centered logo (minimal; landing pages)\n\n```html\n<header class=\"tr-hdr tr-hdr--center\">\n  <a class=\"tr-hdr-logo\" href=\"/\" aria-label=\"{{SITE_NAME}} — till startsidan\">\n    <img src=\"{{LOGO_URL}}\" alt=\"{{SITE_NAME}}\" />\n  </a>\n</header>\n<style>\n.tr-hdr--center{background:var(--color-surface,#fff);display:flex;justify-content:center;padding:clamp(1rem,2.5vw,1.6rem) 1.5rem}\n.tr-hdr-logo{display:inline-block;line-height:0;transition:transform .15s ease}\n.tr-hdr-logo:hover{transform:translateY(-1px)}\n.tr-hdr-logo img{height:clamp(40px,6vw,58px);width:auto;display:block}\n</style>\n```\n\n**Why it's robust:** no `overflow:hidden` anywhere near the logo (the #1 cause of a\nclipped wordmark); the logo sizes by `height` with `width:auto` so it never\ndistorts and never gets cropped; symmetric padding so it can't collide with the\nsection below. If you want a tinted header, set a solid `background` — don't add a\nglow that has to be clipped.\n\n## Header B — Logo left + links right (no-JS responsive menu)\n\n```html\n<header class=\"tr-hdr tr-hdr--nav\">\n  <div class=\"tr-hdr-inner\">\n    <a class=\"tr-hdr-logo\" href=\"/\" aria-label=\"{{SITE_NAME}} — till startsidan\">\n      <img src=\"{{LOGO_URL}}\" alt=\"{{SITE_NAME}}\" />\n    </a>\n    <input type=\"checkbox\" id=\"tr-nav-toggle\" class=\"tr-nav-toggle\" aria-hidden=\"true\" />\n    <label for=\"tr-nav-toggle\" class=\"tr-nav-burger\" aria-label=\"Meny\"><span></span><span></span><span></span></label>\n    <nav class=\"tr-hdr-nav\" aria-label=\"Huvudmeny\">\n      <a href=\"/\">Start</a>\n      <a href=\"#\">Sidan ett</a>\n      <a href=\"#\">Sidan två</a>\n      <a class=\"tr-hdr-cta\" href=\"#kontakt\">Kontakta oss</a>\n    </nav>\n  </div>\n</header>\n<style>\n.tr-hdr--nav{background:var(--color-surface,#fff);border-bottom:1px solid rgba(0,0,0,.06)}\n.tr-hdr-inner{max-width:1160px;margin:0 auto;padding:.9rem 1.5rem;display:flex;align-items:center;justify-content:space-between;gap:1rem;flex-wrap:wrap}\n.tr-hdr-logo{line-height:0}\n.tr-hdr-logo img{height:clamp(36px,4.6vw,50px);width:auto;display:block}\n.tr-hdr-nav{display:flex;align-items:center;gap:clamp(1rem,2.4vw,2rem);font-family:var(--font-heading),sans-serif;font-weight:600}\n.tr-hdr-nav a{color:var(--color-text,#1a1a1a);text-decoration:none}\n.tr-hdr-nav a:hover{color:var(--color-primary,#1F4FB8)}\n.tr-hdr-cta{background:var(--color-primary,#1F4FB8);color:var(--color-primary-fg,#fff);padding:.6rem 1.2rem;border-radius:999px}\n.tr-hdr-cta:hover{filter:brightness(1.05);color:var(--color-primary-fg,#fff)}\n.tr-nav-toggle{display:none}\n.tr-nav-burger{display:none;flex-direction:column;gap:5px;cursor:pointer;padding:.4rem}\n.tr-nav-burger span{width:24px;height:2px;background:var(--color-text,#1a1a1a);border-radius:2px}\n@media(max-width:760px){\n  .tr-nav-burger{display:flex}\n  .tr-hdr-nav{flex-basis:100%;flex-direction:column;align-items:stretch;gap:.2rem;max-height:0;overflow:hidden;transition:max-height .25s ease}\n  .tr-hdr-nav a{padding:.7rem .2rem}\n  .tr-nav-toggle:checked ~ .tr-hdr-nav{max-height:60vh}\n}\n</style>\n```\n\n**Why it's robust:** the mobile menu is a pure-CSS checkbox toggle — no JS to break,\nno library. The `overflow:hidden` is ONLY on the collapsing nav list (never on the\nheader or the logo), so the logo is always fully visible. Links use site colour\nvariables so it matches the brand automatically. The header wraps (`flex-wrap`) so\nnothing overflows the viewport on narrow screens.\n\n---\n\n## Footer A — Centered minimal\n\n```html\n<footer class=\"tr-ftr tr-ftr--center\">\n  <div class=\"tr-ftr-inner\">\n    <div class=\"tr-ftr-brand\">{{SITE_NAME}}</div>\n    <p class=\"tr-ftr-tag\">{{TAGLINE}}</p>\n    <p class=\"tr-ftr-contact\"><a href=\"mailto:{{EMAIL}}\">{{EMAIL}}</a></p>\n    <p class=\"tr-ftr-copy\">© {{YEAR}} {{SITE_NAME}}</p>\n  </div>\n</footer>\n<style>\n.tr-ftr--center{background:var(--color-primary,#163C8C);color:rgba(255,255,255,.78)}\n.tr-ftr--center .tr-ftr-inner{max-width:1120px;margin:0 auto;padding:2.6rem 1.5rem;text-align:center;display:grid;gap:.45rem}\n.tr-ftr-brand{font-family:var(--font-heading),sans-serif;font-weight:800;font-size:1.35rem;color:#fff}\n.tr-ftr-tag{margin:0;font-size:1rem;color:rgba(255,255,255,.85)}\n.tr-ftr-contact{margin:.15rem 0 0}\n.tr-ftr-contact a{color:#fff;text-decoration:none;font-weight:600}\n.tr-ftr-contact a:hover{text-decoration:underline}\n.tr-ftr-copy{margin:.8rem 0 0;font-size:.85rem;color:rgba(255,255,255,.55)}\n</style>\n```\n\n## Footer B — Three columns (brand · links · contact)\n\n```html\n<footer class=\"tr-ftr tr-ftr--cols\">\n  <div class=\"tr-ftr-grid\">\n    <div class=\"tr-ftr-col\">\n      <div class=\"tr-ftr-brand\">{{SITE_NAME}}</div>\n      <p class=\"tr-ftr-tag\">{{TAGLINE}}</p>\n    </div>\n    <nav class=\"tr-ftr-col\" aria-label=\"Sidfot\">\n      <a href=\"/\">Start</a>\n      <a href=\"#\">Sidan ett</a>\n      <a href=\"#\">Sidan två</a>\n    </nav>\n    <div class=\"tr-ftr-col\">\n      <p class=\"tr-ftr-contact\"><a href=\"mailto:{{EMAIL}}\">{{EMAIL}}</a></p>\n    </div>\n  </div>\n  <p class=\"tr-ftr-copy\">© {{YEAR}} {{SITE_NAME}}</p>\n</footer>\n<style>\n.tr-ftr--cols{background:var(--color-primary,#163C8C);color:rgba(255,255,255,.78)}\n.tr-ftr--cols .tr-ftr-grid{max-width:1120px;margin:0 auto;padding:3rem 1.5rem 1.4rem;display:grid;grid-template-columns:1.4fr 1fr 1fr;gap:2rem}\n.tr-ftr--cols .tr-ftr-brand{font-family:var(--font-heading),sans-serif;font-weight:800;font-size:1.35rem;color:#fff;margin-bottom:.4rem}\n.tr-ftr--cols .tr-ftr-tag{margin:0;color:rgba(255,255,255,.8);max-width:34ch}\n.tr-ftr--cols .tr-ftr-col{display:grid;gap:.5rem;align-content:start}\n.tr-ftr--cols nav a{color:rgba(255,255,255,.85);text-decoration:none}\n.tr-ftr--cols nav a:hover{color:#fff;text-decoration:underline}\n.tr-ftr-contact a{color:#fff;text-decoration:none;font-weight:600}\n.tr-ftr--cols .tr-ftr-copy{max-width:1120px;margin:0 auto;padding:0 1.5rem 2.4rem;font-size:.85rem;color:rgba(255,255,255,.55)}\n@media(max-width:680px){.tr-ftr--cols .tr-ftr-grid{grid-template-columns:1fr;gap:1.4rem}}\n</style>\n```\n\n**Why these footers are robust:** the columns collapse to one at 680px (no\nhorizontal scroll); all colours come from `--color-*` with fallbacks; the contact\nis a real `mailto:` link; nothing relies on fixed heights. Swap `--color-primary`\nfor a custom dark if the brand's primary is too light for white text.\n\n---\n\n## Restyling notes\n\n- The logo always comes from `read_site_settings → logo`. If it's `null`, set it\n  first (upload + `update_site_settings`) — don't hard-code a path.\n- For a **shaped transition** from the header/footer into the page, don't build a\n  wave band by hand — that belongs to the adjacent `core/section` via its\n  `divider_top` / `divider_bottom` (see `tr-redesign-branch`).\n- Keep the brand mark + a way home. Even a dramatic redesign keeps the logo\n  linking to `/`.\n",
     "tr-imagegen": "---\nname: tr-imagegen\ndescription: Use when the user wants to generate images for a Typeroll site with AI models (Gemini, OpenAI, Higgsfield) — hero images, illustrations, section backgrounds, og-images. Triggers on \"generera bilder\", \"generate images\", \"skapa en hero-bild\", \"AI-bilder\", \"bildgenerering\", or when a brief calls for imagery that doesn't exist in assets/. Covers the local lab loop (generate → review → pick) and uploading winners to the Typeroll media library.\n---\n\n# Generate images for a Typeroll site (local lab → media library)\n\nImage generation runs **locally** in the site workdir — provider API keys\nlive in the folder's `.env`, candidates land in `images/lab/`, and only\nthe picked winners are uploaded to the Typeroll media library via the\nregular media tools (see `tr-images` for the upload/variants half).\n\nWhy local: you can look at the candidates (Read the files), iterate on\nprompts cheaply, and never ship a key or a reject anywhere.\n\n## Folder convention\n\n```\n<site>/\n├── .env                  # provider keys — GITIGNORED, never committed\n├── prompts/\n│   └── image-style.md    # the site's image style profile (see below)\n└── images/\n    └── lab/              # generated candidates — gitignored, disposable\n```\n\n`.env` keys (only the ones the user has — check before assuming):\n\n```\nGEMINI_API_KEY=...\nOPENAI_API_KEY=...\n```\n\n(Higgsfield needs no key here — it connects as an MCP server, see below.)\n\nLoad them per-command (`source .env` doesn't persist between Bash calls):\n\n```bash\nexport $(grep -v '^#' .env | xargs)   # prepend to each generation command\n```\n\n## The style profile — prompts/image-style.md\n\nEvery site gets ONE style profile that you **prepend to every image\nprompt**. This is what keeps 20 images generated across 5 sessions\nlooking like one site. Derive it from `assets/brand.md` + the brief if\nit doesn't exist yet, and confirm it with the user before generating at\nscale. Keep it short (5–10 lines): art direction, palette, mood,\nphotography vs illustration, what to avoid.\n\nExample shape:\n\n```markdown\n# Bildstil — <Sajtnamn>\nVarm, folklig illustration med mjuka rundade former. Platt 2D med\nsubtila skuggor — ingen 3D, ingen fotorealism. Palett: kobolt #1F4FB8,\nsol #FFC83D, grädde #FFF8EC; accenter sparsamt. Människor: enkla,\ninkluderande, glada — inga karikatyrer. Undvik: stockfoto-känsla, text\ni bilden, logotyper, watermarks.\n```\n\n## Generate candidates\n\nName candidates descriptively: `images/lab/<motiv>-<modell>-<n>.png`.\nGenerate 3–6 candidates per slot (mix models when several keys exist),\nthen **Read the files to actually look at them** before showing the\nuser your shortlist.\n\n### Gemini (gemini-2.5-flash-image)\n\n```bash\ncurl -s \"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent\" \\\n  -H \"x-goog-api-key: $GEMINI_API_KEY\" -H 'Content-Type: application/json' \\\n  -d '{\n    \"contents\": [{\"parts\": [{\"text\": \"<STYLE PROFILE>\\n\\n<MOTIF PROMPT>\"}]}],\n    \"generationConfig\": {\"imageConfig\": {\"aspectRatio\": \"16:9\"}}\n  }' | jq -r '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data' \\\n  | base64 -d > images/lab/hero-gemini-1.png\n```\n\nAspect ratios: `1:1`, `16:9`, `4:3`, `3:4`, `9:16`. Gemini also does\nimage *editing* — pass an existing image as an `inlineData` part plus an\ninstruction to restyle/extend it (useful for \"same illustration but\nwinter\").\n\n### OpenAI (gpt-image-1)\n\n```bash\ncurl -s https://api.openai.com/v1/images/generations \\\n  -H \"Authorization: Bearer $OPENAI_API_KEY\" -H 'Content-Type: application/json' \\\n  -d '{\n    \"model\": \"gpt-image-1\",\n    \"prompt\": \"<STYLE PROFILE>\\n\\n<MOTIF PROMPT>\",\n    \"size\": \"1536x1024\",\n    \"quality\": \"high\"\n  }' | jq -r '.data[0].b64_json' | base64 -d > images/lab/hero-openai-1.png\n```\n\nSizes: `1024x1024`, `1536x1024` (landscape), `1024x1536` (portrait).\n\n### Higgsfield (MCP server — no API key)\n\nHiggsfield exposes its models through a hosted MCP server at\n`https://mcp.higgsfield.ai/mcp` (OAuth-protected — first connect opens a\nbrowser login; no secret lands in any file). Add it next to the\ntyperoll server in the site folder's `.mcp.json`:\n\n```json\n\"higgsfield\": { \"type\": \"http\", \"url\": \"https://mcp.higgsfield.ai/mcp\" }\n```\n\nThen use its tools directly — list what's available rather than\nassuming tool names. Save/download outputs into `images/lab/` with the\nsame naming convention, and prepend the style profile to prompts here\ntoo. **Headless caveat:** OAuth-protected MCP servers need an existing\nlogin session on the machine — connect once interactively before\nrelying on it in a headless run. (The same URL works as a custom\nconnector in Claude Desktop, for editors who don't use Claude Code.)\n\n## Review → pick → upload\n\n1. **Look at every candidate** (Read the image files) and write one line\n   per candidate in `build-log.md` (keep/reject + why).\n2. Show the user the shortlist (file paths) and let them pick — unless\n   they've delegated the pick to you.\n3. Upload winners with the regular media tools (see `tr-images`):\n   - small files → `upload_media_inline` (base64);\n   - larger → `create_upload_url` + HTTP PUT + `finalize_media`.\n   Give real `alt` text and a descriptive filename at upload time.\n4. `generate_image_variants` for responsive sizes when the image is\n   placed full-width.\n5. Place via `update_block` (`core/image`, hero fields, …) or page HTML.\n6. `images/lab/` is disposable — leave rejects there; never upload them.\n\n## Pitfalls\n\n- **Never put text in generated images** (headlines, buttons) — text is\n  HTML's job; generated text renders as gibberish in non-English. And\n  models sneak text onto surfaces where it \"feels natural\" — jerseys,\n  signs, banners, packages — even when the prompt doesn't ask for any.\n  Forbid it explicitly in the prompt (\"plain unmarked clothing, no\n  text, no letters, no numbers anywhere in the image\") and make the\n  same rule part of every site's style profile.\n- **Don't commit `.env` or `images/lab/`** — both are gitignored by the\n  kit convention; keep it that way.\n- **Cost discipline:** generation costs real money per image. Batch\n  thoughtfully (3–6 per slot, not 20), and reuse via Gemini's\n  edit-an-image mode instead of regenerating from scratch.\n- **Licensing/provenance:** AI-generated imagery is fine for site\n  decoration, but never generate fake \"photos\" of real people, products\n  the customer doesn't sell, or anything presented as documentary fact.\n- **The style profile is the contract.** If the user rejects a batch on\n  style grounds, fix `prompts/image-style.md` first, then regenerate —\n  don't just tweak one prompt.\n",
     "tr-images": "---\nname: tr-images\ndescription: Use when the user asks for an image, hero, illustration, or logo to be created and embedded in a Typeroll page. Covers the two-step signed-URL upload flow so the agent doesn't try to POST bytes through the CMS API (it can't).\n---\n\n# Add images to a Typeroll site\n\nThe Typeroll API does NOT accept image bytes directly. Uploads go\nthrough a signed PUT URL straight to Cloudflare R2, and the API only\nsees the metadata. Two-step flow:\n\n## Recipe\n\n### 1. Get an image\n\nOptions, in order of preference:\n\na. **Reuse an existing one.** `list_media` returns CDN URLs for every\n   image already on this site. Search the list before generating\n   anything — saves bandwidth and keeps the visual catalog tight.\n\nb. **Generate locally.** The user's Claude Code installation has\n   access to whatever image-gen tools they've configured (DALL-E,\n   Midjourney, Stable Diffusion, Replicate, etc.). Generate and save\n   to a tempfile.\n\nc. **Source from the web** with appropriate licensing (the user is\n   responsible for clearing rights). Save locally before upload.\n\n### 2. Mint a signed upload URL\n\n```\ncreate_upload_url filename=\"hero-services.png\"\n                  content_type=\"image/png\"\n                  size=<bytes>\n                  alt_text=\"Office worker reviewing documents at a desk\"\n```\n\nReturns:\n\n```json\n{\n  \"upload_url\": \"https://...r2.cloudflarestorage.com/.../signed-...\",\n  \"cdn_url\": \"https://cdn.example.com/orgs/.../images/...png\",\n  \"key\": \"orgs/.../images/...png\",\n  \"media_id\": \"abc123\",\n  \"expires_in\": 300\n}\n```\n\nThe signed URL is valid for 5 minutes. The media doc is already\nregistered — even before the upload completes — so it'll show in\n`list_media` immediately.\n\n### 3. PUT the bytes\n\nOutside the MCP, hit the signed URL directly:\n\n```\nPUT <upload_url>\nContent-Type: <same content_type as in step 2>\nBody: <file bytes>\n```\n\nIn a shell:\n\n```bash\ncurl -sS -X PUT --data-binary @hero-services.png \"$UPLOAD_URL\"\n```\n\nOptionally with `-H \"Content-Type: image/png\"` if you need to override what R2 will infer. **Nothing else.**\n\n> The signed URL embeds checksum-related query parameters\n> (`x-amz-checksum-crc32`, `x-amz-sdk-checksum-algorithm`) for legacy\n> SDK compatibility, but `X-Amz-SignedHeaders=host` — only the host\n> header is part of the signature. Sending the `x-amz-*` values as\n> request headers yields `403 SignatureDoesNotMatch`. Don't. Just\n> `--data-binary` the file at the URL.\n\nOr from JS (Claude Code can run a one-line script):\n\n```js\nawait fetch(uploadUrl, {\n  method: 'PUT',\n  headers: { 'Content-Type': contentType },\n  body: await fs.readFile(path),\n});\n```\n\nParallelise N uploads via shell `&` + `wait`:\n\n```bash\nwhile IFS=$'\\t' read -r filename signed_url; do\n  curl -sS -X PUT --data-binary @\"$filename\" \"$signed_url\" &\ndone < manifest.tsv\nwait\n```\n\nA 200 OK from R2 means the image is now live at `cdn_url`.\n\n### 4. Patch metadata (alt text, etc.)\n\nYou set `alt_text` at create time, but if you generate the image first\nand only THEN realize what to caption it as, patch later:\n\n```\nupdate_media media_id=<id> alt_text=\"...\" filename=\"hero-services-v2.png\"\n```\n\n### 4b. Fill missing alt-text on existing media\n\nWhen a customer has uploaded a bunch of images without alt-text (very\ncommon after a WP migration), don't make it up — use vision:\n\n```\nlist_media                                 → find items with empty alt_text\nsuggest_alt_text_context media_id=<id>     → returns { image_url, suggested_prompt,\n                                              language, used_on_pages, current_alt_text }\n# Pass image_url + the returned suggested_prompt to YOUR OWN vision\n# capability (you can fetch the URL and pass bytes to vision).\nupdate_media media_id=<id> alt_text=\"<what vision returned>\"\n```\n\nThe prompt is tuned for SEO-grade output: short (5-15 words), no \"image\nof / picture of\" filler, written in the site's content language,\ndecorative images return empty string. Run it sequentially on a\nlist_media batch and you can fix alt-text gaps across a whole site\nwithout burning your context on prompt design. The platform does NOT\nrun vision on your behalf — your model does, your usage.\n\n### 5. Embed in a page\n\n`read_page` the target, insert `<img>` in the right spot:\n\n```html\n<img src=\"<cdn_url>\"\n     alt=\"<alt_text>\"\n     style=\"width: 100%; height: auto; display: block; margin: 2rem 0;\" />\n```\n\nThen `update_page` with the new HTML. Or, if you're generating a hero\nfor a brand-new page, include the `<img>` directly in `create_page`'s\n`html_content`.\n\n## Pitfalls\n\n- **Always set `alt_text`.** Empty alt is bad for SEO + accessibility.\n  Default to a one-sentence description of what's in the image.\n- **`<script>` etc. in SVGs.** The page sanitizer drops `<script>`\n  inside SVG, so an icon set that includes script-based animations\n  won't render correctly. Use static SVG or a JPG/PNG export.\n- **CSS background-image references aren't dedup'd.** If you set the\n  same image as a CSS background on multiple pages, the alt-text +\n  metadata are page-irrelevant. The sanitizer allows\n  `background-image: url(...)` in inline styles, but think about\n  whether an `<img>` is actually better.\n- **Source URL leakage.** If you generated the image from a prompt\n  that contains internal info, don't bake that prompt into the\n  filename. Use a descriptive but generic filename.\n\n## Format choice\n\n- **PNG** for logos, icons with hard edges, anything with text.\n- **JPG** for photos. Smaller file, better for big hero images.\n- **WebP** if the target audience runs modern browsers (95%+ in 2026).\n- **SVG** for icons + simple illustrations. Vector scales perfectly.\n- **PDF** is supported by `create_upload_url` for document downloads;\n  link with `<a href>`, not `<img>`.\n",
@@ -25,6 +25,6 @@ export const BUNDLED_SKILLS = {
     "tr-seo": "---\nname: tr-seo\ndescription: Use when the user asks to improve SEO, fix meta tags, add structured data, check page titles, or audit the site's search visibility. Triggers on \"SEO\", \"meta descriptions\", \"Google ranking\", \"structured data\", \"JSON-LD\", \"sitemap\", \"sökoptimering\", or \"hjälp mig synas på Google\".\n---\n\n# SEO audit and improvements for a Typeroll site\n\n> **The buffer model (draft writes).** Every content write in this recipe\n> (pages, blocks, partials, collection items) lands in an unsaved per-doc\n> DRAFT — deploys and plain previews only see SAVED content. For recipe-style\n> build work, pass `save: true` on write calls (the work is pre-approved by\n> the task itself), or run `commit_working_copy` per doc before any\n> `trigger_deploy`. Preview your drafts with `include_working_copy: true`.\n\n\n## What Typeroll handles automatically\n\n- `<html lang>` from site `language` setting (per-page override via `language` field)\n- `<title>` = `page.seo_title || page.title + settings.default_seo_suffix`\n- `<meta name=\"description\">` from `page.seo_description`\n- `<meta name=\"robots\">` from `page.noindex`\n- `<meta property=\"og:*\">` Open Graph tags from seo_title, seo_description, og_image\n- `<link rel=\"canonical\">` from `page.canonical_url` (falls back to the page's own URL)\n- Article schema from `kind: \"article\"` + `author` + `date_published`\n- Page schema from `kind: \"page\"` (default)\n- `robots.txt` from `settings.robots_txt`\n- Sanitized HTML that preserves semantic structure\n\n## Recipe\n\n### 1. Audit current state\n\n```\nlist_pages status=\"all\"\nread_site_settings\n```\n\nFor each page, check:\n- Is `seo_title` set? (if not, Google uses `title` + suffix — often fine)\n- Is `seo_description` set? (150–160 chars, unique per page, includes keywords)\n- Is `og_image` set for the homepage and key landing pages?\n- Does the page have exactly one `<h1>`?\n\n### 2. Fix missing meta descriptions\n\n```\nbatch_update_pages updates=[\n  {page_id: \"home\",     patch: {seo_description: \"Acme designar rum...\"}},\n  {page_id: \"om-oss\",   patch: {seo_description: \"Vi är ett...\"}},\n  {page_id: \"tjanster\", patch: {seo_description: \"Våra tjänster...\"}}\n]\n```\n\nGuidelines:\n- 150–160 characters\n- Include the most important keyword naturally\n- Make it a compelling reason to click, not a summary of the page's nav\n\n### 3. Fix page titles\n\nSEO title = what Google shows in search results.\n\nIf `settings.default_seo_suffix` is set (e.g. \" — Acme Studio\"), every\npage whose `seo_title` is empty will show `title + suffix`. That's usually\nfine for inner pages; set an explicit `seo_title` only when you want\nsomething different.\n\n```\nupdate_site_settings {\"default_seo_suffix\": \" — Acme Studio\"}\n\nupdate_page page_id=\"home\" patch={\n  \"seo_title\": \"Acme Studio — Inredningsdesign i Stockholm\"\n}\n```\n\n### 4. Add Open Graph images\n\nSet `og_image` on pages that get shared on social media. If the site has\na branded hero image, upload it:\n\n```\nupload_media_from_url url=\"https://...\" alt=\"Acme Studio — Inredningsdesign\"\n# → returns cdn_url\n\nbatch_update_pages updates=[\n  {page_id: \"home\",     patch: {og_image: \"<cdn_url>\"}},\n  {page_id: \"om-oss\",   patch: {og_image: \"<cdn_url>\"}}\n]\n```\n\nOG image dimensions: 1200×630px ideal. The platform doesn't resize —\nuse a correctly-sized source image.\n\n### 5. Add structured data (JSON-LD)\n\nTyperoll auto-generates Article and Page schema, but you can override or\nextend with custom JSON-LD per page. Example: LocalBusiness on the homepage.\n\n```\nupdate_page page_id=\"home\" patch={\n  \"json_ld\": \"{\\\"@context\\\":\\\"https://schema.org\\\",\\\"@type\\\":\\\"LocalBusiness\\\",\\\"name\\\":\\\"Acme Studio\\\",\\\"url\\\":\\\"https://acme.se\\\",\\\"telephone\\\":\\\"+46812345\\\",\\\"address\\\":{\\\"@type\\\":\\\"PostalAddress\\\",\\\"streetAddress\\\":\\\"Drottninggatan 1\\\",\\\"addressLocality\\\":\\\"Stockholm\\\",\\\"postalCode\\\":\\\"111 51\\\",\\\"addressCountry\\\":\\\"SE\\\"}}\"\n}\n```\n\n**Important:** JSON-LD goes in the `json_ld` field as a JSON *string*\n(not a nested object). The renderer injects it inside\n`<script type=\"application/ld+json\">`.\n\nCommon schemas worth adding:\n- Homepage: `LocalBusiness` or `Organization`\n- About: `AboutPage`\n- Contact: `ContactPage`\n- Blog articles: auto-generated from `kind:\"article\"` + `author`\n- Events: `Event` with `startDate`, `location`\n- Products: `Product` with `offers`\n\n### 6. robots.txt\n\nThe default robots.txt allows all crawlers. Update if needed:\n\n```\nupdate_site_settings {\n  \"robots_txt\": \"User-agent: *\\nAllow: /\\nSitemap: https://acme.se/sitemap.xml\"\n}\n```\n\nTyperoll doesn't generate a sitemap automatically in phase 1. If the\ncustomer needs one, create a `/sitemap` page with HTML that lists all\npublished pages, or write a static `sitemap.xml` as a page with\n`slug: \"sitemap.xml\"` and HTML-encoded XML (not recommended for large sites).\n\n### 7. Canonical URLs\n\nSet `canonical_url` when a page has a duplicate (e.g. the same content\naccessible via two slugs after a migration):\n\n```\nupdate_page page_id=\"tjansterna\" patch={\n  \"canonical_url\": \"https://acme.se/tjanster\",\n  \"noindex\": true\n}\n```\n\n### 8. Language settings\n\n```\nupdate_site_settings {\"language\": \"sv\"}\n```\n\nPer-page override for multilingual content:\n```\nupdate_page page_id=\"about-en\" patch={\"language\": \"en\"}\n```\n\n### 9. Heading audit\n\nUse `search_pages` to find structural problems:\n\n```\nsearch_pages contains=\"<h1\"      # pages that have at least one H1\n```\n\nThen `read_page` on pages that seem to have none or multiple. Fix via\n`update_page patch={html_content: \"<corrected HTML>\"}`.\n\n### 10. Deploy\n\n```\ntrigger_deploy\nget_deploy_status job_id=<id>\n```\n\n## Pitfalls\n\n- **Don't stuff keywords.** Write descriptions for humans. Google ignores\n  `<meta name=\"keywords\">` (not a field in Typeroll anyway).\n- **JSON-LD is a string, not a nested field.** Pass the entire schema as\n  a JSON-encoded string in `json_ld`. The server escapes `</script` before\n  injection.\n- **OG images need absolute URLs.** The `cdn.typeroll.com` URLs are always\n  absolute — use those.\n- **`canonical_url` + `noindex` together.** If you noindex a page AND set\n  canonical, the canonical is redundant (noindexed pages don't pass equity).\n  Use one or the other.\n- **Default suffix on homepage looks odd.** \"Acme Studio — Acme Studio\"\n  happens when title=\"Acme Studio\" and suffix=\" — Acme Studio\". Set an\n  explicit `seo_title` for the homepage.\n"
 };
 export const BUNDLED_DOCS = {
-    "agents": "# AGENTS.md — Working on a Typeroll site\n\nYou are connected to a Typeroll site through `@typeroll/mcp-server`.\nThis file is your briefing: what the system is, what conventions matter,\nwhat tools to reach for first.\n\nIf anything below conflicts with what you observe in the tools, trust the\ntools — the platform may have moved since this was written.\n\n**Start here for site-shaped tasks.** When the user wants to build,\nmigrate, redesign, or brand a site, call `list_skills` first — the server\nadvertises its own step-by-step playbook (`tr-new-site`, `tr-migrate-wp`,\n`tr-brand`, …). Then `read_skill name=…` loads the full recipe. These are\nlocal reads; no API key or site context required.\n\n**Branch first for anything larger than a small edit.** Before a redesign,\na multi-page change, or trying out a new design direction, run\n`create_branch name=\"…\"` and pass the returned id as `version=<id>` on every\nsubsequent read/write. The work stays off the live `main` version until you\n`merge_branch` it — nothing ships until you decide it should. Branches default\n`robots_blocked:true` and get their own deploy URL for stakeholder review.\nIt's the cheapest insurance there is; when in doubt, branch. The\n`tr-redesign-branch` skill walks the whole flow. (Small, low-risk single edits\ncan go straight to main.)\n\n## What this is\n\nTyperoll is a static-site CMS: content lives in a database, the user\nedits it through an in-app editor, and a deploy step compiles everything\nto a fast static site hosted on Cloudflare Pages. The in-app chat handles\nsingle-page or single-block edits by the editor audience. You — through\nthis MCP — handle the work that doesn't fit there: site-wide redesigns,\nbulk content updates, structural migrations, directory imports.\n\nThe MCP server is a thin wrapper around the public REST API. Each tool\nmaps to one HTTP endpoint; the actual logic runs in the customer's portal\n(SaaS or self-hosted).\n\n## The data model in 90 seconds\n\n- **Pages.** Title, slug, status (`draft | review | unlisted | published`),\n  body content + SEO fields. Two body shapes selectable per page via\n  `content_mode`:\n  - `blocks` (DEFAULT for new pages) — `blocks: Block[]` tree of typed\n    blocks (heading, prose, section, columns, image, button, plus any\n    user/third-party block types installed on the site). Use the\n    block-mutation tools (`add_block`, `update_block`, `move_block`,\n    `remove_block`) for structural changes.\n  - `html` — body lives in `html_content` as a single HTML string.\n    Useful when you have hand-written markup to drop in directly.\n\n  Slug is a single path segment — no slashes. `about` → `/about`,\n  `kontakt` → `/kontakt`, empty string `\"\"` → homepage. The v1 API\n  rejects `services/design` and other slash-containing slugs with\n  \"Invalid slug … slugs must not contain slashes.\" For nested URLs\n  like `/blog/{slug}` or `/services/{slug}`, the right primitive is a\n  **collection with `route_template`** (see the `tr-blog` and\n  `tr-directory` skills) — not a flat page with a slashed slug.\n\n- **Partials = global blocks.** Three kinds:\n  - `header` — auto-injected at the top of every page.\n  - `footer` — auto-injected at the bottom of every page.\n  - `free` — reusable HTML you drop into a page with\n    `<x-include name=\"block-id\" />`. Free blocks are how you avoid\n    duplicating HTML across HTML-mode pages.\n\n  Partials themselves also support `content_mode='blocks'` — pass a\n  `blocks: Block[]` tree to `update_partial` and the renderer composes\n  it the same way as a page. Useful for header/footer authored with\n  block types.\n\n- **Collections.** Repeatable content types (blog, team, events,\n  products, restaurants for a directory site, etc.). Each has a schema\n  (`fields[]`) and optional **per-item routing** via `route_template`\n  (e.g. `/restaurants/{slug}`). When set, every published item gets its\n  own static URL rendered through `item_template_html`. Set\n  `route_template=\"\"` to opt out and keep the collection listing-only.\n\n- **Settings.** Site name, tagline, logo, favicon, colors, fonts,\n  contact info, social links, SEO suffix, default meta description\n  (`default_meta_description` — site-wide fallback for pages without a\n  `seo_description`; tagline is the last resort), plus `scripts_head`,\n  `scripts_body_end`, `custom_css` (writable via the API — your bearer\n  token authorises shipping arbitrary CSS/JS to the live site, just\n  like editing a partial's HTML does).\n\n- **Analytics attribution.** `read_funnel_attribution` and\n  `update_funnel_attribution` manage the Analytics module's validated,\n  site-level rules that forward\n  allowlisted query parameters to exact HTTPS link targets. Optional\n  first-/last-touch storage is consent-gated and off unless configured. Read\n  the current config before writing. For advertising pass-through, preserve\n  only incoming or stored values and omit `fallback`: a fallback creates\n  synthetic attribution when no campaign value exists. The API rejects such\n  fallbacks unless `allow_synthetic_fallbacks=true` explicitly acknowledges\n  that behavior. A target `click_event` is recorded as a consent-gated,\n  first-party conversion when Analytics is enabled; the event contains only\n  the declared funnel, destination, page path, and resolved allowlisted\n  attribution. Navigation never waits for analytics.\n\n- **Page templates.** A `PageTemplate` is a Block[] tree that wraps a\n  page's body. The template contains exactly one block of type\n  `template_content_slot` — at render time that block gets replaced by\n  the page's own `blocks`. Set `Page.template = \"<template-id>\"` to\n  apply a template to a page.\n\n- **Block types.** A site has three sources of block types:\n  - **Core** (origin: 'core', ids like `core/section`) — shipped in\n    the platform, always available.\n  - **User** (origin: 'user') — created in the portal's block-types UI.\n  - **Third-party** (origin: 'third_party') — imported from .tcblocks\n    packages via `import_block_types`.\n\n  `list_block_types` returns ALL of them in one list as a lightweight\n  summary: each entry's id, label, category, container/slot info, origin,\n  and full field schema (names, types, defaults) — but NOT the render-time\n  template/styles/script (omitted so the list stays within token budget as\n  the library grows). Use `read_block_type` for one block's markup, or pass\n  `full:true` to inline it for every block. Always call this FIRST before\n  working with blocks — never hardcode block ids or field names, the\n  available set is per-site.\n\n  **The core library is larger than you'd guess (~30+ blocks): `core/image`,\n  `core/media_card`, `core/gallery`, `core/hero`, `core/feature_grid`,\n  `core/icon_box`, `core/cta`, `core/testimonial`, `core/accordion`, …** Before\n  you report a block as \"missing\" or reach for a `core/html` workaround, call\n  `list_block_types` and check — a real build once hand-built every illustration\n  in `core/html` and filed a false \"no image block\" gap because the library was\n  never enumerated. Prefer a native block; `core/html` is the last resort.\n\n  Block-library specifics worth knowing (template_capabilities_version\n  0.15.0):\n  - **`core/media_card`** — image + text side by side (image left/right,\n    width third/two-fifths/half, heading + richtext + button, optional\n    card background/radius; stacks image-on-top below 720px). Use it for\n    the classic \"photo next to copy\" layout instead of hand-building\n    section+grid+html.\n  - **`core/search`** (0.29.0+) — site search over the deployed site.\n    Place the block anywhere; the deploy pipeline detects it, runs\n    Pagefind over the built HTML, and the block loads the search UI at\n    visit time. Index = page content only (nav/footer excluded); noindex\n    pages stay out. Editor preview shows a placeholder note (the index\n    only exists on the deployed site).\n  - **Archive pagination** (0.29.0+): a collection listing\n    (`core/collection_list` / `core/repeater`) with `paginate: N` renders\n    N items per page + a pager, and the build generates `/page/2/`… routes\n    automatically. `paginate` supersedes `limit`; one paginated listing\n    per page.\n  - **`core/feature_row`** (0.29.0+) — the full-width \"zig-zag\"\n    feature/step row: balanced halves that hug the center gutter (no\n    wide-screen dead air), natural-aspect image (never cover-cropped —\n    that's media_card's card look), eyebrow + heading + richtext +\n    button pair, `image_side: left|right` per row, `stack_order`\n    controls what comes first on mobile. Prefer it over `core/columns`\n    with an unbalanced ratio + width-capped text for these rows.\n  - **`core/hero` and `core/cta` render their buttons server-side** via\n    `primary_label`/`primary_url` + `secondary_label`/`secondary_url`.\n    (The old `buttons` array relied on client hydration that never\n    existed — if you see `data-buttons` in stored content it renders\n    nothing; rebuild with the explicit fields.)\n  - **`core/image` gets responsive `<picture>` automatically at build\n    time** — the deploy pipeline's SEO transform converts CDN `<img>`\n    into `<picture>` with AVIF/WebP srcset variants. You do NOT need\n    `core/html` for responsive images; just point `src` at an uploaded\n    media URL (run `generate_image_variants` first) and optionally set\n    `radius`. Note: the in-portal preview shows the plain `<img>` — the\n    `<picture>` upgrade appears on the deployed site.\n  - **Icons render inline SVG** (since template_capabilities_version\n    0.16.0). Every `type: 'icon'` schema field — on `core/icon`,\n    `core/icon_box`, `core/step_card`, and custom block types — renders\n    a stroke-based inline SVG when the value is a name from\n    `get_site_capabilities → core_icon_names` (a curated Lucide subset:\n    `check`, `star`, `shield-check`, `mail`, `arrow-right`, `zap`,\n    `truck`, `chart-line`, …). Any other value (emoji, plain text) is\n    rendered as escaped text, so emoji stand-ins keep working. Icons\n    size with `font-size` (the SVG is 1em) and paint with\n    `currentColor`. Custom block templates opt in by placing the derived\n    raw token `{{{<field>_svg}}}` where the icon should appear. On\n    pre-0.16.0 portals icons don't render — use emoji or CSS markers.\n    `core/tabs` label icons are the remaining gap (tab strip is built\n    client-side).\n  - **Grids with a partial last row: set `last_row: 'center'`** (since\n    template_capabilities_version 0.16.5). Five equal cards in a 3-col\n    `core/grid` (or 7 in 4, ...) left-align the orphans by default; with\n    `last_row: 'center'` the last row auto-centers. THE DESIGN RULE: when\n    N peer cards don't divide by the column count, center the last row or\n    change the column count — NEVER invent a \"wide\"/full-width variant of\n    one peer card just to fill the hole. Special treatment is a content\n    decision, not a layout patch.\n  - **`core/section` is natively full-bleed on block pages** (since\n    template_capabilities_version 0.14.0): the section's background runs\n    edge-to-edge and meets the header with zero gap; content inside is\n    constrained by the section's own inner container (`width` field:\n    narrow/normal/wide/full). Never use 100vw negative-margin hacks.\n    Top-level blocks that are NOT sections still get a classic centered\n    container as fallback. Anchor ids and custom classes via\n    `style_overrides` are safe on full-bleed sections since 0.15.3 —\n    they merge into the `<section>` element itself. On 0.14.x–0.15.2\n    they wrapped the section in a `<div>`, which silently disabled\n    full-bleed for that section.\n  - **Shaped section transitions** (since template_capabilities_version\n    0.24.0): `core/section` takes `divider_top` / `divider_bottom`\n    (`none | wave | curve | tilt`). The platform paints the divider in the\n    section's OWN `background` and overlaps the neighbour by 1px, so a\n    cream↔colour transition renders seam-free. **Use this for waves/curves —\n    never hand-roll a divider band in `core/html`** (a separate stacked shape\n    seams against the next section as a sub-pixel hairline in Chrome). Put the\n    divider on the section whose colour should \"rise/dip\" into the neighbour\n    (usually the lower section's `divider_top`).\n  - **`core/html`** is the raw-HTML escape hatch for block-mode pages —\n    one `html` field rendered verbatim (then sanitized like HTML-mode\n    content). Use it for the genuinely unique thing no block covers.\n    Prefer real blocks when one fits.\n  - **Structured records at scale** (template_capabilities_version ≥ 0.31.0).\n    Four things landed together for directory-shaped sites:\n    - `collection_completeness` — **start an enrichment pass here**, not by\n      paging every item. Returns per-field gap counts plus the N worst\n      records (missing fields, never-verified fields, fields whose last write\n      is older than the staleness window), computed at read time. Fields no\n      API key may write are excluded by default: a gap you can't close is\n      noise.\n    - **Per-field write authority.** A collection field can declare\n      `writable_by` (`portal | owner | agent | app | import`). A write you're\n      not permitted, or one that would overwrite a higher-precedence writer\n      (a human correction, the listed business's own edit), comes back as\n      **409 with the losing field names** — never a silent no-op. Treat that\n      as \"already handled\" and record it; retrying will lose again.\n    - **Item references.** `item_ref` / `item_ref_list` fields point at items\n      in another collection (`ref_collection`). The reverse direction is\n      computed at render time — don't try to maintain backlinks yourself.\n      Render them with a `core/repeater` whose `source_type` is `related`\n      (a ref field on the current item) or `backlinks` (who points at it).\n    - **Taxonomy pages.** `CollectionDef.facets` generates one page per\n      distinct field value. ⚠️ This turns record count into ROUTE count, and\n      route count is what the build timeout measures. `min_items` (default 2)\n      keeps thin-content pages out, and combination pages must be listed\n      explicitly in `facet_combinations` — never assume a cartesian product.\n  - **`core/embed`** (template_capabilities_version ≥ 0.30.0) is\n    `core/html` plus behaviour: an `html` field and a `js` field. Reach\n    for it when a one-off placement needs JavaScript. **A `<script>` tag\n    written into `core/html` — or into any page/block markup — is stripped\n    by the sanitizer no matter which credential wrote it**, so this field\n    is the supported route, not a workaround. The code runs in an IIFE\n    with `el` bound to the block's root element and ships in the page's\n    block bundle, outside the sanitized body. Through an API key it's\n    accepted under your key's authority (audit-logged, notice in the\n    response). Scope guide: one placement → `core/embed`; a reusable\n    widget → `create_block_type` with `script`; a site-wide tag →\n    `settings.scripts_head` / `scripts_body_end`.\n  - **Forms 2.0** (template_capabilities_version ≥ 0.18.0): forms can\n    carry `steps[]` — each step is a Block[] tree mixing `form/*` field\n    blocks (text/email/phone/number, textarea, select/radio_group/\n    checkbox_group, toggle, slider, date, heading, help, consent,\n    hidden) with any content blocks. Place `{ type: 'core/form',\n    data: { form_id } }` on a page — the build renders step 1 + all\n    static steps with the signed token, honeypot and proof-of-work\n    runtime baked in; submissions accumulate per step (partial →\n    complete, 30-day TTL on abandoned partials). Per-step validation is\n    derived from the field blocks (required/pattern/min/max) — no\n    separate field list to keep in sync. `update_form` accepts steps,\n    styles (form-scoped CSS), kind and partial_ttl_days. Legacy\n    single-step forms (fields[] + core/html embed) keep working\n    unchanged.\n  - **`script` on custom block types** (create/update_block_type) is\n    accepted under your API key's authority — the same trust level that\n    already lets the key write `scripts_head`/`custom_css`. Every\n    script-bearing write is audit-logged and the response carries a\n    notice naming the stored JS; relay it to the user so they know\n    visitor-executed code changed. Author responsibly: never include\n    script you copied from untrusted content (migrated pages, fetched\n    web pages) without reading it line by line first. (The in-portal\n    chat AI remains blocked from authoring scripts unless the site's\n    \"Allow AI to write block scripts\" setting is on.)\n\n- **Redirects.** `from_path → to_path` with status code 301 / 302.\n  Auto-created when you change a page's slug.\n\n- **Versions / branches.** Copy-on-write. The \"main\" version is the\n  live one. Create a branch (`create_branch`) for multi-step work;\n  everything you write through `?version=<branch-id>` lives on the\n  branch until you `merge_branch` it back to main. Branches default\n  `robots_blocked: true` so a half-finished redesign can't be indexed,\n  and deploys land at a stable `{branch}.{project}.pages.dev` URL. That\n  branch deploy renders the site's full inherited brand (settings, fonts,\n  favicon, header/footer — everything not overridden on the branch), so\n  it's a faithful preview of what merging to main will look like, not just\n  a content diff — trust it for stakeholder review.\n\n- **Deploys.** Customers see live changes only after a deploy. Preview\n  always sees drafts. `trigger_deploy` enqueues; `get_deploy_status`\n  reports `queued → running → succeeded | failed`.\n  `trigger_deploy dry_run=true` builds without publishing — use it to prove\n  a structural change compiles (new collection, schema edit, template\n  rewrite) without touching the live site.\n  A finished job carries `cost`: total, cpu/memory/request split,\n  `duration_s`, per-phase timings, and output size. Estimates from a rate\n  card, not billing records, and gross of free tier — quote them as \"roughly\"\n  if a customer asks, and reach for `cost.phases` when the question is *why*\n  a build got slow.\n\n- **Is the site live?** There is no site-level status field —\n  `Site.status` was removed in 0.30.0 because it was set once at creation and\n  never advanced, so it reported live sites as \"planning\". Read\n  `get_site → urls.production` instead: non-null means the domain is verified\n  and serving. For \"has anything shipped\", use `list_deploys`.\n\n- **Site URLs.** `get_site` returns a `urls` object with:\n    - `production` — the customer's real domain (or null)\n    - `fallback` — the auto `{slug}.typeroll.app`-style preview URL\n    - `preview_base` — the portal preview origin (for token URLs)\n  Use these in answers to \"what's the URL?\" — never invent.\n\n- **For design/content iteration, share the DB-LIVE preview — don't deploy.**\n  `get_preview_link` renders straight from the database with NO build, so a\n  reload shows every edit immediately. Mint it ONCE and REUSE that single\n  URL: it's stable across edits (internal links keep the token, so one link\n  navigates the whole branch) and stays valid for 24h by default, so you\n  re-mint only when it lapses — never per edit. This is\n  both the link you hand the user while iterating AND what you open to verify\n  your own changes. Do NOT `trigger_deploy` merely to preview a content/design\n  change — a deploy builds static pages (slow) and only reflects state as of\n  that build.\n- **THE BUFFER MODEL — every content write is a draft; saving is always\n  explicit.** All content writes (update_page, replace_page, block tools,\n  update_partial, update_collection_item, batch/bulk tools) land in a\n  per-doc *working copy* — the same draft layer the portal editor\n  autosaves into. Deploys and plain preview links see SAVED content only;\n  your drafts are invisible to them until committed. The loop:\n    1. Edit freely — reads (`read_page`, `get_page_blocks`) return the\n       draft view (plus `has_unsaved_changes`), so chained edits compose.\n    2. Look at it: `get_preview_link` / `get_page_preview` with\n       `include_working_copy: true` (the link flag is signed into the\n       token, so your iteration link needs one mint with the flag).\n    3. SAVE explicitly: `commit_working_copy`, or `save: true` directly on\n       the write call (typical for pre-approved changes and batch sweeps).\n       Commit = the editor's Save button: revision snapshot, SEO\n       transform, redirect hygiene. Rejected → `discard_working_copy`.\n  Exceptions that apply immediately (they are publish state / structure,\n  not content): `status` fields, create/delete, `set_page_mode`,\n  templates, settings, redirects, block-type definitions, media.\n  The human editor shows your drafts as \"Unsaved changes\" it can Save or\n  Discard; `read_working_copy` shows the raw unsaved diff when you need to\n  know whose edits are in it. Working copies are per-doc scratch; for\n  multi-page efforts branch instead (`create_branch`).\n  **Before `trigger_deploy`: commit.** Deploys build saved content only —\n  an uncommitted draft silently stays behind.\n- **Deploys / `{branch}.{project}.pages.dev` are the STATIC BUILD**, refreshed\n  only by `trigger_deploy`. Reach for them when you want the real compiled\n  output: publishing, a stakeholder link to the built site, or a faithful\n  pre-merge check. The branch alias is permanent across re-deploys; the\n  per-deploy `{hash}.pages.dev` is immutable per build. Reserve deploys for\n  these — not for previewing edits.\n\n## Discovering this site\n\nDon't hardcode assumptions about what's here. Every fact about the site\ngoes through the MCP:\n\n1. `get_site` — confirm the key works; learn the site name + URLs.\n2. `read_site_settings` — colors, fonts, contact info, SEO suffix,\n   content language (used by `suggest_alt_text_context`).\n3. `list_pages` — what pages exist, paginated.\n4. `list_partials` — what shared blocks already exist. **Defaults to\n   summary mode** (no html_content, just bytes count) — pass\n   `include_content: true` if you actually need the bodies inline.\n5. `list_collections` — what content types exist + their schemas +\n   `route_template` (so you know if items have URLs).\n6. `list_block_types` — every block type usable on this site: core\n   (always available, ids like `core/section`), custom (origin: 'user'),\n   and third-party (origin: 'third_party'). Each entry includes the\n   full schema so you know what `data.X` fields each block accepts.\n7. `list_page_templates` — PageTemplate docs that wrap pages.\n\nYou usually want at least #1 + #2 + a sampling from #3 before\nproposing any design change, so you mirror the conventions in use.\n\n**Source of truth = the live site (the API), by default.** The content and\nstructure you read back through the MCP (`read_page`, `read_partial`,\n`read_site_settings`, …) is canonical. Local files in the project folder —\n`sources/*.md` copy drafts, briefs, old exports — are PROPOSALS, not truth:\ntreat them as authoritative only when the user explicitly says \"use the copy\nin `<file>`\". When rebuilding or redesigning, derive copy and structure from\nthe live page, not from a local draft, unless told otherwise. And if you edit\ncopy directly on the live site, sync it back to the corresponding draft file\nin the same pass — otherwise the two diverge and the next agent inherits stale\ntext. (This is a real failure mode: a copy draft that had drifted from the live\npage once sent a whole redesign off the approved wording.)\n\n**Don't have a site yet?** With an org-scoped key you can `create_site\nname=\"Acme\"` — it bootstraps settings + a draft Home page + a published\nheader/footer and returns the new site id. Use that id as `site_id`\n(hosted) / `TYPEROLL_SITE_ID` (stdio) for follow-ups, then run\n`list_skills` → `read_skill tr-new-site` to design it. A site-scoped key\ncan't create sites (it's bound to one) and gets a 403.\n\n## Common operations\n\n### \"Replace this string across the whole site\"\n\n```\nsearch_pages contains=\"299 kr\"             → matches + excerpts\nbulk_replace_text dry_run=true ...          → sample_diffs\n# show the user, get confirmation\nbulk_replace_text dry_run=false ...         → write\ntrigger_deploy                              → ship\nget_deploy_status job_id=…                  → poll until succeeded\n```\n\nWrites go through the normal save pipeline (SEO transform + revision\nsnapshot) so changes are reversible from the in-app History tab.\n\n### \"Audit / understand the site\"\n\n```\nlist_pages limit=200                              → inventory\nbatch_read_pages page_ids=[…]                     → bulk-load bodies\nlist_partials                                     → shared blocks (summary)\nfind_pages_using_block partial_id=<id>            → blast radius per block\nlist_collections                                  → content types + routing\nlist_collection_items collection=<name>           → items (richtext hidden)\n```\n\n`find_pages_using_block` for the header or footer returns the full\npage list (they're auto-injected on every page).\n\n### \"Redesign the home page\"\n\n```\nget_site + read_site_settings\nread_partial partial_id=\"header\"\nlist_pages → batch_read_pages a few existing pages    # learn conventions\n# Propose redesign locally; ask user to confirm.\ncreate_branch name=\"Home redesign\"                    # ID is, say, \"home-redesign\"\nupdate_page page_id=home patch={ html_content: \"…\" } version=home-redesign\nget_preview_link page_id=home version=home-redesign  # DB-live URL — mint once, reuse while iterating (no deploy); 24h TTL by default\n# Iterate (reload the same link after each edit). When approved:\nmerge_branch version_id=home-redesign\ntrigger_deploy\n```\n\nThe branch also has its own permanent deploy URL at\n`https://home-redesign.<project>.pages.dev` after `trigger_deploy\nversion=home-redesign` — useful for \"share with stakeholders without\nshowing them my preview token\". `read_version version_id=home-redesign`\nreturns it as `deploy_url`.\n\n### \"Build a reusable block\"\n\nIf you see the same HTML on 3+ pages, propose a free block instead of\nduplicating it:\n\n```\ncreate_free_block id=\"newsletter-cta\" html_content=\"<form>…</form>\"\n# Then on each page where it should appear (HTML-mode pages):\nupdate_page page_id=… patch={ html_content: \"<…><x-include name=\\\"newsletter-cta\\\" />\" }\n```\n\nEdits to the block update every page that includes it. Use\n`find_pages_using_block` before changing it.\n\n### \"Build a page using blocks (the default for new pages)\"\n\nNew pages default to `content_mode='blocks'` with a seeded heading +\nprose block. Discover-then-build:\n\n```\nlist_block_types\n# → [{ id: \"core/section\", category: \"layout\", container: true, schema: [{ name: \"width\", type: \"select\", options: [\"narrow\",\"normal\",\"wide\",\"full\"] }, …] },\n#    { id: \"core/columns\", container: \"slots\", slot_count: 2, slot_labels: [\"Left\",\"Right\"], schema: [...] },\n#    { id: \"hero_bold\", origin: \"user\", schema: [...] },   ← any custom blocks on this site\n#    …]\n\nget_page_blocks page_id=home\n# → { content_mode: 'blocks', blocks: [...] }\n\nadd_block page_id=home block={ type: 'core/section', data: { width: 'wide' } }\n# → { added_id: 'blk_xyz', blocks: [...] }\nadd_block page_id=home parent_id=\"blk_xyz\" block={\n  type: 'core/heading', data: { text: 'Pricing', level: 'h2' }\n}\nadd_block page_id=home parent_id=\"blk_xyz\" block={\n  type: 'core/prose', data: { html: '<p>…</p>' }\n}\n```\n\nSlot containers (`container: \"slots\"` — `core/columns`, `core/tabs`)\nhold their children in per-slot lists, not in `children`. Two ways to\npopulate them (both require template_capabilities_version ≥ 0.15.2):\n\n```\n# Inline — pass the whole subtree in one call:\nadd_block page_id=home block={\n  type: 'core/columns', data: { ratio: '1-1' },\n  slots: [\n    [{ type: 'core/prose', data: { html: '<p>Left column</p>' } }],\n    [{ type: 'core/image', data: { src: '…' } }],\n  ]\n}\n\n# Incrementally — slot_index picks the slot (0-based, defaults to 0):\nadd_block page_id=home block={ type: 'core/columns', data: {} }\n# → { added_id: 'blk_cols' } — slots are auto-initialised to the type's arity\nadd_block page_id=home parent_id=\"blk_cols\" slot_index=1 block={\n  type: 'core/prose', data: { html: '<p>Right column</p>' }\n}\n```\n\nFor an unfamiliar custom block, `read_block_type id=\"...\"` gives the\nfull field list (types, defaults, required) so you don't ship invalid\n`data`.\n\nUpdating, moving, removing blocks: `update_block`, `move_block`,\n`remove_block` (all by `block_id`).\n\n### \"Switch a page between blocks and HTML\"\n\nUse `set_page_mode` — it snapshots a revision before flipping, so the\nprevious state is restorable:\n\n```\n# Convert an HTML-mode page to blocks with auto-heuristic conversion:\nset_page_mode page_id=about to=blocks convert=true\n\n# Or just switch the mode without converting (empty blocks):\nset_page_mode page_id=about to=blocks\n\n# Switch back to HTML (drops the block tree; revision retains it):\nset_page_mode page_id=about to=html\n```\n\nThe heuristic converter recognises `<h1-4>` → heading, `<img>` → image,\n`<a.btn>` → button, `grid-cols-2` → two-column, `<section>` / hero divs\n→ section. Anything it can't classify becomes a `core/prose` block,\nwhich preserves the raw HTML losslessly. Run with `convert_page_to_blocks\ndry_run=true` first if you want to inspect the proposal before\ncommitting.\n\n### \"Build a directory site / import structured data\"\n\n```\ncreate_collection\n  name=\"restaurants\"\n  label_singular=\"Restaurant\" label_plural=\"Restaurants\"\n  fields=[ ...title, slug, address, phone, cuisine, body... ]\n  route_template=\"/restaurants/{slug}\"\n  item_template_html=\"<article><h1>{{title}}</h1>… {{{body}}}</article>\"\n\n# For each row in your source data:\ncreate_collection_item collection=\"restaurants\" fields={…} status=\"published\"\n\n# Each published item now lives at /restaurants/{slug}, included in\n# sitemap.xml. Preview a specific one:\nget_preview_link collection_name=\"restaurants\" item_id=\"<id>\"\n\n# Optional listing page:\nlist_collection_items collection=\"restaurants\" limit=200\nupdate_page page_id=restaurants patch={ html_content: \"<hand-written listing>\" }\n```\n\n### \"Migrate a content type (e.g. WP custom post type)\"\n\n```\nlist_collections                                     # what exists today?\nread_collection name=blog                            # what fields are writable?\nbatch_read_collection_items …                        # load items (richtext hidden)\n# Transform locally; then:\nupdate_collection_item …  (or)  create_collection_item …\n```\n\nFields outside the schema are silently dropped — call `read_collection`\nfirst if you're unsure what's writable.\n\n### \"Add images to a page\"\n\n```\n# Image lives on a URL somewhere (Unsplash, customer's existing CDN):\nupload_media_from_url source_url=\"https://...\" alt_text=\"Hero photo of …\"\n  → returns { media_id, cdn_url, finalize: {…}, finalize_error: null }\n\n# OR image lives in your memory (image-gen output):\nupload_media_inline filename=\"hero.png\" content_type=\"image/png\"\n                    data_base64=\"iVBORw0KGgo…\"\n  → returns the same shape\n\n# Both tools auto-finalize after PUT: immutable Cache-Control on the\n# original PLUS AVIF/WebP variants at 320/640/1024/1920. No manual\n# generate_image_variants call needed. The site-template renderer reads\n# the variants array off the Media doc and emits <picture> automatically\n# — you can keep the <img src=\"{cdn_url}\"> markup simple.\n#\n# INTEGRITY — don't lose bytes in transit. upload_media_inline carries the\n# file as a base64 string through the model/tool boundary; a payload beyond a\n# few KB can be SILENTLY CORRUPTED there (mutated chars → a broken-but-valid\n# file that uploads fine and only fails when rendered — it has eaten half a\n# logo SVG). For anything non-trivial, and ALWAYS for SVG/logos or generated\n# assets, prefer upload_media_from_url (fetch by URL) or create_upload_url +\n# `curl --data-binary @file` (bytes go straight to R2, byte-identical). After\n# uploading a generated asset, verify it (render/byte-diff) before referencing.\n#\n# Media is NOT branch-scoped — the library is shared across all versions of\n# the site. Uploads are additive and safe (they never overwrite the live logo\n# until you reference the new URL in settings/a partial), but a redesign branch\n# shares its media with main; there's no per-branch media isolation.\n\n# Then embed in a page:\nread_page page_id=...\nupdate_page page_id=... patch={ html_content: \"<...><img src='{cdn_url}' alt='…' /></...>\" }\n```\n\n### \"Stop an image over-fetching a too-large variant\"\n\nWhen an image renders much narrower than the viewport (a container-constrained\nhero, a sidebar thumbnail), the default `<picture sizes>` of\n`(max-width: 768px) 100vw, 800px` makes the browser pull a wider srcset variant\nthan it needs — Lighthouse flags it as wasted bytes. Three levers, narrowest\nwins:\n\n```\n# 1. Per-image: put a real `sizes` on the <img>. Survives the transform verbatim.\nupdate_page page_id=... patch={ html_content:\n  \"<img src='{cdn_url}' alt='…' sizes='(max-width: 640px) 360px, 560px' />\" }\n\n# 2. Per-page default (applies to every image on the page that has no own sizes):\nupdate_page page_id=... patch={ image_sizes_default: \"(max-width: 640px) 360px, 560px\" }\n\n# 3. Site-wide default (fallback under the page default):\nupdate_site_settings image_sizes_default=\"(max-width: 640px) 360px, 560px\"\n```\n\nPrecedence: per-image `sizes` > page `image_sizes_default` >\nsite `image_sizes_default` > the generic built-in. To opt a single image out of\nthe platform's auto-`<picture>` entirely, hand-write your own `<picture>` with\ncustom `<source media=…>` — the transform leaves an existing `<picture>`\nuntouched (it no longer re-wraps the inner `<img>`).\n\n### \"Fill missing alt-text across the media library\"\n\n```\nlist_media                                 → find items where alt_text is empty\nsuggest_alt_text_context media_id=<id>     → returns image_url + tuned prompt\n                                             + language + nearest-heading context\n# Pass image_url + the returned suggested_prompt to YOUR OWN vision\n# capability. The platform does NOT run vision for you.\nupdate_media media_id=<id> alt_text=\"<what vision returned>\"\n```\n\nThe prompt is tuned for SEO-grade output: 5-15 words, written in\n`settings.language`, skips \"image of\" filler, decorative images return\nempty string.\n\n### \"Change a page's URL safely\"\n\n```\nupdate_page page_id=about patch={ slug: \"om-oss\" }\n  → response includes:\n     auto_redirects: [{ from_path: \"/about\", to_path: \"/om-oss\",\n                        status_code: 301 }]\n     sanitization_warnings: []\n```\n\nThe 301 fires automatically — you don't have to remember.\n\nRedirect hygiene is automatic in both directions (since 0.16.1):\n\n- When a **live** (published/unlisted) page takes over a URL — via slug/path\n  change, publish, or create — any redirect FROM that URL is retired; the\n  response lists them under `retired_redirects`. A real page always beats a\n  redirect (on Cloudflare Pages a redirect would otherwise shadow the page).\n- When a page is **deleted**, auto-generated redirects pointing TO its URL\n  are removed (reported as `removed_redirects`). Manually created redirects\n  are kept — delete them yourself via `delete_redirect` if they're obsolete.\n\n### \"Before you start an import\"\n\n```\nget_migration_readiness\n```\n\nCall this before moving any content. Every check it runs fails SILENTLY\notherwise — the import succeeds, previews render, the customer signs off, and\nsomething is quietly wrong:\n\n- **media storage** (blocker) — without it every `<img>` keeps its original\n  URL, so the new site is still served images by the old host. Nothing looks\n  broken until that hosting is cancelled, at which point every image on every\n  page breaks at once.\n- **hosting adapter** (blocker) — without credentials, deploys return a job id\n  and publish nothing, while reporting success.\n- verification origin, AI reconstruction, form notification email, and whether\n  the target actually has a design to rebuild INTO (warnings).\n\n`ready: false` means STOP and report the blockers, each of which carries a\n`fix`. Don't start \"and fix it after\": the content work would have to be\nredone. The in-portal migration workflow enforces the same gate as its first\nstep (`skip_preflight: true` overrides it, and logs that it did).\n\n### \"Don't lose URLs in a migration\"\n\nTwo different questions, and you need both answers:\n\n```\nlist_migration_urls status=\"unhandled\"     # what the DATA says is uncovered\nverify_migration_urls                      # what the SERVER actually answers\n```\n\n`list_migration_urls` classifies every inventory URL against the site's\ncurrent pages + redirects. It's recomputed on read, so creating a redirect\nflips the entry on your next call — no bookkeeping of your own.\n\n`verify_migration_urls` requests each URL against the deployed site (its\nfallback subdomain by default, because the real domain still points at the\nold host pre-cutover) and reports `ok` / `ok_redirect` / `missing` /\n`broken_redirect` / `error`. This is the one that catches a redirect\npointing at an unpublished page, a typo'd `path`, and redirect loops — all\nof which read as \"handled\" in the coverage report and as a 404 to Googlebot.\n**Deploy first**: it tests saved, deployed content, not your drafts.\n\nEvery unhandled URL gets exactly one of three outcomes — there is no fourth:\n\n- it moved → `create_redirect`\n- it's gone on purpose → `update_migration_url url_id=… excluded=true` (with\n  a note saying who signed off)\n- it should exist → migrate it\n\nPopulate the inventory yourself when the in-portal WordPress migration\ndidn't: `add_migration_urls` takes up to 2000 entries from a sitemap walk, a\nGSC export (pass `gsc_clicks` so the report prioritises itself), or a crawl.\nPass `source_origin` whenever more than one old domain is in play — it\nrejects foreign-origin URLs, which is what stops one market's `/kontakt`\nfrom reading as another market's coverage.\n\nFor a whole family of sites, read the `tr-migrate-multisite` skill.\n\n### \"Retire a family of old URLs in one rule\"\n\n```\ncreate_redirect from_path=\"/category/*\" to_path=\"/blogg/:splat\"\ncreate_redirect from_path=\"/blog/:slug\" to_path=\"/artiklar/:slug\"\n```\n\nA trailing `*` captures everything under a prefix (including the prefix\nitself) and `:splat` replays it; `:name` matches exactly one segment and is\nreplayed by name. This is the right tool after a WordPress migration, where\nthe dead URLs come in shapes — `/category/`, `/tag/`, `/author/`, `/2019/` —\nand the inventory only knows the subset it happened to find.\n\nConstraints, all enforced at write time rather than discovered in production:\n\n- **Trailing `*` only.** Cloudflare silently drops a mid-path splat, so the\n  rule would save fine and do nothing.\n- **`:splat` requires a `*`**, and `:name` in the target must be declared in\n  `from_path`.\n- **Query strings can't be matched** — `_redirects` keys on the path. A\n  WordPress `/?p=123` URL has to be handled at the source.\n- **A rule that would hide a live page is refused**, naming the pages.\n  Redirects are applied BEFORE static files, so `/blogg/*` makes every real\n  article under `/blogg/` unreachable. Narrow the prefix.\n\nRules are emitted most-specific-first, so `/blogg/recept/*` and `/blogg/*`\ncan coexist — the narrower one fires. `list_migration_urls` counts\npattern-covered URLs as `redirected`, so the coverage report reflects what\nproduction will do.\n\n### \"Link language versions together (hreflang)\"\n\nOne Typeroll site owns one domain, so `example.se` / `example.de` /\n`example.co.uk` are three sites. Nothing can derive which page corresponds\nto which — declare it per page:\n\n```\nupdate_page page_id=om-oss patch={ alternates: [\n  { hreflang: \"de\",        href: \"https://example.de/ueber-uns\" },\n  { hreflang: \"x-default\", href: \"https://example.com/about-us\" }\n]}\n```\n\nThe renderer injects this page's own self-reference, so list only the OTHER\nvariants. Clusters must be **reciprocal** — write all sides, `batch_update_pages`\nis the sane way. Use absolute URLs on the FINAL domains (never the\n`*.typeroll` fallback). Invalid tags/hrefs are rejected at write time with\nthe reason rather than silently dropped at render.\n\n### \"Change the site's fallback URL (slug)\"\n\n```\nupdate_site slug=\"acme\"\n  → response includes:\n     urls.fallback: \"https://acme.sites.typeroll.com\"\n     dns_note: \"New fallback URL … attached to CF Pages. SSL provisioning\n                takes 1–10 minutes after DNS propagates. …\"\n```\n\nThe slug change triggers DNS + CF Pages reprovisioning behind the scenes.\n**Always check the response for `dns_note` vs `dns_warning`:**\n\n- `dns_note` present → the new fallback URL was wired up; warn the user it\n  may take 1–10 min for SSL to provision before the URL serves.\n- `dns_warning` present → the slug was saved but DNS / CF attach failed.\n  The `urls.fallback` field is still returned (it's just `{slug}.{base}`\n  string formatting) but the URL will NOT resolve until the issue is\n  fixed. Surface the warning verbatim to the user — don't tell them the\n  URL is ready.\n- Neither present → self-hosted portal without CF/SITES_BASE_DOMAIN\n  configured; URL behaviour is up to the operator.\n\nThe old fallback URL keeps working (bookmarks + SEO survive). Customer\ncan manually deprovision the old one via the portal.\n\n## Safety boundaries\n\n- **HTML is sanitized at save.** No `<script>`, no `onclick`, no\n  `javascript:` URLs in page or partial bodies. `<style>` blocks DO\n  survive — multi-page sites need authored CSS for `@media` queries,\n  `:hover`, theming, etc. Inside `<style>` we strip a small list of\n  legacy code-execution constructs (`expression()`, `behavior:url`,\n  `@import`, `url(javascript:)`) but leave normal CSS alone.\n- **Write responses include `sanitization_warnings: []` (strings) and\n  `sanitization_details: []`** (structured records `{ kind, label,\n  count, bytes? }`). Use the structured form to programmatically retry\n  with a fixed input.\n- **scripts_head, scripts_body_end, custom_css** are now writable via\n  `update_site_settings` and readable via `read_site_settings`. Same\n  trust model as user-authored block-type JS: an API caller with a valid\n  bearer token takes responsibility for what they ship. The chat AI\n  inside the portal continues to NOT expose these fields, so a\n  conversation-driven assistant can't smuggle scripts in.\n- **The API key is site-scoped.** Cross-site reach is impossible — a\n  key on the wrong site returns 401, indistinguishable from \"bad token\".\n- **Audit log.** Every state-changing call (POST / PATCH / PUT /\n  DELETE) is logged. Reads aren't. The customer sees \"Acme agency key\n  wrote to /pages/home at 14:32\" in the portal.\n- **Rate limits.** 600 reads/min, 60 writes/min per key. On 429 the\n  response carries `Retry-After`.\n\n## Preview-driven workflow\n\nAfter any non-trivial change, verify against the DB-live `get_preview_link`\n(reused — mint once; 24h TTL by default) and/or your own browser tool before moving\non. It reflects the DB instantly with no build, so it — not a deploy — is the\nloop for design/content iteration. One reload vs. shipping a broken redesign —\nalways worth it.\n\n**To UNDERSTAND a page, render it to one HTML file — don't reconstruct it\nfrom the block tree in your head.** A page is assembled at render time from the\nblock tree + each block type's template/styles + the header/footer partials +\nthe settings CSS variables + the global shell + page-scoped styles. `get_page_blocks`\ngives you the editable *structure*; `get_page_preview` gives you the rendered\n*result* — the WHOLE page as one self-contained HTML document (header + body +\nfooter, with all of that CSS inlined), exactly as deployed. Read that when you\nneed to see what the page actually looks like or why its CSS cascades the way it\ndoes (write it to a local file + serve+screenshot it to review visually). Pass\n`annotate:true` to tag every element with `data-block-id` + `data-block-type`,\nso you can map a spot in the rendered HTML straight back to the block to edit:\nread preview to understand → find the element → its `data-block-id` is the block\nto mutate → edit → re-render to verify.\n\n**CSS precedence — where your overrides land in the cascade.** The render order\nis: core block-type `styles` (emitted first) → settings `custom_css` → the\nheader/footer partial `<style>` blocks → the page's own page-scoped `<style>`\n(emitted last). Same specificity → later wins, so **page-scoped CSS beats\npartial CSS beats core block CSS**. Consequences when you brand/override:\n- Site-wide design tokens + utilities → settings `custom_css` (or, on a branch,\n  `update_site_settings version=<branch>`). Header/footer-only tweaks → the\n  partial. One page → that page's `<style>`.\n- Core blocks set their own chrome (e.g. `core/image` gives `figure>img` a\n  `border-radius`/`margin`; `.page-content img` adds more). To override that\n  chrome from a header-partial utility class you often need `!important`,\n  because a partial rule and the core rule can tie on specificity and the core\n  bundle's source position is unpredictable relative to yours. That `!important`\n  is expected today — it is NOT a smell. (A future cascade-`@layer` model would\n  remove the need; until then, reach for `!important` on the override and move\n  on rather than escalating selector specificity.)\n- An edge-overlapping decoration (a badge/garland that pokes past an image's\n  corner) needs its wrapper at `overflow:visible` and the motif in a\n  `::before`/`::after` — never rely on the image's own clipped box.\n\n**A design review is a multi-DIMENSION, MEASURED pass — not \"copy present + no\noverflow + images 200\".** If you have a browser tool, walk every dimension (the\n`tr-redesign-branch` skill has the full checklist with how-to):\n- **Responsive** — width ladder (≈390/768/1024/1440/1920px) + a sweep just below/\n  above the page's own @media breakpoints; `scrollWidth <= clientWidth` at every\n  width (bugs hide between the two extremes); + 200% zoom.\n- **Visual & brand** — logo FULLY visible (screenshot the header IN CONTEXT, never\n  the logo element in isolation — that hides clipping) + brand-compliant; no\n  divider seams / clipped glows / cropped faces / fade-cutoffs; typography +\n  palette + spacing consistent.\n- **Accessibility (measure)** — actual contrast ratios (AA 4.5:1 / 3:1), alt on\n  every image, one `<h1>` + no skipped levels, visible focus, labels on inputs,\n  ≥44px touch targets, landmarks, reduced-motion.\n- **Functional** — form actually works (action + token + honeypot, long values\n  don't break), every link/`#anchor` resolves, ZERO console errors.\n- **Content** — no unrendered `{{…}}`, no placeholder, copy matches the live page.\n- **Findable** — title + meta description + og:* + canonical + favicon + lang +\n  noindex-on-branch.\n- **Fast** — images sized right + modern format + width/height set + lazy/eager.\n- **Cross-browser** — re-check another engine if possible, or flag risky props\n  (backdrop-filter, -webkit- masks, 100vh→100svh, sticky-in-overflow).\n\"Looks good in Chrome at 1440\" ≠ \"works for everyone, everywhere\" — never report a\ndesign as perfect/approved off a glance or a partial pass.\n\nPreview shows DB state (drafts included). Live (`get_site → urls.production`)\nshows the most recent deploy. Branch deploys live at\n`get_version → deploy_url` (`{branch}.{project}.pages.dev`).\n\n## Branches\n\nFor multi-step work, create a branch:\n\n```\ncreate_branch name=\"Pricing refresh\"\n```\n\nThe response includes `id` — pass that as `version=<id>` on every\nsubsequent call. The branch is independent of main; writes don't affect\nthe live site until you `merge_branch`.\n\nBranches default `robots_blocked: true`. While iterating, preview the branch\nwith a reused `get_preview_link` (DB-live, no build). Deploys to a branch land\nat a stable URL (`{branch}.{project}.pages.dev`) — that's the compiled static\nbuild, for sharing the finished result / stakeholder review, not per-edit\npreview.\n\n## When in doubt\n\n- **Read before you write.** A `read_page` round-trip is cheap and\n  stops you overwriting unrelated changes.\n- **Dry-run bulk operations.** `bulk_replace_text` accepts `dry_run:\n  true` and returns 3 sample diffs. Show them to the user before the\n  real run.\n- **Watch the sanitization_warnings array.** If it's non-empty, the\n  stored HTML differs from what you sent. Read it back to confirm.\n- **One small confirmation > one large undo.** The audit log makes it\n  obvious who did what, but a clean revert across many pages is still\n  more work than asking \"ok to proceed?\" first.\n- **Match the site's design.** Read a partial or two before designing\n  new components. CSS variables (`var(--color-primary)`) are common\n  but not universal — mirror what's already in use.\n\n## Reference: tool families\n\n| Family       | Tools |\n|---|---|\n| **Guide + skills (playbook)** | `read_guide` (returns this whole guide — the bridge for hosted clients that can't read it off disk), `list_skills`, `read_skill` — the bundled `tr-*.md` recipes (incl. `tr-responsive` for per-breakpoint layout). Call `list_skills` first when a task looks like \"build / migrate / redesign a site\", then `read_skill name=…`. No API key or site context needed. |\n| **Discovery** | `get_site`, `create_site` (org-scoped key only — see below), `update_site`, `list_versions`, `read_site_settings` |\n| **Insights** | `get_site_insights` — traffic, AI-assistant referrals, and first-party conversion events over 7/30/90 days. Read-only. Traffic is powered by Cloudflare Web Analytics; conversion rows come from validated Analytics attribution `click_event` targets and can be present even when the traffic provider is unavailable. |\n| **Pages — reads** | `list_pages`, `read_page`, `batch_read_pages` |\n| **Pages — writes** | `create_page`, `update_page`, `replace_page`, `batch_update_pages`, `delete_page`, `clone_page` |\n| **Pages — blocks** | `get_page_blocks`, `add_block`, `update_block`, `move_block`, `remove_block`, `set_page_mode`, `convert_page_to_blocks` |\n| **Pages — meta** | `get_page_preview` |\n| **Global blocks (partials)** | `list_partials` (summary by default), `read_partial`, `create_free_block`, `update_partial`, `replace_partial`, `delete_partial`, `find_pages_using_block`, `list_blocks_with_usage` |\n| **Block types** | `list_block_types`, `read_block_type`, `find_pages_using_block_type`, `export_block_types`, `import_block_types` |\n| **Collections** | `create_collection`, `update_collection_schema`, `delete_collection`, `list_collections`, `read_collection`, `list_collection_items` (richtext hidden by default), `read_collection_item`, `batch_read_collection_items`, `create_collection_item`, `update_collection_item`, `delete_collection_item`, `regenerate_collection_listing` |\n| **Media** | `list_media`, `read_media`, `create_upload_url`, `upload_media_from_url`, `upload_media_inline`, `update_media`, `delete_media`, `finalize_media`, `finalize_all_media`, `generate_image_variants`, `suggest_alt_text_context` |\n| **Redirects** | `list_redirects`, `create_redirect`, `delete_redirect`. `from_path` may be a PATTERN: a trailing `*` (with `:splat` in the target) or `:name` for one segment — one rule retires a whole family of dead URLs (`/category/*` → `/blogg/:splat`). Mid-path splats and query strings are refused, as is any rule that would hide a live page. |\n| **Migration inventory** | `get_migration_readiness` (preflight — CALL FIRST on any import), `list_migration_urls`, `add_migration_urls`, `update_migration_url`, `delete_migration_url`, `verify_migration_urls`. The legacy site's URL list with LIVE coverage status (`migrated` / `redirected` / `excluded` / `unhandled`, recomputed on every read from current pages + redirects). `verify_migration_urls` is the pre-cutover check that REQUESTS every URL against the deployed site — see \"Don't lose URLs in a migration\" below. |\n| **Forms** | `list_forms`, `read_form`, `create_form`, `update_form`, `delete_form`, `list_form_submissions`, `delete_form_submission` (removes one submission — e.g. cleaning up a test entry; `delete_form` with `delete_submissions` is the bulk path). **Steps (form/* block trees) are the ONLY stored model** (0.29.0+, `forms_steps_only` capability): pass `steps` for funnels, or `fields` for simple forms — the server converts a flat field list to a single static step (read_form shows the resulting steps). **Placing a form: prefer a `core/form` block with `data.form_id`** — it renders every form with validation + signed token handled. The raw `<form method=\"POST\">` embed with hidden `_token` + honeypot `_hp` (from read/create's `submit_token`/`submit_url`) is the HTML-mode fallback, single-step forms only; no client JS (the sanitizer strips inline `<script>`) |\n| **Settings** | `update_site_settings` (whitelist) |\n| **Analytics attribution** | `read_funnel_attribution`, `update_funnel_attribution` (admin; redeploy after changes) |\n| **Search + bulk** | `search_pages`, `bulk_replace_text` |\n| **Branches** | `create_branch`, `read_version`, `delete_branch`, `merge_branch` |\n| **Deploy** | `trigger_deploy`, `list_deploys`, `get_deploy_status` |\n| **Preview** | `get_preview_link`, `get_page_preview` |\n\nEvery tool's input is validated server-side; the MCP server only does\nauth + shape. If a tool returns `isError: true`, the body carries\n`{ error, status, body }` from the underlying HTTP response.\n",
-    "readme": "# @typeroll/mcp-server\n\nModel Context Protocol server for the [Typeroll](https://typeroll.com)\npublic API. Lets Claude (Desktop / claude.ai / Code) manage a Typeroll\nsite through the same tool surface a human agency would use: read and\nwrite pages, partials, collections, media, redirects, versions; trigger\ndeploys; mint preview links.\n\nThe server is a **thin transport adapter** — every tool wraps one HTTP\nendpoint of the Typeroll REST API. Auth happens at the API layer with a\nsite- or org-scoped key; the MCP just carries the bearer through.\n\n## Two ways to connect\n\n- **Hosted (Claude Desktop / claude.ai) — paste a URL.** No CLI, no\n  Node.js install. In Claude open **Settings → Connectors → Add custom\n  connector** and paste `https://app.typeroll.com/api/mcp`\n  (or `https://<your-self-hosted-portal>/mcp`). Claude opens a consent\n  page; paste your Typeroll API key there.\n- **Stdio (Claude Code) — one `claude mcp add` command.** Best for local\n  dev / agency staff already in a terminal. Instructions below.\n\nThis npm package is the stdio transport. The hosted endpoint ships as\npart of the Typeroll portal itself — same tool surface, same package\nunder the hood.\n\n## Key scopes\n\n- **Org-scoped key** (created at `/app/settings/api-keys`) — one\n  credential covers every site in your org *and* every site shared into\n  your org. The default for the hosted Claude connector. Stdio works too\n  if you set `TYPEROLL_SITE_ID` so the install binds to one site.\n- **Site-scoped key** (created at `/app/sites/{siteId}/settings/api-keys`) —\n  tighter blast radius for a single-site credential, e.g. one you'd\n  hand to a customer for a self-managed site.\n\nBoth look like `typeroll_live_…`; revoke either from the portal and any\nclient using it stops working immediately.\n\n## Stdio quick start (Claude Code)\n\n1. **Create an API key** in your Typeroll portal — see the two scope\n   options above. Org-scoped is the right default.\n\n2. **Add the server to Claude Code.** Drop this into `~/.claude.json`\n   (or your local `.claude/config.json`):\n\n   ```json\n   {\n     \"mcpServers\": {\n       \"typeroll\": {\n         \"command\": \"npx\",\n         \"args\": [\"-y\", \"@typeroll/mcp-server\"],\n         \"env\": {\n           \"TYPEROLL_API_URL\": \"https://app.typeroll.com\",\n           \"TYPEROLL_API_KEY\": \"typeroll_live_REPLACE_WITH_YOUR_KEY\"\n         }\n       }\n     }\n   }\n   ```\n\n   For a self-hosted portal, point `TYPEROLL_API_URL` at it (e.g.\n   `https://cms.example.com`).\n\n   Prefer a scaffold? Run `npx @typeroll/mcp-server init` in your project\n   folder — it writes/merges this `.mcp.json`, copies the skills into\n   `.claude/skills/`, and adds an `AGENTS.md` pointer + imagegen-lab\n   files. Idempotent; `--force` to overwrite. (Skills only:\n   `npx @typeroll/mcp-server install-skills .claude/skills`.)\n\n3. **Tell the agent what kind of work you want.** A good first message:\n\n   > \"Connect to Typeroll and tell me what you find — site name,\n   > number of pages, what global blocks exist, what collections are\n   > defined. Then I'll give you a task.\"\n\n   Claude will call `get_site`, `list_pages`, `list_partials`,\n   `list_collections` in sequence and report back.\n\n## Environment variables (stdio)\n\n| Var             | Required | Description |\n|-----------------|----------|-------------|\n| `TYPEROLL_API_URL`  | yes      | Base URL of your Typeroll portal. |\n| `TYPEROLL_API_KEY`  | yes      | A `typeroll_live_…` bearer token. |\n| `TYPEROLL_SITE_ID`  | sometimes | Pin to a specific site. Required when using an org-scoped key over stdio (the install can only target one site at a time); auto-detected for site-scoped keys. |\n\n## What the agent should read first\n\nThe package ships [AGENTS.md](./AGENTS.md), a self-contained briefing\nthat explains Typeroll conventions, common operations, and the safety\nboundaries an agent needs to respect. Point Claude at it (or include it\nin your project's `CLAUDE.md` / `AGENTS.md`) so it knows when to use\nwhich tool.\n\n## Tool surface\n\nAround 50 tools across these families. See [AGENTS.md](./AGENTS.md) for\nthe full reference + concrete operation recipes.\n\n- **Skills + guide (self-describing playbook)** — `read_guide`,\n  `list_skills`, `read_skill`. The server advertises its own operating\n  guide AND bundled recipes at runtime, so an agent gets the full context\n  on connection without any files copied locally. `read_guide` returns the\n  whole AGENTS.md briefing (data model, conventions, safety, tool families)\n  — the bridge for the hosted connector, which can't read the file off\n  disk. `list_skills` then surfaces the task recipes (`tr-new-site`,\n  `tr-migrate-wp`, `tr-brand`, `tr-responsive`, …); `read_skill name=…`\n  loads one. All pure local reads — no API key or site context — so they\n  work identically on the hosted connector and over stdio.\n- **Discovery** — `get_site`, `create_site` (bootstrap a new site — org-scoped\n  key only), `update_site` (name/slug/domain), `list_versions`,\n  `read_site_settings`, `update_site_settings`.\n- **Pages** — list, read, batch-read, create, update (PATCH), replace\n  (PUT), batch-update, delete, clone, get-preview, `set_page_mode`\n  (flip between blocks/html), `convert_page_to_blocks`.\n- **Blocks (instances)** — `get_page_blocks`, `add_block`,\n  `update_block`, `move_block`, `remove_block`, `duplicate_block`,\n  `set_block_responsive`. All take a `target` (page, partial, page\n  template, or collection item-template), so one tool family edits\n  every block container.\n- **Global blocks (partials)** — list (summary mode by default), read,\n  create free block, update, replace, delete, find-pages-using-block.\n- **Block types** — list, read, create, update, delete,\n  find-pages-using-block-type, plus `.tcblocks` export/import. Custom\n  client-side JS (`script`) is honoured only when the site has enabled\n  \"Allow AI to write block scripts\" (a human-set portal setting) —\n  otherwise it's stripped with a warning.\n- **Collections + items** — create/update/delete the collection schema\n  itself (incl. `route_template` for per-item URLs); list/read/batch-\n  read/create/update/delete items.\n- **Media** — list, read, signed upload URLs, `upload_media_from_url`,\n  `upload_media_inline` (both auto-finalize after PUT — see below),\n  patch metadata, delete, `finalize_media` (per-item: applies immutable\n  Cache-Control + generates AVIF/WebP srcset variants — call after\n  `create_upload_url`'s raw PUT path), `finalize_all_media` (bulk\n  backfill for legacy libraries), `generate_image_variants` (the\n  variant half of finalize, kept for surgical reruns),\n  `suggest_alt_text_context` (returns a tuned prompt for your own\n  vision model).\n- **Redirects** — list, create, delete. Plus automatic 301 on slug change.\n- **Forms** — list, read, create, update, delete, list submissions.\n  Read/create responses include `submit_token` + `submit_url` — a plain\n  `<form method=\"POST\">` with a hidden `_token` input is a fully working\n  no-JS embed (the endpoint answers form posts with an HTML\n  confirmation page).\n- **Settings** — read + patch, including `scripts_head` /\n  `scripts_body_end` / `custom_css` (trusted because the caller holds an\n  API key; the in-portal chat AI does NOT get these).\n- **Search** — `search_pages` with substring or regex.\n- **Bulk** — `bulk_replace_text` with dry-run.\n- **Branches** — create, read, delete, merge. Branch deploys get their\n  own URL at `{branch}.{project}.pages.dev`.\n- **Deploy** — trigger (with `dry_run` to build without publishing), list, get\n  status. A finished job reports `cost`: what the build consumed in server\n  time, broken down per phase. Estimates from a rate card, not billing records.\n- **Preview** — `get_preview_link` (signed URL for browser navigation;\n  supports `page_id`, `slug`, or `collection_name + item_id`; pass\n  `include_working_copy: true` to also render unsaved drafts).\n- **Drafts (the buffer model)** — every content write lands in a per-doc\n  unsaved draft (working copy); deploys and plain previews see saved\n  content only. Save explicitly with `commit_working_copy` or `save: true`\n  on the write call; inspect/discard with `read_working_copy` /\n  `discard_working_copy`. Status changes and structural operations apply\n  immediately.\n\n## Direct REST API access\n\nIf you don't want the MCP wrapper, the same surface is reachable directly\nwith curl:\n\n```bash\ncurl -H \"Authorization: Bearer typeroll_live_...\" \\\n  https://app.typeroll.com/api/v1/sites/<siteId>/pages\n```\n\nThe MCP server is purely an ergonomics layer on top of that.\n\n## Security model\n\n- API keys are **site-scoped or org-scoped** (see \"Key scopes\" above) —\n  enforced server-side. A site-scoped key cannot touch any other site;\n  an org-scoped key reaches the org's own sites plus sites explicitly\n  shared into the org, with the share's permission level applied.\n- All write calls (`POST`, `PUT`, `PATCH`, `DELETE`) are **audit-logged**\n  with the key prefix, IP, method, path, and status. Reads are not\n  logged (cost vs. value).\n- **Rate limits**: 600 reads/min, 60 writes/min per key. 429 responses\n  carry `Retry-After` headers.\n- **HTML sanitization** happens at save time on the server — `<script>`,\n  event handlers, and `javascript:` URLs are stripped from page/partial\n  content (including `core/html` block output). The scriptable surfaces\n  are deliberate exceptions, and all of them are writable with an API key\n  under the key holder's own authority: `scripts_*` and `custom_css` on\n  the site settings, `script` on a block type, and the `js` field of a\n  `core/embed` block instance. Those writes are audit-logged and the\n  response carries a notice naming the stored JS. Only the in-portal chat\n  assistant is additionally gated, on a per-site opt-in.\n- Keys can be **revoked** at any time from the portal. Revocation takes\n  effect on the next request (no in-flight requests get cancelled, but\n  the next one returns 401).\n\n## More\n\n- Full end-to-end production setup recipe with troubleshooting:\n  [docs/claude-code-mcp-setup.md](../../docs/claude-code-mcp-setup.md)\n- Agent operations briefing: [AGENTS.md](./AGENTS.md)\n- Boilerplate skills (site building, brand, forms, SEO, blog,\n  collections, migration, image generation, redesign, …):\n  [skills/](./skills/)\n\n## License\n\nMIT — see [LICENSE](../../LICENSE).\n"
+    "agents": "# AGENTS.md — Working on a Typeroll site\n\nYou are connected to a Typeroll site through `@typeroll/mcp-server`.\nThis file is your briefing: what the system is, what conventions matter,\nwhat tools to reach for first.\n\nIf anything below conflicts with what you observe in the tools, trust the\ntools — the platform may have moved since this was written.\n\n**Start here for site-shaped tasks.** When the user wants to build,\nmigrate, redesign, or brand a site, call `list_skills` first — the server\nadvertises its own step-by-step playbook (`tr-new-site`, `tr-migrate-wp`,\n`tr-brand`, …). Then `read_skill name=…` loads the full recipe. These are\nlocal reads; no API key or site context required.\n\n**Branch first for anything larger than a small edit.** Before a redesign,\na multi-page change, or trying out a new design direction, run\n`create_branch name=\"…\"` and pass the returned id as `version=<id>` on every\nsubsequent read/write. The work stays off the live `main` version until you\n`merge_branch` it — nothing ships until you decide it should. Branches default\n`robots_blocked:true` and get their own deploy URL for stakeholder review.\nIt's the cheapest insurance there is; when in doubt, branch. The\n`tr-redesign-branch` skill walks the whole flow. (Small, low-risk single edits\ncan go straight to main.)\n\n## What this is\n\nTyperoll is a static-site CMS: content lives in a database, the user\nedits it through an in-app editor, and a deploy step compiles everything\nto a fast static site hosted on Cloudflare Pages. The in-app chat handles\nsingle-page or single-block edits by the editor audience. You — through\nthis MCP — handle the work that doesn't fit there: site-wide redesigns,\nbulk content updates, structural migrations, directory imports.\n\nThe MCP server is a thin wrapper around the public REST API. Each tool\nmaps to one HTTP endpoint; the actual logic runs in the customer's portal\n(SaaS or self-hosted).\n\n## The data model in 90 seconds\n\n- **Pages.** Title, slug, status (`draft | review | unlisted | published`),\n  body content + SEO fields. Two body shapes selectable per page via\n  `content_mode`:\n  - `blocks` (DEFAULT for new pages) — `blocks: Block[]` tree of typed\n    blocks (heading, prose, section, columns, image, button, plus any\n    user/third-party block types installed on the site). Use the\n    block-mutation tools (`add_block`, `update_block`, `move_block`,\n    `remove_block`) for structural changes.\n  - `html` — body lives in `html_content` as a single HTML string.\n    Useful when you have hand-written markup to drop in directly.\n\n  Slug is a single path segment — no slashes. `about` → `/about`,\n  `kontakt` → `/kontakt`, empty string `\"\"` → homepage. The v1 API\n  rejects `services/design` and other slash-containing slugs with\n  \"Invalid slug … slugs must not contain slashes.\" For nested URLs\n  like `/blog/{slug}` or `/services/{slug}`, the right primitive is a\n  **collection with `route_template`** (see the `tr-blog` and\n  `tr-directory` skills) — not a flat page with a slashed slug.\n\n- **Partials = global blocks.** Three kinds:\n  - `header` — auto-injected at the top of every page.\n  - `footer` — auto-injected at the bottom of every page.\n  - `free` — reusable HTML you drop into a page with\n    `<x-include name=\"block-id\" />`. Free blocks are how you avoid\n    duplicating HTML across HTML-mode pages.\n\n  Partials themselves also support `content_mode='blocks'` — pass a\n  `blocks: Block[]` tree to `update_partial` and the renderer composes\n  it the same way as a page. Useful for header/footer authored with\n  block types.\n\n- **Collections.** Repeatable content types (blog, team, events,\n  products, restaurants for a directory site, etc.). Each has a schema\n  (`fields[]`) and optional **per-item routing** via `route_template`\n  (e.g. `/restaurants/{slug}`). When set, every published item gets its\n  own static URL rendered through `item_template_html`. Set\n  `route_template=\"\"` to opt out and keep the collection listing-only.\n\n- **Settings.** Site name, tagline, logo, favicon, colors, fonts,\n  contact info, social links, SEO suffix, default meta description\n  (`default_meta_description` — site-wide fallback for pages without a\n  `seo_description`; tagline is the last resort), plus `scripts_head`,\n  `scripts_body_end`, `custom_css` (writable via the API — your bearer\n  token authorises shipping arbitrary CSS/JS to the live site, just\n  like editing a partial's HTML does).\n\n- **Analytics attribution.** `read_funnel_attribution` and\n  `update_funnel_attribution` manage the Analytics module's validated,\n  site-level rules that forward\n  allowlisted query parameters to exact HTTPS link targets. Optional\n  first-/last-touch storage is consent-gated and off unless configured. Read\n  the current config before writing. For advertising pass-through, preserve\n  only incoming or stored values and omit `fallback`: a fallback creates\n  synthetic attribution when no campaign value exists. The API rejects such\n  fallbacks unless `allow_synthetic_fallbacks=true` explicitly acknowledges\n  that behavior. A target `click_event` is recorded as a consent-gated,\n  first-party conversion when Analytics is enabled; the event contains only\n  the declared funnel, destination, page path, and resolved allowlisted\n  attribution. Navigation never waits for analytics.\n\n- **Apps.** `list_apps`, `read_app`, and `update_app` expose the complete\n  code-defined app registry through the same admin API key used for content\n  and deploys. Read the schema before writing. Secret fields stay masked on\n  reads and encrypted at rest; omitted fields preserve their current values.\n  When `affects_build` is true, deploy after the update.\n\n- **Page templates.** A `PageTemplate` is a Block[] tree that wraps a\n  page's body. The template contains exactly one block of type\n  `template_content_slot` — at render time that block gets replaced by\n  the page's own `blocks`. Set `Page.template = \"<template-id>\"` to\n  apply a template to a page.\n\n- **Block types.** A site has three sources of block types:\n  - **Core** (origin: 'core', ids like `core/section`) — shipped in\n    the platform, always available.\n  - **User** (origin: 'user') — created in the portal's block-types UI.\n  - **Third-party** (origin: 'third_party') — imported from .tcblocks\n    packages via `import_block_types`.\n\n  `list_block_types` returns ALL of them in one list as a lightweight\n  summary: each entry's id, label, category, container/slot info, origin,\n  and full field schema (names, types, defaults) — but NOT the render-time\n  template/styles/script (omitted so the list stays within token budget as\n  the library grows). Use `read_block_type` for one block's markup, or pass\n  `full:true` to inline it for every block. Always call this FIRST before\n  working with blocks — never hardcode block ids or field names, the\n  available set is per-site.\n\n  **The core library is larger than you'd guess (~30+ blocks): `core/image`,\n  `core/media_card`, `core/gallery`, `core/hero`, `core/feature_grid`,\n  `core/icon_box`, `core/cta`, `core/testimonial`, `core/accordion`, …** Before\n  you report a block as \"missing\" or reach for a `core/html` workaround, call\n  `list_block_types` and check — a real build once hand-built every illustration\n  in `core/html` and filed a false \"no image block\" gap because the library was\n  never enumerated. Prefer a native block; `core/html` is the last resort.\n\n  Block-library specifics worth knowing (template_capabilities_version\n  0.15.0):\n  - **`core/media_card`** — image + text side by side (image left/right,\n    width third/two-fifths/half, heading + richtext + button, optional\n    card background/radius; stacks image-on-top below 720px). Use it for\n    the classic \"photo next to copy\" layout instead of hand-building\n    section+grid+html.\n  - **`core/search`** (0.29.0+) — site search over the deployed site.\n    Place the block anywhere; the deploy pipeline detects it, runs\n    Pagefind over the built HTML, and the block loads the search UI at\n    visit time. Index = page content only (nav/footer excluded); noindex\n    pages stay out. Editor preview shows a placeholder note (the index\n    only exists on the deployed site).\n  - **Archive pagination** (0.29.0+): a collection listing\n    (`core/collection_list` / `core/repeater`) with `paginate: N` renders\n    N items per page + a pager, and the build generates `/page/2/`… routes\n    automatically. `paginate` supersedes `limit`; one paginated listing\n    per page.\n  - **`core/feature_row`** (0.29.0+) — the full-width \"zig-zag\"\n    feature/step row: balanced halves that hug the center gutter (no\n    wide-screen dead air), natural-aspect image (never cover-cropped —\n    that's media_card's card look), eyebrow + heading + richtext +\n    button pair, `image_side: left|right` per row, `stack_order`\n    controls what comes first on mobile. Prefer it over `core/columns`\n    with an unbalanced ratio + width-capped text for these rows.\n  - **`core/hero` and `core/cta` render their buttons server-side** via\n    `primary_label`/`primary_url` + `secondary_label`/`secondary_url`.\n    (The old `buttons` array relied on client hydration that never\n    existed — if you see `data-buttons` in stored content it renders\n    nothing; rebuild with the explicit fields.)\n  - **`core/image` gets responsive `<picture>` automatically at build\n    time** — the deploy pipeline's SEO transform converts CDN `<img>`\n    into `<picture>` with AVIF/WebP srcset variants. You do NOT need\n    `core/html` for responsive images; just point `src` at an uploaded\n    media URL (run `generate_image_variants` first) and optionally set\n    `radius`. Note: the in-portal preview shows the plain `<img>` — the\n    `<picture>` upgrade appears on the deployed site.\n  - **Icons render inline SVG** (since template_capabilities_version\n    0.16.0). Every `type: 'icon'` schema field — on `core/icon`,\n    `core/icon_box`, `core/step_card`, and custom block types — renders\n    a stroke-based inline SVG when the value is a name from\n    `get_site_capabilities → core_icon_names` (a curated Lucide subset:\n    `check`, `star`, `shield-check`, `mail`, `arrow-right`, `zap`,\n    `truck`, `chart-line`, …). Any other value (emoji, plain text) is\n    rendered as escaped text, so emoji stand-ins keep working. Icons\n    size with `font-size` (the SVG is 1em) and paint with\n    `currentColor`. Custom block templates opt in by placing the derived\n    raw token `{{{<field>_svg}}}` where the icon should appear. On\n    pre-0.16.0 portals icons don't render — use emoji or CSS markers.\n    `core/tabs` label icons are the remaining gap (tab strip is built\n    client-side).\n  - **Grids with a partial last row: set `last_row: 'center'`** (since\n    template_capabilities_version 0.16.5). Five equal cards in a 3-col\n    `core/grid` (or 7 in 4, ...) left-align the orphans by default; with\n    `last_row: 'center'` the last row auto-centers. THE DESIGN RULE: when\n    N peer cards don't divide by the column count, center the last row or\n    change the column count — NEVER invent a \"wide\"/full-width variant of\n    one peer card just to fill the hole. Special treatment is a content\n    decision, not a layout patch.\n  - **`core/section` is natively full-bleed on block pages** (since\n    template_capabilities_version 0.14.0): the section's background runs\n    edge-to-edge and meets the header with zero gap; content inside is\n    constrained by the section's own inner container (`width` field:\n    narrow/normal/wide/full). Never use 100vw negative-margin hacks.\n    Top-level blocks that are NOT sections still get a classic centered\n    container as fallback. Anchor ids and custom classes via\n    `style_overrides` are safe on full-bleed sections since 0.15.3 —\n    they merge into the `<section>` element itself. On 0.14.x–0.15.2\n    they wrapped the section in a `<div>`, which silently disabled\n    full-bleed for that section.\n  - **Shaped section transitions** (since template_capabilities_version\n    0.24.0): `core/section` takes `divider_top` / `divider_bottom`\n    (`none | wave | curve | tilt`). The platform paints the divider in the\n    section's OWN `background` and overlaps the neighbour by 1px, so a\n    cream↔colour transition renders seam-free. **Use this for waves/curves —\n    never hand-roll a divider band in `core/html`** (a separate stacked shape\n    seams against the next section as a sub-pixel hairline in Chrome). Put the\n    divider on the section whose colour should \"rise/dip\" into the neighbour\n    (usually the lower section's `divider_top`).\n  - **`core/html`** is the raw-HTML escape hatch for block-mode pages —\n    one `html` field rendered verbatim (then sanitized like HTML-mode\n    content). Use it for the genuinely unique thing no block covers.\n    Prefer real blocks when one fits.\n  - **Structured records at scale** (template_capabilities_version ≥ 0.31.0).\n    Four things landed together for directory-shaped sites:\n    - `collection_completeness` — **start an enrichment pass here**, not by\n      paging every item. Returns per-field gap counts plus the N worst\n      records (missing fields, never-verified fields, fields whose last write\n      is older than the staleness window), computed at read time. Fields no\n      API key may write are excluded by default: a gap you can't close is\n      noise.\n    - **Per-field write authority.** A collection field can declare\n      `writable_by` (`portal | owner | agent | app | import`). A write you're\n      not permitted, or one that would overwrite a higher-precedence writer\n      (a human correction, the listed business's own edit), comes back as\n      **409 with the losing field names** — never a silent no-op. Treat that\n      as \"already handled\" and record it; retrying will lose again.\n    - **Item references.** `item_ref` / `item_ref_list` fields point at items\n      in another collection (`ref_collection`). The reverse direction is\n      computed at render time — don't try to maintain backlinks yourself.\n      Render them with a `core/repeater` whose `source_type` is `related`\n      (a ref field on the current item) or `backlinks` (who points at it).\n    - **Taxonomy pages.** `CollectionDef.facets` generates one page per\n      distinct field value. ⚠️ This turns record count into ROUTE count, and\n      route count is what the build timeout measures. `min_items` (default 2)\n      keeps thin-content pages out, and combination pages must be listed\n      explicitly in `facet_combinations` — never assume a cartesian product.\n  - **`core/embed`** (template_capabilities_version ≥ 0.30.0) is\n    `core/html` plus behaviour: an `html` field and a `js` field. Reach\n    for it when a one-off placement needs JavaScript. **A `<script>` tag\n    written into `core/html` — or into any page/block markup — is stripped\n    by the sanitizer no matter which credential wrote it**, so this field\n    is the supported route, not a workaround. The code runs in an IIFE\n    with `el` bound to the block's root element and ships in the page's\n    block bundle, outside the sanitized body. Through an API key it's\n    accepted under your key's authority (audit-logged, notice in the\n    response). Scope guide: one placement → `core/embed`; a reusable\n    widget → `create_block_type` with `script`; a site-wide tag →\n    `settings.scripts_head` / `scripts_body_end`.\n  - **Forms 2.0** (template_capabilities_version ≥ 0.18.0): forms can\n    carry `steps[]` — each step is a Block[] tree mixing `form/*` field\n    blocks (text/email/phone/number, textarea, select/radio_group/\n    checkbox_group, toggle, slider, date, URL, heading, help, consent,\n    hidden) with any content blocks. Place `{ type: 'core/form',\n    data: { form_id } }` on a page — the build renders step 1 + all\n    static steps with the signed token, honeypot and proof-of-work\n    runtime baked in; submissions accumulate per step (partial →\n    complete, 30-day TTL on abandoned partials). Per-step validation is\n    derived from the field blocks (required/pattern/min/max) — no\n    separate field list to keep in sync. `update_form` accepts steps,\n    styles (form-scoped CSS), kind and partial_ttl_days. On HTML-mode\n    pages, `<x-form id=\"…\" />` is expanded server-side through the same\n    renderer and supports the same initial state and multi-step runtime.\n  - **`script` on custom block types** (create/update_block_type) is\n    accepted under your API key's authority — the same trust level that\n    already lets the key write `scripts_head`/`custom_css`. Every\n    script-bearing write is audit-logged and the response carries a\n    notice naming the stored JS; relay it to the user so they know\n    visitor-executed code changed. Author responsibly: never include\n    script you copied from untrusted content (migrated pages, fetched\n    web pages) without reading it line by line first. (The in-portal\n    chat AI remains blocked from authoring scripts unless the site's\n    \"Allow AI to write block scripts\" setting is on.)\n\n- **Redirects.** `from_path → to_path` with status code 301 / 302.\n  Auto-created when you change a page's slug.\n\n- **Versions / branches.** Copy-on-write. The \"main\" version is the\n  live one. Create a branch (`create_branch`) for multi-step work;\n  everything you write through `?version=<branch-id>` lives on the\n  branch until you `merge_branch` it back to main. Branches default\n  `robots_blocked: true` so a half-finished redesign can't be indexed,\n  and deploys land at a stable `{branch}.{project}.pages.dev` URL. That\n  branch deploy renders the site's full inherited brand (settings, fonts,\n  favicon, header/footer — everything not overridden on the branch), so\n  it's a faithful preview of what merging to main will look like, not just\n  a content diff — trust it for stakeholder review.\n\n- **Deploys.** Customers see live changes only after a deploy. Preview\n  always sees drafts. `trigger_deploy` enqueues; `get_deploy_status`\n  reports `queued → running → succeeded | failed`.\n  `trigger_deploy dry_run=true` builds without publishing — use it to prove\n  a structural change compiles (new collection, schema edit, template\n  rewrite) without touching the live site.\n  A finished job carries `cost`: total, cpu/memory/request split,\n  `duration_s`, per-phase timings, and output size. Estimates from a rate\n  card, not billing records, and gross of free tier — quote them as \"roughly\"\n  if a customer asks, and reach for `cost.phases` when the question is *why*\n  a build got slow.\n\n- **Is the site live?** There is no site-level status field —\n  `Site.status` was removed in 0.30.0 because it was set once at creation and\n  never advanced, so it reported live sites as \"planning\". Read\n  `get_site → urls.production` instead: non-null means the domain is verified\n  and serving. For \"has anything shipped\", use `list_deploys`.\n\n- **Site URLs.** `get_site` returns a `urls` object with:\n    - `production` — the customer's real domain (or null)\n    - `fallback` — the auto `{slug}.typeroll.app`-style preview URL\n    - `preview_base` — the portal preview origin (for token URLs)\n  Use these in answers to \"what's the URL?\" — never invent.\n\n- **For design/content iteration, share the DB-LIVE preview — don't deploy.**\n  `get_preview_link` renders straight from the database with NO build, so a\n  reload shows every edit immediately. Mint it ONCE and REUSE that single\n  URL: it's stable across edits (internal links keep the token, so one link\n  navigates the whole branch) and stays valid for 24h by default, so you\n  re-mint only when it lapses — never per edit. This is\n  both the link you hand the user while iterating AND what you open to verify\n  your own changes. Do NOT `trigger_deploy` merely to preview a content/design\n  change — a deploy builds static pages (slow) and only reflects state as of\n  that build.\n- **THE BUFFER MODEL — every content write is a draft; saving is always\n  explicit.** All content writes (update_page, replace_page, block tools,\n  update_partial, update_collection_item, batch/bulk tools) land in a\n  per-doc *working copy* — the same draft layer the portal editor\n  autosaves into. Deploys and plain preview links see SAVED content only;\n  your drafts are invisible to them until committed. The loop:\n    1. Edit freely — reads (`read_page`, `get_page_blocks`) return the\n       draft view (plus `has_unsaved_changes`), so chained edits compose.\n    2. Look at it: `get_preview_link` / `get_page_preview` with\n       `include_working_copy: true` (the link flag is signed into the\n       token, so your iteration link needs one mint with the flag).\n    3. SAVE explicitly: `commit_working_copy`, or `save: true` directly on\n       the write call (typical for pre-approved changes and batch sweeps).\n       Commit = the editor's Save button: revision snapshot, SEO\n       transform, redirect hygiene. Rejected → `discard_working_copy`.\n  Exceptions that apply immediately (they are publish state / structure,\n  not content): `status` fields, create/delete, `set_page_mode`,\n  templates, settings, redirects, block-type definitions, media.\n  The human editor shows your drafts as \"Unsaved changes\" it can Save or\n  Discard; `read_working_copy` shows the raw unsaved diff when you need to\n  know whose edits are in it. Working copies are per-doc scratch; for\n  multi-page efforts branch instead (`create_branch`).\n  **Before `trigger_deploy`: commit.** Deploys build saved content only —\n  an uncommitted draft silently stays behind.\n- **Deploys / `{branch}.{project}.pages.dev` are the STATIC BUILD**, refreshed\n  only by `trigger_deploy`. Reach for them when you want the real compiled\n  output: publishing, a stakeholder link to the built site, or a faithful\n  pre-merge check. The branch alias is permanent across re-deploys; the\n  per-deploy `{hash}.pages.dev` is immutable per build. Reserve deploys for\n  these — not for previewing edits.\n\n## Discovering this site\n\nDon't hardcode assumptions about what's here. Every fact about the site\ngoes through the MCP:\n\n1. `get_site` — confirm the key works; learn the site name + URLs.\n2. `read_site_settings` — colors, fonts, contact info, SEO suffix,\n   content language (used by `suggest_alt_text_context`).\n3. `list_pages` — what pages exist, paginated.\n4. `list_partials` — what shared blocks already exist. **Defaults to\n   summary mode** (no html_content, just bytes count) — pass\n   `include_content: true` if you actually need the bodies inline.\n5. `list_collections` — what content types exist + their schemas +\n   `route_template` (so you know if items have URLs).\n6. `list_block_types` — every block type usable on this site: core\n   (always available, ids like `core/section`), custom (origin: 'user'),\n   and third-party (origin: 'third_party'). Each entry includes the\n   full schema so you know what `data.X` fields each block accepts.\n7. `list_page_templates` — PageTemplate docs that wrap pages.\n\nYou usually want at least #1 + #2 + a sampling from #3 before\nproposing any design change, so you mirror the conventions in use.\n\n**Source of truth = the live site (the API), by default.** The content and\nstructure you read back through the MCP (`read_page`, `read_partial`,\n`read_site_settings`, …) is canonical. Local files in the project folder —\n`sources/*.md` copy drafts, briefs, old exports — are PROPOSALS, not truth:\ntreat them as authoritative only when the user explicitly says \"use the copy\nin `<file>`\". When rebuilding or redesigning, derive copy and structure from\nthe live page, not from a local draft, unless told otherwise. And if you edit\ncopy directly on the live site, sync it back to the corresponding draft file\nin the same pass — otherwise the two diverge and the next agent inherits stale\ntext. (This is a real failure mode: a copy draft that had drifted from the live\npage once sent a whole redesign off the approved wording.)\n\n**Don't have a site yet?** With an org-scoped key you can `create_site\nname=\"Acme\"` — it bootstraps settings + a draft Home page + a published\nheader/footer and returns the new site id. Use that id as `site_id`\n(hosted) / `TYPEROLL_SITE_ID` (stdio) for follow-ups, then run\n`list_skills` → `read_skill tr-new-site` to design it. A site-scoped key\ncan't create sites (it's bound to one) and gets a 403.\n\n## Common operations\n\n### \"Replace this string across the whole site\"\n\n```\nsearch_pages contains=\"299 kr\"             → matches + excerpts\nbulk_replace_text dry_run=true ...          → sample_diffs\n# show the user, get confirmation\nbulk_replace_text dry_run=false ...         → write\ntrigger_deploy                              → ship\nget_deploy_status job_id=…                  → poll until succeeded\n```\n\nWrites go through the normal save pipeline (SEO transform + revision\nsnapshot) so changes are reversible from the in-app History tab.\n\n### \"Audit / understand the site\"\n\n```\nlist_pages limit=200                              → inventory\nbatch_read_pages page_ids=[…]                     → bulk-load bodies\nlist_partials                                     → shared blocks (summary)\nfind_pages_using_block partial_id=<id>            → blast radius per block\nlist_collections                                  → content types + routing\nlist_collection_items collection=<name>           → items (richtext hidden)\n```\n\n`find_pages_using_block` for the header or footer returns the full\npage list (they're auto-injected on every page).\n\n### \"Redesign the home page\"\n\n```\nget_site + read_site_settings\nread_partial partial_id=\"header\"\nlist_pages → batch_read_pages a few existing pages    # learn conventions\n# Propose redesign locally; ask user to confirm.\ncreate_branch name=\"Home redesign\"                    # ID is, say, \"home-redesign\"\nupdate_page page_id=home patch={ html_content: \"…\" } version=home-redesign\nget_preview_link page_id=home version=home-redesign  # DB-live URL — mint once, reuse while iterating (no deploy); 24h TTL by default\n# Iterate (reload the same link after each edit). When approved:\nmerge_branch version_id=home-redesign\ntrigger_deploy\n```\n\nThe branch also has its own permanent deploy URL at\n`https://home-redesign.<project>.pages.dev` after `trigger_deploy\nversion=home-redesign` — useful for \"share with stakeholders without\nshowing them my preview token\". `read_version version_id=home-redesign`\nreturns it as `deploy_url`.\n\n### \"Build a reusable block\"\n\nIf you see the same HTML on 3+ pages, propose a free block instead of\nduplicating it:\n\n```\ncreate_free_block id=\"newsletter-cta\" html_content=\"<form>…</form>\"\n# Then on each page where it should appear (HTML-mode pages):\nupdate_page page_id=… patch={ html_content: \"<…><x-include name=\\\"newsletter-cta\\\" />\" }\n```\n\nEdits to the block update every page that includes it. Use\n`find_pages_using_block` before changing it.\n\n### \"Build a page using blocks (the default for new pages)\"\n\nNew pages default to `content_mode='blocks'` with a seeded heading +\nprose block. Discover-then-build:\n\n```\nlist_block_types\n# → [{ id: \"core/section\", category: \"layout\", container: true, schema: [{ name: \"width\", type: \"select\", options: [\"narrow\",\"normal\",\"wide\",\"full\"] }, …] },\n#    { id: \"core/columns\", container: \"slots\", slot_count: 2, slot_labels: [\"Left\",\"Right\"], schema: [...] },\n#    { id: \"hero_bold\", origin: \"user\", schema: [...] },   ← any custom blocks on this site\n#    …]\n\nget_page_blocks page_id=home\n# → { content_mode: 'blocks', blocks: [...] }\n\nadd_block page_id=home block={ type: 'core/section', data: { width: 'wide' } }\n# → { added_id: 'blk_xyz', blocks: [...] }\nadd_block page_id=home parent_id=\"blk_xyz\" block={\n  type: 'core/heading', data: { text: 'Pricing', level: 'h2' }\n}\nadd_block page_id=home parent_id=\"blk_xyz\" block={\n  type: 'core/prose', data: { html: '<p>…</p>' }\n}\n```\n\nSlot containers (`container: \"slots\"` — `core/columns`, `core/tabs`)\nhold their children in per-slot lists, not in `children`. Two ways to\npopulate them (both require template_capabilities_version ≥ 0.15.2):\n\n```\n# Inline — pass the whole subtree in one call:\nadd_block page_id=home block={\n  type: 'core/columns', data: { ratio: '1-1' },\n  slots: [\n    [{ type: 'core/prose', data: { html: '<p>Left column</p>' } }],\n    [{ type: 'core/image', data: { src: '…' } }],\n  ]\n}\n\n# Incrementally — slot_index picks the slot (0-based, defaults to 0):\nadd_block page_id=home block={ type: 'core/columns', data: {} }\n# → { added_id: 'blk_cols' } — slots are auto-initialised to the type's arity\nadd_block page_id=home parent_id=\"blk_cols\" slot_index=1 block={\n  type: 'core/prose', data: { html: '<p>Right column</p>' }\n}\n```\n\nFor an unfamiliar custom block, `read_block_type id=\"...\"` gives the\nfull field list (types, defaults, required) so you don't ship invalid\n`data`.\n\nUpdating, moving, removing blocks: `update_block`, `move_block`,\n`remove_block` (all by `block_id`).\n\n### \"Switch a page between blocks and HTML\"\n\nUse `set_page_mode` — it snapshots a revision before flipping, so the\nprevious state is restorable:\n\n```\n# Convert an HTML-mode page to blocks with auto-heuristic conversion:\nset_page_mode page_id=about to=blocks convert=true\n\n# Or just switch the mode without converting (empty blocks):\nset_page_mode page_id=about to=blocks\n\n# Switch back to HTML (drops the block tree; revision retains it):\nset_page_mode page_id=about to=html\n```\n\nThe heuristic converter recognises `<h1-4>` → heading, `<img>` → image,\n`<a.btn>` → button, `grid-cols-2` → two-column, `<section>` / hero divs\n→ section. Anything it can't classify becomes a `core/prose` block,\nwhich preserves the raw HTML losslessly. Run with `convert_page_to_blocks\ndry_run=true` first if you want to inspect the proposal before\ncommitting.\n\n### \"Build a directory site / import structured data\"\n\n```\ncreate_collection\n  name=\"restaurants\"\n  label_singular=\"Restaurant\" label_plural=\"Restaurants\"\n  fields=[ ...title, slug, address, phone, cuisine, body... ]\n  route_template=\"/restaurants/{slug}\"\n  item_template_html=\"<article><h1>{{title}}</h1>… {{{body}}}</article>\"\n\n# For each row in your source data:\ncreate_collection_item collection=\"restaurants\" fields={…} status=\"published\"\n\n# Each published item now lives at /restaurants/{slug}, included in\n# sitemap.xml. Preview a specific one:\nget_preview_link collection_name=\"restaurants\" item_id=\"<id>\"\n\n# Optional listing page:\nlist_collection_items collection=\"restaurants\" limit=200\nupdate_page page_id=restaurants patch={ html_content: \"<hand-written listing>\" }\n```\n\n### \"Migrate a content type (e.g. WP custom post type)\"\n\n```\nlist_collections                                     # what exists today?\nread_collection name=blog                            # what fields are writable?\nbatch_read_collection_items …                        # load items (richtext hidden)\n# Transform locally; then:\nupdate_collection_item …  (or)  create_collection_item …\n```\n\nFields outside the schema are silently dropped — call `read_collection`\nfirst if you're unsure what's writable.\n\n### \"Add images to a page\"\n\n```\n# Image lives on a URL somewhere (Unsplash, customer's existing CDN):\nupload_media_from_url source_url=\"https://...\" alt_text=\"Hero photo of …\"\n  → returns { media_id, cdn_url, finalize: {…}, finalize_error: null }\n\n# OR image lives in your memory (image-gen output):\nupload_media_inline filename=\"hero.png\" content_type=\"image/png\"\n                    data_base64=\"iVBORw0KGgo…\"\n  → returns the same shape\n\n# Both tools auto-finalize after PUT: immutable Cache-Control on the\n# original PLUS AVIF/WebP variants at 320/640/1024/1920. No manual\n# generate_image_variants call needed. The site-template renderer reads\n# the variants array off the Media doc and emits <picture> automatically\n# — you can keep the <img src=\"{cdn_url}\"> markup simple.\n#\n# INTEGRITY — don't lose bytes in transit. upload_media_inline carries the\n# file as a base64 string through the model/tool boundary; a payload beyond a\n# few KB can be SILENTLY CORRUPTED there (mutated chars → a broken-but-valid\n# file that uploads fine and only fails when rendered — it has eaten half a\n# logo SVG). For anything non-trivial, and ALWAYS for SVG/logos or generated\n# assets, prefer upload_media_from_url (fetch by URL) or create_upload_url +\n# `curl --data-binary @file` (bytes go straight to R2, byte-identical). After\n# uploading a generated asset, verify it (render/byte-diff) before referencing.\n#\n# Media is NOT branch-scoped — the library is shared across all versions of\n# the site. Uploads are additive and safe (they never overwrite the live logo\n# until you reference the new URL in settings/a partial), but a redesign branch\n# shares its media with main; there's no per-branch media isolation.\n\n# Then embed in a page:\nread_page page_id=...\nupdate_page page_id=... patch={ html_content: \"<...><img src='{cdn_url}' alt='…' /></...>\" }\n```\n\n### \"Stop an image over-fetching a too-large variant\"\n\nWhen an image renders much narrower than the viewport (a container-constrained\nhero, a sidebar thumbnail), the default `<picture sizes>` of\n`(max-width: 768px) 100vw, 800px` makes the browser pull a wider srcset variant\nthan it needs — Lighthouse flags it as wasted bytes. Three levers, narrowest\nwins:\n\n```\n# 1. Per-image: put a real `sizes` on the <img>. Survives the transform verbatim.\nupdate_page page_id=... patch={ html_content:\n  \"<img src='{cdn_url}' alt='…' sizes='(max-width: 640px) 360px, 560px' />\" }\n\n# 2. Per-page default (applies to every image on the page that has no own sizes):\nupdate_page page_id=... patch={ image_sizes_default: \"(max-width: 640px) 360px, 560px\" }\n\n# 3. Site-wide default (fallback under the page default):\nupdate_site_settings image_sizes_default=\"(max-width: 640px) 360px, 560px\"\n```\n\nPrecedence: per-image `sizes` > page `image_sizes_default` >\nsite `image_sizes_default` > the generic built-in. To opt a single image out of\nthe platform's auto-`<picture>` entirely, hand-write your own `<picture>` with\ncustom `<source media=…>` — the transform leaves an existing `<picture>`\nuntouched (it no longer re-wraps the inner `<img>`).\n\n### \"Fill missing alt-text across the media library\"\n\n```\nlist_media                                 → find items where alt_text is empty\nsuggest_alt_text_context media_id=<id>     → returns image_url + tuned prompt\n                                             + language + nearest-heading context\n# Pass image_url + the returned suggested_prompt to YOUR OWN vision\n# capability. The platform does NOT run vision for you.\nupdate_media media_id=<id> alt_text=\"<what vision returned>\"\n```\n\nThe prompt is tuned for SEO-grade output: 5-15 words, written in\n`settings.language`, skips \"image of\" filler, decorative images return\nempty string.\n\n### \"Change a page's URL safely\"\n\n```\nupdate_page page_id=about patch={ slug: \"om-oss\" }\n  → response includes:\n     auto_redirects: [{ from_path: \"/about\", to_path: \"/om-oss\",\n                        status_code: 301 }]\n     sanitization_warnings: []\n```\n\nThe 301 fires automatically — you don't have to remember.\n\nRedirect hygiene is automatic in both directions (since 0.16.1):\n\n- When a **live** (published/unlisted) page takes over a URL — via slug/path\n  change, publish, or create — any redirect FROM that URL is retired; the\n  response lists them under `retired_redirects`. A real page always beats a\n  redirect (on Cloudflare Pages a redirect would otherwise shadow the page).\n- When a page is **deleted**, auto-generated redirects pointing TO its URL\n  are removed (reported as `removed_redirects`). Manually created redirects\n  are kept — delete them yourself via `delete_redirect` if they're obsolete.\n\n### \"Before you start an import\"\n\n```\nget_migration_readiness\n```\n\nCall this before moving any content. Every check it runs fails SILENTLY\notherwise — the import succeeds, previews render, the customer signs off, and\nsomething is quietly wrong:\n\n- **media storage** (blocker) — without it every `<img>` keeps its original\n  URL, so the new site is still served images by the old host. Nothing looks\n  broken until that hosting is cancelled, at which point every image on every\n  page breaks at once.\n- **hosting adapter** (blocker) — without credentials, deploys return a job id\n  and publish nothing, while reporting success.\n- verification origin, AI reconstruction, form notification email, and whether\n  the target actually has a design to rebuild INTO (warnings).\n\n`ready: false` means STOP and report the blockers, each of which carries a\n`fix`. Don't start \"and fix it after\": the content work would have to be\nredone. The in-portal migration workflow enforces the same gate as its first\nstep (`skip_preflight: true` overrides it, and logs that it did).\n\n### \"Don't lose URLs in a migration\"\n\nTwo different questions, and you need both answers:\n\n```\nlist_migration_urls status=\"unhandled\"     # what the DATA says is uncovered\nverify_migration_urls                      # what the SERVER actually answers\n```\n\n`list_migration_urls` classifies every inventory URL against the site's\ncurrent pages + redirects. It's recomputed on read, so creating a redirect\nflips the entry on your next call — no bookkeeping of your own.\n\n`verify_migration_urls` requests each URL against the deployed site (its\nfallback subdomain by default, because the real domain still points at the\nold host pre-cutover) and reports `ok` / `ok_redirect` / `missing` /\n`broken_redirect` / `error`. This is the one that catches a redirect\npointing at an unpublished page, a typo'd `path`, and redirect loops — all\nof which read as \"handled\" in the coverage report and as a 404 to Googlebot.\n**Deploy first**: it tests saved, deployed content, not your drafts.\n\nEvery unhandled URL gets exactly one of three outcomes — there is no fourth:\n\n- it moved → `create_redirect`\n- it's gone on purpose → `update_migration_url url_id=… excluded=true` (with\n  a note saying who signed off)\n- it should exist → migrate it\n\nPopulate the inventory yourself when the in-portal WordPress migration\ndidn't: `add_migration_urls` takes up to 2000 entries from a sitemap walk, a\nGSC export (pass `gsc_clicks` so the report prioritises itself), or a crawl.\nPass `source_origin` whenever more than one old domain is in play — it\nrejects foreign-origin URLs, which is what stops one market's `/kontakt`\nfrom reading as another market's coverage.\n\nFor a whole family of sites, read the `tr-migrate-multisite` skill.\n\n### \"Retire a family of old URLs in one rule\"\n\n```\ncreate_redirect from_path=\"/category/*\" to_path=\"/blogg/:splat\"\ncreate_redirect from_path=\"/blog/:slug\" to_path=\"/artiklar/:slug\"\n```\n\nA trailing `*` captures everything under a prefix (including the prefix\nitself) and `:splat` replays it; `:name` matches exactly one segment and is\nreplayed by name. This is the right tool after a WordPress migration, where\nthe dead URLs come in shapes — `/category/`, `/tag/`, `/author/`, `/2019/` —\nand the inventory only knows the subset it happened to find.\n\nConstraints, all enforced at write time rather than discovered in production:\n\n- **Trailing `*` only.** Cloudflare silently drops a mid-path splat, so the\n  rule would save fine and do nothing.\n- **`:splat` requires a `*`**, and `:name` in the target must be declared in\n  `from_path`.\n- **Query strings can't be matched** — `_redirects` keys on the path. A\n  WordPress `/?p=123` URL has to be handled at the source.\n- **A rule that would hide a live page is refused**, naming the pages.\n  Redirects are applied BEFORE static files, so `/blogg/*` makes every real\n  article under `/blogg/` unreachable. Narrow the prefix.\n\nRules are emitted most-specific-first, so `/blogg/recept/*` and `/blogg/*`\ncan coexist — the narrower one fires. `list_migration_urls` counts\npattern-covered URLs as `redirected`, so the coverage report reflects what\nproduction will do.\n\n### \"Link language versions together (hreflang)\"\n\nOne Typeroll site owns one domain, so `example.se` / `example.de` /\n`example.co.uk` are three sites. Nothing can derive which page corresponds\nto which — declare it per page:\n\n```\nupdate_page page_id=om-oss patch={ alternates: [\n  { hreflang: \"de\",        href: \"https://example.de/ueber-uns\" },\n  { hreflang: \"x-default\", href: \"https://example.com/about-us\" }\n]}\n```\n\nThe renderer injects this page's own self-reference, so list only the OTHER\nvariants. Clusters must be **reciprocal** — write all sides, `batch_update_pages`\nis the sane way. Use absolute URLs on the FINAL domains (never the\n`*.typeroll` fallback). Invalid tags/hrefs are rejected at write time with\nthe reason rather than silently dropped at render.\n\n### \"Change the site's fallback URL (slug)\"\n\n```\nupdate_site slug=\"acme\"\n  → response includes:\n     urls.fallback: \"https://acme.sites.typeroll.com\"\n     dns_note: \"New fallback URL … attached to CF Pages. SSL provisioning\n                takes 1–10 minutes after DNS propagates. …\"\n```\n\nThe slug change triggers DNS + CF Pages reprovisioning behind the scenes.\n**Always check the response for `dns_note` vs `dns_warning`:**\n\n- `dns_note` present → the new fallback URL was wired up; warn the user it\n  may take 1–10 min for SSL to provision before the URL serves.\n- `dns_warning` present → the slug was saved but DNS / CF attach failed.\n  The `urls.fallback` field is still returned (it's just `{slug}.{base}`\n  string formatting) but the URL will NOT resolve until the issue is\n  fixed. Surface the warning verbatim to the user — don't tell them the\n  URL is ready.\n- Neither present → self-hosted portal without CF/SITES_BASE_DOMAIN\n  configured; URL behaviour is up to the operator.\n\nThe old fallback URL keeps working (bookmarks + SEO survive). Customer\ncan manually deprovision the old one via the portal.\n\n## Safety boundaries\n\n- **HTML is sanitized at save.** No `<script>`, no `onclick`, no\n  `javascript:` URLs in page or partial bodies. `<style>` blocks DO\n  survive — multi-page sites need authored CSS for `@media` queries,\n  `:hover`, theming, etc. Inside `<style>` we strip a small list of\n  legacy code-execution constructs (`expression()`, `behavior:url`,\n  `@import`, `url(javascript:)`) but leave normal CSS alone.\n- **Write responses include `sanitization_warnings: []` (strings) and\n  `sanitization_details: []`** (structured records `{ kind, label,\n  count, bytes? }`). Use the structured form to programmatically retry\n  with a fixed input.\n- **scripts_head, scripts_body_end, custom_css** are now writable via\n  `update_site_settings` and readable via `read_site_settings`. Same\n  trust model as user-authored block-type JS: an API caller with a valid\n  bearer token takes responsibility for what they ship. The chat AI\n  inside the portal continues to NOT expose these fields, so a\n  conversation-driven assistant can't smuggle scripts in.\n- **The API key is site-scoped.** Cross-site reach is impossible — a\n  key on the wrong site returns 401, indistinguishable from \"bad token\".\n- **Audit log.** Every state-changing call (POST / PATCH / PUT /\n  DELETE) is logged. Reads aren't. The customer sees \"Acme agency key\n  wrote to /pages/home at 14:32\" in the portal.\n- **Rate limits.** 600 reads/min, 60 writes/min per key. On 429 the\n  response carries `Retry-After`.\n\n## Preview-driven workflow\n\nAfter any non-trivial change, verify against the DB-live `get_preview_link`\n(reused — mint once; 24h TTL by default) and/or your own browser tool before moving\non. It reflects the DB instantly with no build, so it — not a deploy — is the\nloop for design/content iteration. One reload vs. shipping a broken redesign —\nalways worth it.\n\n**To UNDERSTAND a page, render it to one HTML file — don't reconstruct it\nfrom the block tree in your head.** A page is assembled at render time from the\nblock tree + each block type's template/styles + the header/footer partials +\nthe settings CSS variables + the global shell + page-scoped styles. `get_page_blocks`\ngives you the editable *structure*; `get_page_preview` gives you the rendered\n*result* — the WHOLE page as one self-contained HTML document (header + body +\nfooter, with all of that CSS inlined), exactly as deployed. Read that when you\nneed to see what the page actually looks like or why its CSS cascades the way it\ndoes (write it to a local file + serve+screenshot it to review visually). Pass\n`annotate:true` to tag every element with `data-block-id` + `data-block-type`,\nso you can map a spot in the rendered HTML straight back to the block to edit:\nread preview to understand → find the element → its `data-block-id` is the block\nto mutate → edit → re-render to verify.\n\n**CSS precedence — where your overrides land in the cascade.** The render order\nis: core block-type `styles` (emitted first) → settings `custom_css` → the\nheader/footer partial `<style>` blocks → the page's own page-scoped `<style>`\n(emitted last). Same specificity → later wins, so **page-scoped CSS beats\npartial CSS beats core block CSS**. Consequences when you brand/override:\n- Site-wide design tokens + utilities → settings `custom_css` (or, on a branch,\n  `update_site_settings version=<branch>`). Header/footer-only tweaks → the\n  partial. One page → that page's `<style>`.\n- Core blocks set their own chrome (e.g. `core/image` gives `figure>img` a\n  `border-radius`/`margin`; `.page-content img` adds more). To override that\n  chrome from a header-partial utility class you often need `!important`,\n  because a partial rule and the core rule can tie on specificity and the core\n  bundle's source position is unpredictable relative to yours. That `!important`\n  is expected today — it is NOT a smell. (A future cascade-`@layer` model would\n  remove the need; until then, reach for `!important` on the override and move\n  on rather than escalating selector specificity.)\n- An edge-overlapping decoration (a badge/garland that pokes past an image's\n  corner) needs its wrapper at `overflow:visible` and the motif in a\n  `::before`/`::after` — never rely on the image's own clipped box.\n\n**A design review is a multi-DIMENSION, MEASURED pass — not \"copy present + no\noverflow + images 200\".** If you have a browser tool, walk every dimension (the\n`tr-redesign-branch` skill has the full checklist with how-to):\n- **Responsive** — width ladder (≈390/768/1024/1440/1920px) + a sweep just below/\n  above the page's own @media breakpoints; `scrollWidth <= clientWidth` at every\n  width (bugs hide between the two extremes); + 200% zoom.\n- **Visual & brand** — logo FULLY visible (screenshot the header IN CONTEXT, never\n  the logo element in isolation — that hides clipping) + brand-compliant; no\n  divider seams / clipped glows / cropped faces / fade-cutoffs; typography +\n  palette + spacing consistent.\n- **Accessibility (measure)** — actual contrast ratios (AA 4.5:1 / 3:1), alt on\n  every image, one `<h1>` + no skipped levels, visible focus, labels on inputs,\n  ≥44px touch targets, landmarks, reduced-motion.\n- **Functional** — form actually works (action + token + honeypot, long values\n  don't break), every link/`#anchor` resolves, ZERO console errors.\n- **Content** — no unrendered `{{…}}`, no placeholder, copy matches the live page.\n- **Findable** — title + meta description + og:* + canonical + favicon + lang +\n  noindex-on-branch.\n- **Fast** — images sized right + modern format + width/height set + lazy/eager.\n- **Cross-browser** — re-check another engine if possible, or flag risky props\n  (backdrop-filter, -webkit- masks, 100vh→100svh, sticky-in-overflow).\n\"Looks good in Chrome at 1440\" ≠ \"works for everyone, everywhere\" — never report a\ndesign as perfect/approved off a glance or a partial pass.\n\nPreview shows DB state (drafts included). Live (`get_site → urls.production`)\nshows the most recent deploy. Branch deploys live at\n`get_version → deploy_url` (`{branch}.{project}.pages.dev`).\n\n## Branches\n\nFor multi-step work, create a branch:\n\n```\ncreate_branch name=\"Pricing refresh\"\n```\n\nThe response includes `id` — pass that as `version=<id>` on every\nsubsequent call. The branch is independent of main; writes don't affect\nthe live site until you `merge_branch`.\n\nBranches default `robots_blocked: true`. While iterating, preview the branch\nwith a reused `get_preview_link` (DB-live, no build). Deploys to a branch land\nat a stable URL (`{branch}.{project}.pages.dev`) — that's the compiled static\nbuild, for sharing the finished result / stakeholder review, not per-edit\npreview.\n\n## When in doubt\n\n- **Read before you write.** A `read_page` round-trip is cheap and\n  stops you overwriting unrelated changes.\n- **Dry-run bulk operations.** `bulk_replace_text` accepts `dry_run:\n  true` and returns 3 sample diffs. Show them to the user before the\n  real run.\n- **Watch the sanitization_warnings array.** If it's non-empty, the\n  stored HTML differs from what you sent. Read it back to confirm.\n- **One small confirmation > one large undo.** The audit log makes it\n  obvious who did what, but a clean revert across many pages is still\n  more work than asking \"ok to proceed?\" first.\n- **Match the site's design.** Read a partial or two before designing\n  new components. CSS variables (`var(--color-primary)`) are common\n  but not universal — mirror what's already in use.\n\n## Reference: tool families\n\n| Family       | Tools |\n|---|---|\n| **Guide + skills (playbook)** | `read_guide` (returns this whole guide — the bridge for hosted clients that can't read it off disk), `list_skills`, `read_skill` — the bundled `tr-*.md` recipes (incl. `tr-responsive` for per-breakpoint layout). Call `list_skills` first when a task looks like \"build / migrate / redesign a site\", then `read_skill name=…`. No API key or site context needed. |\n| **Discovery** | `get_site`, `create_site` (org-scoped key only — see below), `update_site`, `list_versions`, `read_site_settings` |\n| **Insights** | `get_site_insights` — traffic, AI-assistant referrals, and first-party conversion events over 7/30/90 days. Read-only. Traffic is powered by Cloudflare Web Analytics; conversion rows come from validated Analytics attribution `click_event` targets and can be present even when the traffic provider is unavailable. |\n| **Pages — reads** | `list_pages`, `read_page`, `batch_read_pages` |\n| **Pages — writes** | `create_page`, `update_page`, `replace_page`, `batch_update_pages`, `delete_page`, `clone_page` |\n| **Pages — blocks** | `get_page_blocks`, `add_block`, `update_block`, `move_block`, `remove_block`, `set_page_mode`, `convert_page_to_blocks` |\n| **Pages — meta** | `get_page_preview` |\n| **Global blocks (partials)** | `list_partials` (summary by default), `read_partial`, `create_free_block`, `update_partial`, `replace_partial`, `delete_partial`, `find_pages_using_block`, `list_blocks_with_usage` |\n| **Block types** | `list_block_types`, `read_block_type`, `find_pages_using_block_type`, `export_block_types`, `import_block_types` |\n| **Collections** | `create_collection`, `update_collection_schema`, `delete_collection`, `list_collections`, `read_collection`, `list_collection_items` (richtext hidden by default), `read_collection_item`, `batch_read_collection_items`, `create_collection_item`, `update_collection_item`, `delete_collection_item`, `regenerate_collection_listing` |\n| **Media** | `list_media`, `read_media`, `create_upload_url`, `upload_media_from_url`, `upload_media_inline`, `update_media`, `delete_media`, `finalize_media`, `finalize_all_media`, `generate_image_variants`, `suggest_alt_text_context` |\n| **Redirects** | `list_redirects`, `create_redirect`, `delete_redirect`. `from_path` may be a PATTERN: a trailing `*` (with `:splat` in the target) or `:name` for one segment — one rule retires a whole family of dead URLs (`/category/*` → `/blogg/:splat`). Mid-path splats and query strings are refused, as is any rule that would hide a live page. |\n| **Migration inventory** | `get_migration_readiness` (preflight — CALL FIRST on any import), `list_migration_urls`, `add_migration_urls`, `update_migration_url`, `delete_migration_url`, `verify_migration_urls`. The legacy site's URL list with LIVE coverage status (`migrated` / `redirected` / `excluded` / `unhandled`, recomputed on every read from current pages + redirects). `verify_migration_urls` is the pre-cutover check that REQUESTS every URL against the deployed site — see \"Don't lose URLs in a migration\" below. |\n| **Forms** | `list_forms`, `read_form`, `create_form`, `update_form`, `delete_form`, `list_form_submissions`, `delete_form_submission` (removes one submission — e.g. cleaning up a test entry; `delete_form` with `delete_submissions` is the bulk path). **Steps (form/* block trees) are the ONLY stored model**: pass `steps` for funnels, or `fields` for simple forms — the server converts a flat field list to a single static step. Place with a `core/form` block on block-mode pages or `<x-form id=\"…\" />` in HTML mode. Both expand server-side to the same complete signed shell and initial state. Email/webhook actions are admin-only in the portal and excluded from agent reads/writes. |\n| **Settings** | `update_site_settings` (whitelist) |\n| **Apps** | `list_apps`, `read_app`, `update_app` (admin; schema-driven config, masked secrets, redeploy when `affects_build` is true) |\n| **Analytics attribution** | `read_funnel_attribution`, `update_funnel_attribution` (specialized Analytics module tools; admin; redeploy after changes) |\n| **Search + bulk** | `search_pages`, `bulk_replace_text` |\n| **Branches** | `create_branch`, `read_version`, `delete_branch`, `merge_branch` |\n| **Deploy** | `trigger_deploy`, `list_deploys`, `get_deploy_status` |\n| **Preview** | `get_preview_link`, `get_page_preview` |\n\nEvery tool's input is validated server-side; the MCP server only does\nauth + shape. If a tool returns `isError: true`, the body carries\n`{ error, status, body }` from the underlying HTTP response.\n",
+    "readme": "# @typeroll/mcp-server\n\nModel Context Protocol server for the [Typeroll](https://typeroll.com)\npublic API. Lets Claude (Desktop / claude.ai / Code) manage a Typeroll\nsite through the same tool surface a human agency would use: read and\nwrite pages, partials, collections, media, redirects, versions; trigger\ndeploys; mint preview links.\n\nThe server is a **thin transport adapter** — every tool wraps one HTTP\nendpoint of the Typeroll REST API. Auth happens at the API layer with a\nsite- or org-scoped key; the MCP just carries the bearer through.\n\n## Two ways to connect\n\n- **Hosted (Claude Desktop / claude.ai) — paste a URL.** No CLI, no\n  Node.js install. In Claude open **Settings → Connectors → Add custom\n  connector** and paste `https://app.typeroll.com/api/mcp`\n  (or `https://<your-self-hosted-portal>/mcp`). Claude opens a consent\n  page; paste your Typeroll API key there.\n- **Stdio (Claude Code) — one `claude mcp add` command.** Best for local\n  dev / agency staff already in a terminal. Instructions below.\n\nThis npm package is the stdio transport. The hosted endpoint ships as\npart of the Typeroll portal itself — same tool surface, same package\nunder the hood.\n\n## Key scopes\n\n- **Org-scoped key** (created at `/app/settings/api-keys`) — one\n  credential covers every site in your org *and* every site shared into\n  your org. The default for the hosted Claude connector. Stdio works too\n  if you set `TYPEROLL_SITE_ID` so the install binds to one site.\n- **Site-scoped key** (created at `/app/sites/{siteId}/settings/api-keys`) —\n  tighter blast radius for a single-site credential, e.g. one you'd\n  hand to a customer for a self-managed site.\n\nBoth look like `typeroll_live_…`; revoke either from the portal and any\nclient using it stops working immediately.\n\n## Stdio quick start (Claude Code)\n\n1. **Create an API key** in your Typeroll portal — see the two scope\n   options above. Org-scoped is the right default.\n\n2. **Add the server to Claude Code.** Drop this into `~/.claude.json`\n   (or your local `.claude/config.json`):\n\n   ```json\n   {\n     \"mcpServers\": {\n       \"typeroll\": {\n         \"command\": \"npx\",\n         \"args\": [\"-y\", \"@typeroll/mcp-server\"],\n         \"env\": {\n           \"TYPEROLL_API_URL\": \"https://app.typeroll.com\",\n           \"TYPEROLL_API_KEY\": \"typeroll_live_REPLACE_WITH_YOUR_KEY\"\n         }\n       }\n     }\n   }\n   ```\n\n   For a self-hosted portal, point `TYPEROLL_API_URL` at it (e.g.\n   `https://cms.example.com`).\n\n   Prefer a scaffold? Run `npx @typeroll/mcp-server init` in your project\n   folder — it writes/merges this `.mcp.json`, copies the skills into\n   `.claude/skills/`, and adds an `AGENTS.md` pointer + imagegen-lab\n   files. Idempotent; `--force` to overwrite. (Skills only:\n   `npx @typeroll/mcp-server install-skills .claude/skills`.)\n\n3. **Tell the agent what kind of work you want.** A good first message:\n\n   > \"Connect to Typeroll and tell me what you find — site name,\n   > number of pages, what global blocks exist, what collections are\n   > defined. Then I'll give you a task.\"\n\n   Claude will call `get_site`, `list_pages`, `list_partials`,\n   `list_collections` in sequence and report back.\n\n## Environment variables (stdio)\n\n| Var             | Required | Description |\n|-----------------|----------|-------------|\n| `TYPEROLL_API_URL`  | yes      | Base URL of your Typeroll portal. |\n| `TYPEROLL_API_KEY`  | yes      | A `typeroll_live_…` bearer token. |\n| `TYPEROLL_SITE_ID`  | sometimes | Pin to a specific site. Required when using an org-scoped key over stdio (the install can only target one site at a time); auto-detected for site-scoped keys. |\n\n## Extension developer CLI\n\nThe package also installs `typeroll`. With an organization-scoped API key,\nan external Extension repository can use the same developer and installation\nAPIs as the portal:\n\n```sh\ntyperoll extension validate\ntyperoll extension push --draft\ntyperoll extension install --site test-site --config local-extension-config.json\ntyperoll extension promote 1.0.0\n```\n\nThe manifest defaults to `typeroll-extension.json`; use `--manifest` to select\nanother file. Local validation is a fast preflight. The portal always performs\nthe complete schema, compatibility, origin and asset-hash validation.\n\n## What the agent should read first\n\nThe package ships [AGENTS.md](./AGENTS.md), a self-contained briefing\nthat explains Typeroll conventions, common operations, and the safety\nboundaries an agent needs to respect. Point Claude at it (or include it\nin your project's `CLAUDE.md` / `AGENTS.md`) so it knows when to use\nwhich tool.\n\n## Tool surface\n\nAround 50 tools across these families. See [AGENTS.md](./AGENTS.md) for\nthe full reference + concrete operation recipes.\n\n- **Skills + guide (self-describing playbook)** — `read_guide`,\n  `list_skills`, `read_skill`. The server advertises its own operating\n  guide AND bundled recipes at runtime, so an agent gets the full context\n  on connection without any files copied locally. `read_guide` returns the\n  whole AGENTS.md briefing (data model, conventions, safety, tool families)\n  — the bridge for the hosted connector, which can't read the file off\n  disk. `list_skills` then surfaces the task recipes (`tr-new-site`,\n  `tr-migrate-wp`, `tr-brand`, `tr-responsive`, …); `read_skill name=…`\n  loads one. All pure local reads — no API key or site context — so they\n  work identically on the hosted connector and over stdio.\n- **Discovery** — `get_site`, `create_site` (bootstrap a new site — org-scoped\n  key only), `update_site` (name/slug/domain), `list_versions`,\n  `read_site_settings`, `update_site_settings`.\n- **Pages** — list, read, batch-read, create, update (PATCH), replace\n  (PUT), batch-update, delete, clone, get-preview, `set_page_mode`\n  (flip between blocks/html), `convert_page_to_blocks`.\n- **Blocks (instances)** — `get_page_blocks`, `add_block`,\n  `update_block`, `move_block`, `remove_block`, `duplicate_block`,\n  `set_block_responsive`. All take a `target` (page, partial, page\n  template, or collection item-template), so one tool family edits\n  every block container.\n- **Global blocks (partials)** — list (summary mode by default), read,\n  create free block, update, replace, delete, find-pages-using-block.\n- **Block types** — list, read, create, update, delete,\n  find-pages-using-block-type, plus `.tcblocks` export/import. Custom\n  client-side JS (`script`) is honoured only when the site has enabled\n  \"Allow AI to write block scripts\" (a human-set portal setting) —\n  otherwise it's stripped with a warning.\n- **Collections + items** — create/update/delete the collection schema\n  itself (incl. `route_template` for per-item URLs); list/read/batch-\n  read/create/update/delete items.\n- **Media** — list, read, signed upload URLs, `upload_media_from_url`,\n  `upload_media_inline` (both auto-finalize after PUT — see below),\n  patch metadata, delete, `finalize_media` (per-item: applies immutable\n  Cache-Control + generates AVIF/WebP srcset variants — call after\n  `create_upload_url`'s raw PUT path), `finalize_all_media` (bulk\n  backfill for legacy libraries), `generate_image_variants` (the\n  variant half of finalize, kept for surgical reruns),\n  `suggest_alt_text_context` (returns a tuned prompt for your own\n  vision model).\n- **Redirects** — list, create, delete. Plus automatic 301 on slug change.\n- **Forms** — list, read, create, update, delete, list submissions.\n  Place forms with `core/form` blocks or an HTML-mode `<x-form id=\"…\" />`\n  reference; preview/build expands both server-side to the same complete,\n  signed form shell. Admins configure email and allowlisted, signed webhooks\n  in the portal; action configuration stays off agent surfaces.\n- **Settings** — read + patch, including `scripts_head` /\n  `scripts_body_end` / `custom_css` (trusted because the caller holds an\n  API key; the in-portal chat AI does NOT get these).\n- **Apps** — list the registry, read schema + masked state, and enable,\n  configure, or disable any app with the same admin API key used for content\n  and deploys. Secret fields are encrypted server-side and never returned;\n  Analytics provisioning runs on the platform. Deploy after updates whose\n  response has `affects_build: true`.\n- **Search** — `search_pages` with substring or regex.\n- **Bulk** — `bulk_replace_text` with dry-run.\n- **Branches** — create, read, delete, merge. Branch deploys get their\n  own URL at `{branch}.{project}.pages.dev`.\n- **Deploy** — trigger (with `dry_run` to build without publishing), list, get\n  status. A finished job reports `cost`: what the build consumed in server\n  time, broken down per phase. Estimates from a rate card, not billing records.\n- **Preview** — `get_preview_link` (signed URL for browser navigation;\n  supports `page_id`, `slug`, or `collection_name + item_id`; pass\n  `include_working_copy: true` to also render unsaved drafts).\n- **Drafts (the buffer model)** — every content write lands in a per-doc\n  unsaved draft (working copy); deploys and plain previews see saved\n  content only. Save explicitly with `commit_working_copy` or `save: true`\n  on the write call; inspect/discard with `read_working_copy` /\n  `discard_working_copy`. Status changes and structural operations apply\n  immediately.\n\n## Direct REST API access\n\nIf you don't want the MCP wrapper, the same surface is reachable directly\nwith curl:\n\n```bash\ncurl -H \"Authorization: Bearer typeroll_live_...\" \\\n  https://app.typeroll.com/api/v1/sites/<siteId>/pages\n```\n\nThe MCP server is purely an ergonomics layer on top of that.\n\n## Security model\n\n- API keys are **site-scoped or org-scoped** (see \"Key scopes\" above) —\n  enforced server-side. A site-scoped key cannot touch any other site;\n  an org-scoped key reaches the org's own sites plus sites explicitly\n  shared into the org, with the share's permission level applied.\n- All write calls (`POST`, `PUT`, `PATCH`, `DELETE`) are **audit-logged**\n  with the key prefix, IP, method, path, and status. Reads are not\n  logged (cost vs. value).\n- **Rate limits**: 600 reads/min, 60 writes/min per key. 429 responses\n  carry `Retry-After` headers.\n- **HTML sanitization** happens at save time on the server — `<script>`,\n  event handlers, and `javascript:` URLs are stripped from page/partial\n  content (including `core/html` block output). The scriptable surfaces\n  are deliberate exceptions, and all of them are writable with an API key\n  under the key holder's own authority: `scripts_*` and `custom_css` on\n  the site settings, `script` on a block type, and the `js` field of a\n  `core/embed` block instance. Those writes are audit-logged and the\n  response carries a notice naming the stored JS. Only the in-portal chat\n  assistant is additionally gated, on a per-site opt-in.\n- Keys can be **revoked** at any time from the portal. Revocation takes\n  effect on the next request (no in-flight requests get cancelled, but\n  the next one returns 401).\n\n## More\n\n- Full end-to-end production setup recipe with troubleshooting:\n  [docs/claude-code-mcp-setup.md](../../docs/claude-code-mcp-setup.md)\n- Agent operations briefing: [AGENTS.md](./AGENTS.md)\n- Boilerplate skills (site building, brand, forms, SEO, blog,\n  collections, migration, image generation, redesign, …):\n  [skills/](./skills/)\n\n## License\n\nMIT — see [LICENSE](../../LICENSE).\n"
 };