@dogsbay/format-astro 0.2.0-beta.8 → 0.2.0-beta.81

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.
package/dist/project.js CHANGED
@@ -4,14 +4,29 @@
4
4
  * Takes ExportPage[] + NavItem[] and generates a complete Astro project
5
5
  * with static .astro pages using real Dogsbay components.
6
6
  */
7
- import { existsSync, mkdirSync, writeFileSync, readFileSync, cpSync, readdirSync, statSync, } from "node:fs";
7
+ import { existsSync, mkdirSync, writeFileSync, readFileSync, copyFileSync, cpSync, readdirSync, statSync, } from "node:fs";
8
8
  import { join, dirname, relative, resolve } from "node:path";
9
9
  import { fileURLToPath } from "node:url";
10
10
  import { treeToDogsbayMd } from "@dogsbay/format-dogsbay-md";
11
- import { treeToAstro } from "./serialize.js";
11
+ import { treeToAstro, TONE_CLASSES } from "./serialize.js";
12
12
  import { buildLlmsTxt, buildSectionLlmsTxt, buildLlmsFullTxt } from "./llms-txt.js";
13
- import { normalizeBasePath, basePathSegments, buildCurrentPath } from "./base-path.js";
14
- import { detectLeadingNodes } from "./lead.js";
13
+ import { buildSitemap, buildSitemapIndex } from "./sitemap.js";
14
+ import { normalizeBasePath, basePathSegments, buildCurrentPath, withBasePath, parseSiteUrl, combinePrefix, } from "./base-path.js";
15
+ /**
16
+ * Combined URL prefix = urlBase (Astro `base` from site.url path) +
17
+ * basePath (filesystem layout prefix). Every URL emitter (nav,
18
+ * sitemap, llms.txt, .md mirror, _headers, taxonomy) uses this for
19
+ * href output. Filesystem-layout consumers (mkdir, page output
20
+ * paths) keep using basePath alone — Astro's `base` config adds the
21
+ * urlBase prefix at route time.
22
+ *
23
+ * See plans/astro-base-from-site-url.md.
24
+ */
25
+ function combinedPrefix(options) {
26
+ const { urlBase } = parseSiteUrl(options.siteUrl);
27
+ return combinePrefix(urlBase, normalizeBasePath(options.basePath));
28
+ }
29
+ import { detectLeadingNodes, deriveDescription } from "./lead.js";
15
30
  /**
16
31
  * Recursively prefix all hrefs in a nav tree.
17
32
  * Turns `<basePath>/foo` into `<basePath>/{section}/foo`. The
@@ -49,6 +64,39 @@ function prefixNavHrefs(items, section, basePath) {
49
64
  function escapeRegex(s) {
50
65
  return s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
51
66
  }
67
+ /**
68
+ * Prefix the combined URL base onto root-absolute nav hrefs.
69
+ *
70
+ * Most importers emit nav hrefs already carrying the combined prefix —
71
+ * the dogsbay-md importer walks source paths through
72
+ * `fileToHref(file, hrefPrefix=combined)`, so OpenShift's `nav.yml`
73
+ * (`welcome/foo.md`) resolves straight to `/<base>/welcome/foo`. But an
74
+ * importer that produces a nav.json of *pre-resolved* hrefs at a
75
+ * different base — e.g. `dogsbay convert --from docusaurus --to
76
+ * dogsbay-md`, whose hrefs are root-absolute (`/about/foo`) — would
77
+ * otherwise emit unprefixed nav links that 404 under a site base, and
78
+ * break prev/next (pagination matches against the combined-prefixed
79
+ * `currentPath`).
80
+ *
81
+ * Running every nav href through `rewriteHref` here is the single
82
+ * chokepoint that makes nav base-correct for ALL importers. It's
83
+ * idempotent — hrefs already starting with the prefix (or external /
84
+ * anchor / protocol-relative) are left untouched — so importer-side
85
+ * prefixing keeps working unchanged, and an empty prefix is a no-op.
86
+ */
87
+ function prefixNavBaseHrefs(items, prefix) {
88
+ if (!prefix)
89
+ return items;
90
+ return items.map((item) => {
91
+ const result = { ...item };
92
+ if (result.href)
93
+ result.href = rewriteHref(result.href, prefix);
94
+ if (result.children) {
95
+ result.children = prefixNavBaseHrefs(result.children, prefix);
96
+ }
97
+ return result;
98
+ });
99
+ }
52
100
  /**
53
101
  * Rewrite internal hrefs in a page's tree nodes.
54
102
  * Prepends `prefix` to root-relative links (e.g. `/workers/...` → `/docs/workers/...`).
@@ -97,6 +145,55 @@ function rewriteHref(href, prefix) {
97
145
  return href;
98
146
  return prefix + href;
99
147
  }
148
+ /**
149
+ * Rewrite image srcs in inline nodes + raw HTML to include the
150
+ * combined URL prefix.
151
+ *
152
+ * Astro auto-prefixes `<a href>` and image imports going through
153
+ * `<AstroImage>`, but raw `<img src="...">` HTML in template
154
+ * output is left untouched. The serializer emits raw `<img>` for
155
+ * inline images and falls back to it for non-optimized block
156
+ * images, so we have to prefix manually before serialization to
157
+ * make `/_assets/...` paths resolve under subpath-mounted deploys
158
+ * (GH Pages project pages, multi-mount Cloudflare).
159
+ *
160
+ * Symmetric with rewriteTreeHrefs — same skip-rules (external,
161
+ * anchors, already-prefixed). Block images keep their prefix
162
+ * stripped back off for the `imageMap[...]` lookup key (see
163
+ * paragraphToAstro in serialize.ts) so Astro's image optimization
164
+ * still finds the source.
165
+ */
166
+ function rewriteTreeImageSrcs(nodes, prefix) {
167
+ for (const node of nodes) {
168
+ if (node.inline) {
169
+ rewriteInlineImageSrcs(node.inline, prefix);
170
+ }
171
+ if (node.html) {
172
+ node.html = rewriteHtmlImageSrcs(node.html, prefix);
173
+ }
174
+ if (node.children) {
175
+ rewriteTreeImageSrcs(node.children, prefix);
176
+ }
177
+ }
178
+ }
179
+ function rewriteInlineImageSrcs(nodes, prefix) {
180
+ for (const node of nodes) {
181
+ if (node.type === "image" && typeof node.src === "string") {
182
+ node.src = rewriteHref(node.src, prefix);
183
+ }
184
+ else if (node.type === "link") {
185
+ // Links wrap inline children (which may include images) — same
186
+ // recursion shape as rewriteInlineHrefs.
187
+ rewriteInlineImageSrcs(node.children, prefix);
188
+ }
189
+ else if (node.type === "highlight" && node.children) {
190
+ rewriteInlineImageSrcs(node.children, prefix);
191
+ }
192
+ }
193
+ }
194
+ function rewriteHtmlImageSrcs(html, prefix) {
195
+ return html.replace(/(<img\b[^>]*\ssrc=")(\/[^"]+)"/g, (_match, before, src) => `${before}${rewriteHref(src, prefix)}"`);
196
+ }
100
197
  /**
101
198
  * Build a `wrangler.jsonc` for Cloudflare Workers Static Assets.
102
199
  *
@@ -147,6 +244,125 @@ function buildWranglerConfig(siteName, options) {
147
244
  lines.push(`}`);
148
245
  return lines.join("\n") + "\n";
149
246
  }
247
+ /**
248
+ * Build the GitHub Actions workflow YAML for `actions/deploy-pages`.
249
+ *
250
+ * The workflow:
251
+ * 1. Checks out the repo on every push to the default branch.
252
+ * 2. Installs node + pnpm at the Astro project directory, runs
253
+ * `dogsbay site build` (via `pnpm dlx` since Dogsbay is a
254
+ * global CLI, not a project dep), then `pnpm run build`
255
+ * (which runs `astro build && pagefind`).
256
+ * 3. Uploads `<astroDirRel>/dist` as a Pages artifact via
257
+ * `actions/upload-pages-artifact`.
258
+ * 4. Deploys via `actions/deploy-pages`.
259
+ *
260
+ * `astroDirRel` is the path of the Astro output relative to the
261
+ * repo root (typically "astro" — the default config has
262
+ * `output: ./astro`). Empty string is allowed when the project is
263
+ * flat (outputDir === projectDir); the workflow degrades naturally
264
+ * by omitting the `defaults: working-directory` block.
265
+ *
266
+ * Author edits — extra build steps, secrets, deploy gating — survive
267
+ * subsequent `dogsbay site build` runs because the file is written
268
+ * write-if-missing (see emitDeployArtifacts). To start over, delete
269
+ * the workflow file and rebuild.
270
+ *
271
+ * Note on basePath: GitHub Pages serves project sites at
272
+ * `https://<user>.github.io/<repo>/`. Authors who want their docs at
273
+ * the repo root should set `site.basePath: /<repo-name>` (or empty
274
+ * for user/org pages). The platform's basePath plumbing handles all
275
+ * URL rewriting; this workflow doesn't need to know about it.
276
+ */
277
+ function buildGitHubPagesWorkflow(astroDirRel) {
278
+ // When the Astro output IS the project root, drop the working-
279
+ // directory block and reference cache + artifact paths without a
280
+ // prefix. This is the flat-layout case (rare for site-init flows;
281
+ // common for `dogsbay convert` outputs that get manually wired up).
282
+ const isFlat = astroDirRel === "" || astroDirRel === ".";
283
+ const workingDirBlock = isFlat
284
+ ? ""
285
+ : `
286
+ defaults:
287
+ run:
288
+ working-directory: ${astroDirRel}`;
289
+ const cacheDep = isFlat
290
+ ? "pnpm-lock.yaml"
291
+ : `${astroDirRel}/pnpm-lock.yaml`;
292
+ const artifactPath = isFlat ? "dist" : `${astroDirRel}/dist`;
293
+ return `# Deploy to GitHub Pages.
294
+ # Generated by \`dogsbay site init --deploy=github-pages\` (or by
295
+ # adding \`deploy: { target: github-pages }\` to dogsbay.config.yml
296
+ # and running \`dogsbay site build\`). Author edits survive every
297
+ # subsequent build — the file is never overwritten. To regenerate
298
+ # from template, delete the file and rebuild.
299
+ #
300
+ # Repo settings: Settings → Pages → Source = "GitHub Actions".
301
+ name: Deploy to GitHub Pages
302
+
303
+ on:
304
+ push:
305
+ branches: [main]
306
+ workflow_dispatch:
307
+
308
+ permissions:
309
+ contents: read
310
+ pages: write
311
+ id-token: write
312
+
313
+ # Allow only one concurrent deployment, skipping queued runs.
314
+ concurrency:
315
+ group: pages
316
+ cancel-in-progress: false
317
+
318
+ jobs:
319
+ build:
320
+ runs-on: ubuntu-latest${workingDirBlock}
321
+ steps:
322
+ - uses: actions/checkout@v4
323
+
324
+ - uses: pnpm/action-setup@v4
325
+ with:
326
+ version: 10
327
+
328
+ - uses: actions/setup-node@v4
329
+ with:
330
+ # Astro 6 requires Node ^20.19.5 || >=22.12.0; pin 22 for
331
+ # forward-compat (Node 20 LTS is fine for Astro 5 sites
332
+ # but the Dogsbay scaffold targets Astro 6).
333
+ node-version: 22
334
+ cache: pnpm
335
+ cache-dependency-path: ${cacheDep}
336
+
337
+ - name: Install dependencies
338
+ run: pnpm install --frozen-lockfile
339
+
340
+ # \`dogsbay\` is a global CLI, not a project dep — pnpm dlx
341
+ # fetches it on demand. To pin a version, replace with e.g.
342
+ # \`pnpm dlx dogsbay@0.2.0-beta.18 site build\`.
343
+ - name: Build with Dogsbay
344
+ run: pnpm dlx dogsbay@beta site build
345
+
346
+ - name: Build Astro site
347
+ run: pnpm run build
348
+
349
+ - name: Upload Pages artifact
350
+ uses: actions/upload-pages-artifact@v3
351
+ with:
352
+ path: ${artifactPath}
353
+
354
+ deploy:
355
+ needs: build
356
+ runs-on: ubuntu-latest
357
+ environment:
358
+ name: github-pages
359
+ url: \${{ steps.deployment.outputs.page_url }}
360
+ steps:
361
+ - name: Deploy to GitHub Pages
362
+ id: deployment
363
+ uses: actions/deploy-pages@v4
364
+ `;
365
+ }
150
366
  /**
151
367
  * Construct the SiteConfig object that gets serialized to
152
368
  * `src/data/site.json`. Backward-compatible: existing fields keep their
@@ -156,8 +372,6 @@ function buildSiteConfig(siteName, options) {
156
372
  const cfg = {
157
373
  siteName,
158
374
  repoUrl: options.repoUrl || "",
159
- editUri: options.editUri || "blob/main/docs/",
160
- copyright: options.copyright || "",
161
375
  };
162
376
  if (options.siteUrl)
163
377
  cfg.siteUrl = options.siteUrl;
@@ -169,6 +383,15 @@ function buildSiteConfig(siteName, options) {
169
383
  cfg.twitterHandle = options.twitterHandle;
170
384
  if (options.themeColor)
171
385
  cfg.themeColor = options.themeColor;
386
+ // editUri + copyright follow the same omit-on-empty pattern as the
387
+ // optional fields above; previously they were always written
388
+ // (editUri defaulted to "blob/main/docs/", copyright to ""), which
389
+ // left zombie config in src/data/site.json. Downstream guards already
390
+ // treat empty / undefined as "don't render" so this is purely a tidy.
391
+ if (options.editUri)
392
+ cfg.editUri = options.editUri;
393
+ if (options.copyright)
394
+ cfg.copyright = options.copyright;
172
395
  if (options.brandKeywords && options.brandKeywords.length > 0) {
173
396
  cfg.brandKeywords = options.brandKeywords;
174
397
  }
@@ -187,11 +410,58 @@ function buildSiteConfig(siteName, options) {
187
410
  }
188
411
  if (options.taxonomyIndexPaths &&
189
412
  Object.keys(options.taxonomyIndexPaths).length > 0) {
190
- cfg.taxonomyIndexPaths = options.taxonomyIndexPaths;
413
+ // Bake basePath into every emitted indexPath so consumers
414
+ // (TypeBadge / StatusBadge / future components) compose hrefs
415
+ // like `${indexPath}/<value>/` and resolve under the configured
416
+ // site base. Without the prefix, `/by-type/tutorial/` 404s on
417
+ // any site with `site.basePath` set. Caller passes raw config
418
+ // values (`/by-type`, `/tags`, etc.) — basePath threading is
419
+ // this emitter's responsibility, matching how `page.url` is
420
+ // already prefixed in the taxonomy data file.
421
+ // Taxonomy index paths are baked into site.json so components
422
+ // (TagList, TaxonomyIndex, TypeBadge) emit correct hrefs at
423
+ // runtime. Use combined so these resolve under the host's
424
+ // served subpath.
425
+ const taxoPrefix = combinedPrefix(options);
426
+ cfg.taxonomyIndexPaths = Object.fromEntries(Object.entries(options.taxonomyIndexPaths).map(([name, raw]) => [
427
+ name,
428
+ withBasePath(taxoPrefix, raw),
429
+ ]));
191
430
  }
192
431
  if (options.taxonomyDisplay &&
193
432
  Object.keys(options.taxonomyDisplay).length > 0) {
194
- cfg.taxonomyDisplay = options.taxonomyDisplay;
433
+ // Flatten prefix labels into top-level entries so the
434
+ // search-facets resolver finds them after DocsLayout splits
435
+ // slash-nested tags into per-prefix Pagefind filter divs.
436
+ //
437
+ // Input:
438
+ // tags.prefixes = { difficulty: { label, color }, ... }
439
+ // tags.labels = { "difficulty/1": "Beginner", "difficulty/2": ... }
440
+ // Output additions (kept alongside the original `tags` entry):
441
+ // difficulty.labels = { "1": "Beginner", "2": ... }
442
+ //
443
+ // Resolver does `display[facetName].labels[value]` — facet name
444
+ // is now `difficulty`, value is `1`, → "Beginner". See
445
+ // plans/per-prefix-search-facets.md.
446
+ const flat = { ...options.taxonomyDisplay };
447
+ const tagsDisplay = options.taxonomyDisplay.tags;
448
+ if (tagsDisplay?.prefixes) {
449
+ for (const prefix of Object.keys(tagsDisplay.prefixes)) {
450
+ if (flat[prefix])
451
+ continue; // top-level entry wins
452
+ const leafLabels = {};
453
+ if (tagsDisplay.labels) {
454
+ const needle = `${prefix}/`;
455
+ for (const [slug, label] of Object.entries(tagsDisplay.labels)) {
456
+ if (slug.startsWith(needle)) {
457
+ leafLabels[slug.slice(needle.length)] = label;
458
+ }
459
+ }
460
+ }
461
+ flat[prefix] = { labels: leafLabels };
462
+ }
463
+ }
464
+ cfg.taxonomyDisplay = flat;
195
465
  }
196
466
  return cfg;
197
467
  }
@@ -257,6 +527,74 @@ function ensureDirectoryStructure(outputDir, basePath) {
257
527
  export function emitSiteConfig(outputDir, siteName, options) {
258
528
  mkdirSync(join(outputDir, "src", "data"), { recursive: true });
259
529
  writeFileSync(join(outputDir, "src", "data", "site.json"), JSON.stringify(buildSiteConfig(siteName, options), null, 2));
530
+ // Auto-generated companion to astro.config.mjs. Carries the
531
+ // site/base values derived from dogsbay.config.yml's site.url so
532
+ // changes propagate without --force-rescaffolding the main
533
+ // astro.config.mjs (which is scaffold-once and may have author
534
+ // edits — custom integrations, build hooks, etc.). The main
535
+ // config imports `dogsbaySite` + `dogsbayBase` from here.
536
+ // See plans/astro-base-from-site-url.md.
537
+ const { origin, urlBase: astroBase } = parseSiteUrl(options.siteUrl);
538
+ const hasSiteUrl = Boolean(options.siteUrl && /^https?:\/\//.test(options.siteUrl));
539
+ const dogsbaySiteJson = hasSiteUrl
540
+ ? JSON.stringify(origin ?? options.siteUrl)
541
+ : "undefined";
542
+ const dogsbayBaseJson = astroBase ? JSON.stringify(astroBase) : "undefined";
543
+ // build.inlineStylesheets — defaults to "auto" (Astro's own
544
+ // default; matches our docs-first bias since theme.css is ~120KB
545
+ // and externalizing it lets the file cache cross-page). Authors
546
+ // wanting "always" / "never" set it via dogsbay.config.yml's
547
+ // build.inlineStylesheets. See docs/perf-tuning.md.
548
+ const dogsbayInline = options.inlineStylesheets ?? "auto";
549
+ writeFileSync(join(outputDir, "astro.config.dogsbay.mjs"), [
550
+ "// Auto-generated by `dogsbay site build` — DO NOT EDIT.",
551
+ "// Tracks site.url + derived Astro base + build behaviour from",
552
+ "// dogsbay.config.yml. Edit dogsbay.config.yml and rebuild;",
553
+ "// edits to this file will be overwritten on the next build.",
554
+ `export const dogsbaySite = ${dogsbaySiteJson};`,
555
+ `export const dogsbayBase = ${dogsbayBaseJson};`,
556
+ `export const dogsbayInlineStylesheets = ${JSON.stringify(dogsbayInline)};`,
557
+ "",
558
+ ].join("\n"));
559
+ // Migration check: pre-beta.20 sites have an astro.config.mjs that
560
+ // doesn't import the companion. Without the import, the values
561
+ // emitted above are unused and Astro's `base` stays unset — the
562
+ // exact bug this work was meant to close. Warn loudly, with the
563
+ // patch the user needs to apply, until astro.config.mjs is
564
+ // updated. We don't auto-patch because the file may have author
565
+ // edits (custom integrations, build hooks).
566
+ const astroConfigPath = join(outputDir, "astro.config.mjs");
567
+ if (existsSync(astroConfigPath)) {
568
+ const astroConfigSrc = readFileSync(astroConfigPath, "utf-8");
569
+ if (!astroConfigSrc.includes("astro.config.dogsbay.mjs")) {
570
+ console.warn([
571
+ "",
572
+ " ⚠ astro.config.mjs is missing the dogsbay companion import.",
573
+ " Without it, Astro's `base` config stays unset and assets",
574
+ " served from a host subpath (GH Pages project pages,",
575
+ " multi-mount Cloudflare) will 404.",
576
+ "",
577
+ " Add these two lines to astro.config.mjs:",
578
+ "",
579
+ ' import {',
580
+ ' dogsbaySite,',
581
+ ' dogsbayBase,',
582
+ ' dogsbayInlineStylesheets,',
583
+ ' } from "./astro.config.dogsbay.mjs";',
584
+ "",
585
+ " export default defineConfig({",
586
+ " ...(dogsbaySite ? { site: dogsbaySite } : {}),",
587
+ " ...(dogsbayBase ? { base: dogsbayBase } : {}),",
588
+ " build: { inlineStylesheets: dogsbayInlineStylesheets },",
589
+ " // ...your existing config...",
590
+ " });",
591
+ "",
592
+ " OR regenerate from template (overwrites your edits):",
593
+ " dogsbay site init . --scaffold-only --force",
594
+ "",
595
+ ].join("\n"));
596
+ }
597
+ }
260
598
  }
261
599
  export function emitSiteScaffold(outputDir, siteName, options, writeScaffold) {
262
600
  let scaffoldFilesSkipped = 0;
@@ -300,6 +638,7 @@ export function emitSiteScaffold(outputDir, siteName, options, writeScaffold) {
300
638
  };
301
639
  // Per-deploy-target additions to package.json
302
640
  const isCloudflare = options.deploy === "cloudflare-workers";
641
+ const isGitHubPages = options.deploy === "github-pages";
303
642
  const deployScripts = isCloudflare
304
643
  ? { deploy: "pnpm build && wrangler deploy" }
305
644
  : {};
@@ -325,14 +664,25 @@ export function emitSiteScaffold(outputDir, siteName, options, writeScaffold) {
325
664
  },
326
665
  dependencies: {
327
666
  astro: "^6.0.0",
328
- "@astrojs/sitemap": "^3.0.0",
667
+ // Sitemap is emitted directly by Dogsbay into
668
+ // public/<basePath>/sitemap-{index,0}.xml so multi-mount
669
+ // deploys don't collide at the host root. We deliberately
670
+ // do NOT depend on @astrojs/sitemap (it hardcodes output to
671
+ // dist/ root, which is what we're moving away from).
329
672
  // Pagefind is invoked from the build script (see scripts.build above).
330
673
  // Lives in dependencies (not devDependencies) so production builds
331
674
  // include it; the produced search index is shipped statically and
332
675
  // doesn't load this dep at runtime.
333
676
  pagefind: "^1.4.0",
334
677
  tailwindcss: "^4.0.0",
335
- "@tailwindcss/vite": "^4.0.0",
678
+ // Pinned to 4.2.x — `@tailwindcss/vite` 4.3.x ships an
679
+ // oxcResolvePlugin shape that breaks Astro 6's
680
+ // rolldown-vite ("Missing field tsconfigPaths in
681
+ // oxcResolvePlugin"). Surfaced during the FastAPI import
682
+ // (~150-page MkDocs site) on a fresh `dogsbay site init`.
683
+ // Drop the ~ when Astro 6 picks up a compatible rolldown
684
+ // build OR @tailwindcss/vite restores the prior shape.
685
+ "@tailwindcss/vite": "~4.2.2",
336
686
  "tailwind-variants": "^0.3.0",
337
687
  shiki: "^4.0.0",
338
688
  "@shikijs/transformers": "^4.0.0",
@@ -348,6 +698,15 @@ export function emitSiteScaffold(outputDir, siteName, options, writeScaffold) {
348
698
  "@dogsbay/primitives": dogsbayDep("primitives"),
349
699
  "@dogsbay/icons": dogsbayDep("icons"),
350
700
  "@dogsbay/elements": dogsbayDep("elements"),
701
+ // Transitive of `@dogsbay/primitives` (via
702
+ // `@floating-ui/dom`). Listed at the top level because
703
+ // npm doesn't hoist the second-level transitive when
704
+ // `@dogsbay/primitives` is linked via `file:` (the
705
+ // `--local` monorepo mode + the canary publish flow on
706
+ // GH Pages CI both hit this). Surfaced during the
707
+ // FastAPI import: Rollup failed with "Cannot resolve
708
+ // @floating-ui/core" at astro build time.
709
+ "@floating-ui/core": "^1.7.0",
351
710
  },
352
711
  // Pin transitive Vite to 7. Vite 8 just released; Astro 6
353
712
  // peer-deps Vite 7 and prints a warning when 8 is hoisted.
@@ -374,6 +733,18 @@ export function emitSiteScaffold(outputDir, siteName, options, writeScaffold) {
374
733
  scaffoldFilesSkipped++;
375
734
  }
376
735
  }
736
+ // GitHub Pages deploy artifacts — workflow + .nojekyll. The actual
737
+ // emission lives in `emitDeployArtifacts` so site-build can also
738
+ // call it on existing sites without going through scaffold (a user
739
+ // adds `deploy: github-pages` to dogsbay.config.yml and reruns
740
+ // `site build` to get the workflow). At scaffold-time we pass
741
+ // forceOverwrite=writeScaffold so `--force` regenerates from
742
+ // template; on regular builds it stays write-if-missing.
743
+ if (isGitHubPages) {
744
+ emitDeployArtifacts(outputDir, options, {
745
+ forceOverwrite: writeScaffold,
746
+ });
747
+ }
377
748
  // Generate astro.config.mjs
378
749
  // `preserveSymlinks: true` is used with --local to pin local file: deps to
379
750
  // their on-disk paths. Inside a pnpm workspace this breaks Astro's internal
@@ -385,52 +756,50 @@ export function emitSiteScaffold(outputDir, siteName, options, writeScaffold) {
385
756
  preserveSymlinks: true,
386
757
  },`
387
758
  : "";
388
- // Sitemap integration is conditional: requires an absolute site URL so
389
- // <loc> entries can be properly absolute. Without siteUrl, the sitemap
390
- // step is skipped (the import + integration call are simply omitted from
391
- // the generated config). Sitemap also filters out frontmatter-noindex pages.
392
- const hasSiteUrl = Boolean(options.siteUrl && /^https?:\/\//.test(options.siteUrl));
393
- const sitemapImport = hasSiteUrl ? `import sitemap from "@astrojs/sitemap";\n` : "";
394
- // Strip any path component from site.url before emitting. The
395
- // config validator already rejects `site.url` containing a path
396
- // when `basePath` is non-empty (canonical URLs would double-count
397
- // the prefix); this is a defensive normalisation for the case
398
- // where the validator is bypassed or basePath is empty.
759
+ // siteUrl gates absolute-URL emission (sitemap <loc> entries,
760
+ // canonical tags). Without one, both are skipped relative URLs
761
+ // are still correct, the sitemap is just not generated.
399
762
  //
400
- // Note: we deliberately do NOT emit Astro's `base:` field. With
401
- // the current file emission (pages live under
402
- // `src/pages/<basePath>/...`), adding `base` would cause Astro
403
- // to doubly-prefix every route. Switching to `base`-driven
404
- // routing is a separate refactor — see plans/configurable-base-path.md.
405
- let siteField = "";
406
- if (hasSiteUrl) {
407
- let originOnly;
408
- try {
409
- const u = new URL(options.siteUrl);
410
- originOnly = `${u.protocol}//${u.host}`;
411
- }
412
- catch {
413
- originOnly = options.siteUrl;
414
- }
415
- siteField = `\n site: ${JSON.stringify(originOnly)},`;
416
- }
417
- const integrationsField = hasSiteUrl ? `\n integrations: [sitemap()],` : "";
418
- // astro.config.mjs scaffold-once. Maintainer adds custom integrations.
419
- // The plugin-aliases import is for the Dogsbay plugin API: each
420
- // build emits `astro.config.plugins.mjs` exporting `pluginAliases`,
421
- // a Vite alias map for `virtual:dogsbay-plugin-config/<id>` modules.
422
- // When no plugins use defineClientConfig the map is empty and the
423
- // spread is a no-op. See plans/plugin-api.md.
763
+ // Sitemap is emitted directly by Dogsbay (see emitSitemapFiles)
764
+ // into public/<basePath>/sitemap-*.xml. We deliberately do NOT
765
+ // wire @astrojs/sitemap here; that integration hardcodes output
766
+ // to dist/ root, breaking multi-mount deploys.
767
+ const hasSiteUrl = Boolean(options.siteUrl && /^https?:\/\//.test(options.siteUrl));
768
+ // site.url's path component (if any) becomes Astro's `base`. The
769
+ // origin alone goes into `site`. This split lets dogsbay model
770
+ // both axes independently:
771
+ // - Astro's `base` (= urlBase) controls the URL prefix Astro
772
+ // bakes into HTML asset references (`<basePath>/_astro/...`)
773
+ // and the routes Astro generates from src/pages.
774
+ // - dogsbay's basePath controls the filesystem layout
775
+ // (`src/pages/<basePath>/...`).
776
+ // The two compose at emit time — combining for nav hrefs,
777
+ // sitemap, llms.txt, etc. See plans/astro-base-from-site-url.md.
778
+ const { origin, urlBase: astroBase } = parseSiteUrl(options.siteUrl);
779
+ // astro.config.mjs — scaffold-once, but the site/base values flow
780
+ // through a separate auto-generated file (`astro.config.dogsbay.mjs`,
781
+ // emitted unconditionally below) so dogsbay-derived values stay in
782
+ // sync with `dogsbay.config.yml` even on existing sites where the
783
+ // main config is preserved. Same pattern as
784
+ // `astro.config.plugins.mjs` the import line is the load-bearing
785
+ // bit; the auto-file is what changes.
424
786
  if (writeScaffold) {
425
787
  writeFileSync(join(outputDir, "astro.config.mjs"), `import { defineConfig } from "astro/config";
426
788
  import tailwindcss from "@tailwindcss/vite";
427
- ${sitemapImport}import { pluginAliases, pluginFsAllow } from "./astro.config.plugins.mjs";
789
+ import { pluginAliases, pluginFsAllow } from "./astro.config.plugins.mjs";
790
+ import {
791
+ dogsbaySite,
792
+ dogsbayBase,
793
+ dogsbayInlineStylesheets,
794
+ } from "./astro.config.dogsbay.mjs";
428
795
 
429
- export default defineConfig({${siteField}
796
+ export default defineConfig({
797
+ ...(dogsbaySite ? { site: dogsbaySite } : {}),
798
+ ...(dogsbayBase ? { base: dogsbayBase } : {}),
430
799
  output: "static",
431
800
  build: {
432
- inlineStylesheets: "always",
433
- },${integrationsField}
801
+ inlineStylesheets: dogsbayInlineStylesheets,
802
+ },
434
803
  vite: {
435
804
  plugins: [tailwindcss()],
436
805
  resolve: {
@@ -452,6 +821,9 @@ export default defineConfig({${siteField}
452
821
  else {
453
822
  scaffoldFilesSkipped++;
454
823
  }
824
+ // astro.config.dogsbay.mjs is emitted by emitSiteConfig (called
825
+ // above and on every site build) so site/base values stay in
826
+ // sync without a re-scaffold. See its definition for rationale.
455
827
  // Always seed an empty astro.config.plugins.mjs so the import in
456
828
  // astro.config.mjs resolves before the first plugin-emitting
457
829
  // build. Subsequent builds replace it via emitPluginRuntime.
@@ -522,7 +894,12 @@ export default defineConfig({${siteField}
522
894
  */
523
895
  export async function emitAstroPages(pages, nav, outputDir, options) {
524
896
  const siteName = options.siteName || "Documentation";
897
+ // basePath = filesystem layout prefix (where pages live under
898
+ // src/pages/...). combined = the URL prefix HTML hrefs need
899
+ // (urlBase + basePath). The two diverge whenever site.url has a
900
+ // path component (GH Pages project pages, multi-mount Cloudflare).
525
901
  const basePath = normalizeBasePath(options.basePath);
902
+ const combined = combinedPrefix(options);
526
903
  const baseSegments = basePathSegments(basePath);
527
904
  // Ensure dirs exist (callers may invoke us without going through the
528
905
  // full exportAstroProject orchestrator, e.g. dogsbay convert at Step 7).
@@ -543,13 +920,36 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
543
920
  // Remove existing entry for this section (full replace)
544
921
  existingNav = existingNav.filter((item) => item.label?.toLowerCase() !== siteName.toLowerCase()
545
922
  && item.label?.toLowerCase() !== section.toLowerCase());
546
- const prefixedNav = prefixNavHrefs(nav, section, basePath);
923
+ // Nav hrefs already carry the `combined` prefix (the importer
924
+ // emits them via fileToHref(file, hrefPrefix=combined)).
925
+ // prefixNavHrefs takes the existing prefix and weaves a section
926
+ // segment into it.
927
+ const prefixedNav = prefixNavHrefs(nav, section, combined);
547
928
  const sectionLabel = siteName
548
929
  || section.split("-").map(w => w.charAt(0).toUpperCase() + w.slice(1)).join(" ");
549
930
  existingNav.push({ label: sectionLabel, children: prefixedNav });
550
931
  outputNav = existingNav;
551
932
  }
933
+ // Safety net: ensure every nav href carries the combined URL base,
934
+ // regardless of which importer produced the nav. Idempotent — a nav
935
+ // already prefixed at import time (dogsbay-md walking source paths) is
936
+ // left untouched; a nav of pre-baked root-absolute hrefs (docusaurus
937
+ // convert) gets the base it was missing. Keeps nav hrefs aligned with
938
+ // the combined-prefixed `currentPath` so prev/next still matches.
939
+ outputNav = prefixNavBaseHrefs(outputNav, combined);
552
940
  writeFileSync(join(outputDir, "src", "data", "nav.json"), JSON.stringify(outputNav, null, 2));
941
+ // Also publish nav.json under public/_dogsbay/ so the
942
+ // client-mode <DocsNavClient /> can fetch it at runtime
943
+ // (Astro copies public/ to dist/ as-is at build). The src/data
944
+ // copy stays for build-time imports (pagination, prev/next
945
+ // calculation in each page) — the public copy is the on-the-wire
946
+ // copy that the browser ever sees. Kept identical (same bytes,
947
+ // same shape); the duplicate is cheap (one file) and keeps the
948
+ // build-time and runtime worlds cleanly separated. See
949
+ // plans/client-rendered-nav.md.
950
+ const publicNavDir = join(outputDir, "public", "_dogsbay");
951
+ mkdirSync(publicNavDir, { recursive: true });
952
+ writeFileSync(join(publicNavDir, "nav.json"), JSON.stringify(outputNav));
553
953
  // Static assets (images etc.) — content-tier; always copy from the
554
954
  // user's source dir. If they removed an asset, we want it gone here
555
955
  // too. Skipped when sourceDir isn't supplied (programmatic callers
@@ -557,21 +957,93 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
557
957
  if (options.sourceDir) {
558
958
  copyAssets(options.sourceDir, outputDir, options.imageOptimization);
559
959
  }
960
+ // External asset mounts (e.g. a Docusaurus `static/` dir) → public/_assets/,
961
+ // preserving internal structure so the importer's `/_assets/...` image refs
962
+ // resolve. Used by the convert --to astro path; the dogsbay-md → site build
963
+ // path instead lands these under content/_assets and copyAssets picks them up.
964
+ if (options.assetMounts) {
965
+ for (const mount of options.assetMounts) {
966
+ if (existsSync(mount.dir)) {
967
+ cpSync(mount.dir, join(outputDir, "public", "_assets"), { recursive: true });
968
+ }
969
+ }
970
+ }
560
971
  let generated = 0;
972
+ // Tracks whether a real root home page (`content/index.md` → slug
973
+ // "index") was emitted. Drives the root-served-site redirect fallback
974
+ // below: a corpus whose nav root is e.g. `welcome/index` (OpenShift)
975
+ // has no `/` page, so without a redirect `/` 404s. See
976
+ // plans/dir-index-slug-nav-drop.md.
977
+ let rootIndexEmitted = false;
978
+ const generatedPaths = new Set();
561
979
  const pagesDir = join(outputDir, "src", "pages", ...baseSegments);
562
980
  const useImageOpt = options.imageOptimization ?? false;
563
- // hrefPrefix is the same string as basePath. rewriteHref handles the
564
- // empty-basePath case correctly: any link starting with "/" matches
565
- // the early-return guard, so root-relative links pass through
566
- // unrewritten when the site is served at host root.
567
- const hrefPrefix = basePath;
981
+ // hrefPrefix is the COMBINED prefix (urlBase + basePath) what
982
+ // rendered HTML hrefs need so internal links resolve under the
983
+ // host's served subpath AND under the dogsbay basePath. For
984
+ // simple host-apex deploys with basePath, urlBase is empty so
985
+ // combined === basePath (back-compat). For GH Pages project pages
986
+ // and multi-mount Cloudflare, combined adds the urlBase layer.
987
+ const hrefPrefix = combined;
988
+ // Route-exclusion gate. Two signals, either is enough:
989
+ // - frontmatter `_fragment: true` (loader-stamped — the importer
990
+ // knew this .md was include-only, not a navigable page)
991
+ // - excludeFromRoutes match against the slug (project-declared,
992
+ // in dogsbay.config.yml — for content under conventional
993
+ // fragment dirs like `modules/`, `_attributes/`, `snippets/`)
994
+ // Excluded pages stay on disk under content/ (so includes resolve)
995
+ // but don't produce .astro / .md.ts routes. See plans/build-at-scale.md.
996
+ //
997
+ // Match semantics:
998
+ // - A single-segment pattern (no `/`) matches ANY occurrence of
999
+ // that segment in the slug path. So `modules` excludes both
1000
+ // `modules/foo` AND `welcome/modules/foo` AND
1001
+ // `drupal-build/openshift-enterprise/ai/includes/x`. This is
1002
+ // what AsciiBinder corpora want — fragment dirs symlink into
1003
+ // section dirs so the same `includes/` etc. appears at many
1004
+ // depths.
1005
+ // - A multi-segment pattern (contains `/`) matches as a leading
1006
+ // prefix (exact path-prefix). `welcome/_internal` excludes
1007
+ // only that specific path tree, not arbitrary `_internal/`
1008
+ // elsewhere.
1009
+ const excludePatterns = (options.excludeFromRoutes ?? []).map((p) => p.replace(/^\/+/, "").replace(/\/+$/, ""));
1010
+ function isExcludedSlug(slug) {
1011
+ if (excludePatterns.length === 0)
1012
+ return false;
1013
+ const segments = slug.split("/");
1014
+ for (const pattern of excludePatterns) {
1015
+ if (pattern.includes("/")) {
1016
+ // Multi-segment → leading-prefix match.
1017
+ if (slug === pattern || slug.startsWith(pattern + "/"))
1018
+ return true;
1019
+ }
1020
+ else {
1021
+ // Single-segment → any-segment match.
1022
+ if (segments.includes(pattern))
1023
+ return true;
1024
+ }
1025
+ }
1026
+ return false;
1027
+ }
568
1028
  for (const page of pages) {
569
1029
  try {
1030
+ // Skip excluded pages before any expensive work (tree rewrite,
1031
+ // serialize, IO).
1032
+ const isFragment = page.frontmatter && page.frontmatter._fragment === true;
1033
+ if (isFragment || isExcludedSlug(page.slug)) {
1034
+ continue;
1035
+ }
570
1036
  // Rewrite internal hrefs to match the output URL structure
571
1037
  rewriteTreeHrefs(page.tree, hrefPrefix);
1038
+ // Same for raw image srcs — Astro doesn't auto-prefix
1039
+ // `<img src="/_assets/...">` so we do it here. Block images
1040
+ // strip the prefix back off for the `imageMap[...]` lookup
1041
+ // (see paragraphToAstro in serialize.ts).
1042
+ rewriteTreeImageSrcs(page.tree, hrefPrefix);
572
1043
  const result = treeToAstro(page.tree, {
573
1044
  imageOptimization: useImageOpt,
574
1045
  codeBlockTitle: options.codeBlockTitle ?? true,
1046
+ combinedPrefix: hrefPrefix,
575
1047
  });
576
1048
  const imageSetup = useImageOpt ? [
577
1049
  '',
@@ -585,11 +1057,32 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
585
1057
  ' imageMap[publicPath] = mod.default;',
586
1058
  '}',
587
1059
  ] : [];
588
- // Per-page meta from frontmatter
1060
+ // Per-page meta from frontmatter. When no description is authored,
1061
+ // derive one from the first prose paragraph so every page emits a
1062
+ // <meta name="description"> (Lighthouse flags pages without one).
1063
+ // See plans/auto-meta-description.md.
589
1064
  const fm = (page.frontmatter ?? {});
590
- const pageDescription = fm.description ?? "";
1065
+ // A writer-authored frontmatter description is meta-only (not in the body),
1066
+ // so it's safe to also render as a visible lede. A DERIVED description is
1067
+ // pulled FROM the body, so rendering it as a lede always duplicates content
1068
+ // (e.g. a page whose first paragraph sits under "## Big picture"). Keep the
1069
+ // two separate: derive for the <meta> tag, but only lede the authored one.
1070
+ const fmDescription = typeof fm.description === "string" && fm.description.trim().length > 0
1071
+ ? fm.description
1072
+ : undefined;
1073
+ const pageDescription = fmDescription || deriveDescription(page.tree) || "";
591
1074
  const pageOgImage = fm.ogImage ?? "";
592
- const pageNoindex = fm.noindex === true || fm.draft === true;
1075
+ // Noindex / nofollow are independent meta directives. Site-level
1076
+ // forces both bits site-wide (staging / compliance lockdown);
1077
+ // page frontmatter can ESCALATE either bit independently but
1078
+ // cannot opt out of a site-level lockdown. `draft: true` keeps
1079
+ // its existing role as a noindex shorthand. See
1080
+ // plans/site-level-robots-meta.md.
1081
+ const pageNoindex = options.noindex === true ||
1082
+ fm.noindex === true ||
1083
+ fm.draft === true;
1084
+ const pageNofollow = options.nofollow === true ||
1085
+ fm.nofollow === true;
593
1086
  // Independent of noindex: pages can be excluded from in-site
594
1087
  // Pagefind search even when external SEs should index them
595
1088
  // (or vice versa). See DocsLayout's prop docs for the
@@ -608,7 +1101,28 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
608
1101
  const pageCategory = Array.isArray(pageMeta?.category)
609
1102
  ? pageMeta.category
610
1103
  : undefined;
611
- const tagsIndexPath = options.tagsIndexPath ?? "/tags";
1104
+ // Custom-taxonomy values lifted from frontmatter into
1105
+ // `meta.taxonomies` by the importer (see `parseMeta` in
1106
+ // `@dogsbay/types`). Surfaced to DocsLayout so it can emit one
1107
+ // `<div data-pagefind-filter="<name>:<value>">` per entry — this
1108
+ // is what makes user-declared taxonomies (`difficulty`, `team`,
1109
+ // anything not in the five built-ins) appear as visible facet
1110
+ // checkboxes in the search dialog. Without this passthrough
1111
+ // they're silently dropped after the importer.
1112
+ const pageTaxonomies = pageMeta?.taxonomies && Object.keys(pageMeta.taxonomies).length > 0
1113
+ ? pageMeta.taxonomies
1114
+ : undefined;
1115
+ // `tagsIndexPath` flows to `<TagList>` for chip hrefs
1116
+ // (`${indexPath}/${tag}/`). Caller passes the raw config value
1117
+ // (e.g. `/tags`); we bake the COMBINED prefix (urlBase from
1118
+ // site.url's path + basePath) here so chips resolve under both
1119
+ // the host's served subpath AND the dogsbay basePath. With
1120
+ // basePath alone, chips 404 on GH Pages project deploys
1121
+ // (basePath="" + non-empty urlBase) — same shape as the
1122
+ // typeBadgeHref / statusBadgeHref composition in DocsLayout,
1123
+ // which already reads combined-prefixed values out of
1124
+ // siteConfig.taxonomyIndexPaths (baked in buildSiteConfig).
1125
+ const tagsIndexPath = withBasePath(combined, options.tagsIndexPath ?? "/tags");
612
1126
  // Auto-lede detection. If the markdown body doesn't already
613
1127
  // start with an H1 / leading paragraph, we ask DocsLayout to
614
1128
  // render the frontmatter title / description at the top of
@@ -622,7 +1136,7 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
622
1136
  page.title.length > 0;
623
1137
  const autoLede = !isRedirect &&
624
1138
  !bodyHasLede &&
625
- pageDescription.length > 0;
1139
+ fmDescription !== undefined;
626
1140
  // Per-page LLM action UI. Site-wide config in
627
1141
  // `options.llmActions`; per-page opt-out via frontmatter
628
1142
  // `llmActions: false`. Skipped on redirect pages (no real
@@ -640,12 +1154,20 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
640
1154
  // available via other means. The "Open in" deep links work
641
1155
  // regardless of mirror availability — agents that can't fetch
642
1156
  // the page just see the URL in their chat.
1157
+ // pageHrefBase uses combined (urlBase + basePath) so the URL
1158
+ // resolves correctly when the host serves dist/ at a subpath
1159
+ // (GH Pages project page, multi-mount Cloudflare).
643
1160
  const pageHrefBase = section
644
- ? (basePath ? `${basePath}/${section}/${page.slug}` : `/${section}/${page.slug}`)
645
- : (basePath ? `${basePath}/${page.slug}` : `/${page.slug}`);
1161
+ ? (combined ? `${combined}/${section}/${page.slug}` : `/${section}/${page.slug}`)
1162
+ : (combined ? `${combined}/${page.slug}` : `/${page.slug}`);
646
1163
  const pageMdHref = `${pageHrefBase}.md`;
647
- const pageMdAbsoluteUrl = options.siteUrl
648
- ? options.siteUrl.replace(/\/$/, "") + pageMdHref
1164
+ // For absolute URLs (the "Copy as MD" deep link), use the
1165
+ // origin (no path) + the full combined path; siteUrl alone
1166
+ // would double-include the urlBase since pageHrefBase already
1167
+ // contains it.
1168
+ const { origin } = parseSiteUrl(options.siteUrl);
1169
+ const pageMdAbsoluteUrl = origin
1170
+ ? origin + pageMdHref
649
1171
  : pageMdHref;
650
1172
  // Markdown body for the Copy button. Reuse the same serializer
651
1173
  // that produces the .md mirror so what the user copies matches
@@ -687,7 +1209,10 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
687
1209
  "",
688
1210
  `const headings = ${JSON.stringify(page.headings || [])};`,
689
1211
  `const nav = navData;`,
690
- `const currentPath = "${buildCurrentPath(basePath, section, page.slug)}";`,
1212
+ // currentPath uses combined so it matches nav.json hrefs
1213
+ // (which are also combined-prefixed). getPagination compares
1214
+ // them as strings; mismatched prefixes break prev/next.
1215
+ `const currentPath = "${buildCurrentPath(combined, section, page.slug)}";`,
691
1216
  // Filter nav to the current (locale, version) bucket
692
1217
  // before computing prev/next — without this, pagination
693
1218
  // walks the global nav and a "Next" link can leak from
@@ -705,12 +1230,14 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
705
1230
  `const description = ${JSON.stringify(pageDescription)} || undefined;`,
706
1231
  `const ogImage = ${JSON.stringify(pageOgImage)} || undefined;`,
707
1232
  `const noindex = ${JSON.stringify(pageNoindex)};`,
1233
+ `const nofollow = ${JSON.stringify(pageNofollow)};`,
708
1234
  `const excludeFromSearch = ${JSON.stringify(pageExcludeFromSearch)};`,
709
1235
  `const pageTags = ${JSON.stringify(pageTags ?? null)};`,
710
1236
  `const pageStatus = ${JSON.stringify(pageStatus ?? null)};`,
711
1237
  `const pageType = ${JSON.stringify(pageTypeStr ?? null)};`,
712
1238
  `const pageAudience = ${JSON.stringify(pageAudience ?? null)};`,
713
1239
  `const pageCategory = ${JSON.stringify(pageCategory ?? null)};`,
1240
+ `const pageTaxonomies = ${JSON.stringify(pageTaxonomies ?? null)};`,
714
1241
  `const tagsIndexPath = ${JSON.stringify(tagsIndexPath)};`,
715
1242
  `const llmActionsProps = ${JSON.stringify(llmActionsEnabled
716
1243
  ? {
@@ -741,6 +1268,7 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
741
1268
  ` twitterHandle={siteConfig.twitterHandle || undefined}`,
742
1269
  ` themeColor={siteConfig.themeColor || undefined}`,
743
1270
  ` noindex={noindex}`,
1271
+ ` nofollow={nofollow}`,
744
1272
  ` excludeFromSearch={excludeFromSearch}`,
745
1273
  ` plausibleDomain={siteConfig.plausible?.domain}`,
746
1274
  ` plausibleScriptUrl={siteConfig.plausible?.scriptUrl}`,
@@ -756,12 +1284,39 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
756
1284
  ` pageType={pageType ?? undefined}`,
757
1285
  ` audience={pageAudience ?? undefined}`,
758
1286
  ` category={pageCategory ?? undefined}`,
1287
+ ` taxonomies={pageTaxonomies ?? undefined}`,
759
1288
  ` autoH1={${autoH1}}`,
760
1289
  ` autoLede={${autoLede}}`,
761
1290
  ` llmActions={llmActionsProps}`,
762
1291
  ` multiSource={${JSON.stringify(page.multiSource ?? null)} ?? undefined}`,
763
1292
  ` switcherMap={switcherMapData}`,
764
- ` basePath={${JSON.stringify(basePath || "/docs")}}`,
1293
+ // basePath here is the COMBINED URL prefix (urlBase from
1294
+ // site.url's path + dogsbay basePath). DocsLayout uses it
1295
+ // for switcher links, the footer llms.txt link, and the
1296
+ // <head> alternate link — all three need the full URL
1297
+ // prefix the host actually serves under. Empty string is
1298
+ // valid (root-served sites with no urlBase or basePath);
1299
+ // don't fall back to "/docs" — that would 404 for those.
1300
+ ` basePath={${JSON.stringify(combined)}}`,
1301
+ // navMode — controls whether DocsLayout server-renders the
1302
+ // full sidebar nav tree per page (`ssr-full`) or emits a
1303
+ // client-hydrated placeholder (`client`, default). The
1304
+ // client mode shrinks per-page HTML dramatically at scale.
1305
+ // See plans/client-rendered-nav.md.
1306
+ ` navMode={${JSON.stringify(options.navMode ?? "client")}}`,
1307
+ // Pagefind index URL — must include the combined prefix or
1308
+ // the loader 404s on subpath-mounted deploys. The pagefind
1309
+ // CLI writes to <astroOutput>/dist/pagefind/ which Astro
1310
+ // serves under its `base` (= urlBase); dogsbay's basePath
1311
+ // adds the second prefix layer. Empty combined → `/pagefind/`.
1312
+ ` pagefindUrl={${JSON.stringify(combined ? `${combined}/pagefind/` : "/pagefind/")}}`,
1313
+ // Favicon — composed with combined prefix so the
1314
+ // <link rel="icon"> resolves on subpath-mounted deploys.
1315
+ // Authors who want a different favicon override via the
1316
+ // `favicon` slot on DocsLayout, or drop the file at
1317
+ // `public/favicon.ico` in their Astro project (which is
1318
+ // what the default points at).
1319
+ ` favicon={${JSON.stringify(combined ? `${combined}/favicon.ico` : "/favicon.ico")}}`,
765
1320
  ` wideLayout={${wideLayout}}`,
766
1321
  `>`,
767
1322
  ` <MarkdownContentStack>`,
@@ -798,6 +1353,9 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
798
1353
  mkdirSync(dirname(pagePath), { recursive: true });
799
1354
  writeFileSync(pagePath, pageLines.join("\n") + "\n");
800
1355
  generated++;
1356
+ if (!section && page.slug === "index")
1357
+ rootIndexEmitted = true;
1358
+ generatedPaths.add(relative(outputDir, pagePath));
801
1359
  // Companion .md endpoint for content negotiation. Prerendered, so
802
1360
  // it's served as a static asset at runtime — no Worker overhead.
803
1361
  //
@@ -833,14 +1391,55 @@ export async function emitAstroPages(pages, nav, outputDir, options) {
833
1391
  }
834
1392
  }
835
1393
  // Generate index redirect at src/pages/index.astro — sends `/` to the
836
- // first nav href. Skipped when basePath is empty: with root-served
837
- // sites, src/pages/index.astro is the actual home page (not a
838
- // redirect target), and writing a redirect would clobber it.
839
- if (basePath !== "") {
840
- const firstHref = findFirstNavHref(nav, basePath);
841
- writeFileSync(join(outputDir, "src", "pages", "index.astro"), `---\nreturn Astro.redirect("${firstHref}");\n---\n`);
842
- }
843
- return { generated, outputNav };
1394
+ // first nav href. Two cases need it:
1395
+ // - basePath set: `/` (host apex) isn't a content page, so redirect
1396
+ // into the basePath subtree.
1397
+ // - basePath empty BUT no root home page emitted (no
1398
+ // `content/index.md`): the corpus's entry point is a nested page
1399
+ // (e.g. OpenShift's `welcome/index`), so `/` would 404 without a
1400
+ // redirect. Only emit here for non-section builds and never when a
1401
+ // real root index page exists (writing a redirect would clobber it).
1402
+ const firstHref = findFirstNavHref(nav, basePath);
1403
+ const needsRootRedirect = basePath !== "" ||
1404
+ (!section && !rootIndexEmitted && firstHref !== "");
1405
+ if (needsRootRedirect) {
1406
+ const indexPath = join(outputDir, "src", "pages", "index.astro");
1407
+ writeFileSync(indexPath, `---\nreturn Astro.redirect("${firstHref}");\n---\n`);
1408
+ generatedPaths.add(relative(outputDir, indexPath));
1409
+ }
1410
+ return { generated, outputNav, generatedPaths };
1411
+ }
1412
+ /**
1413
+ * Copy each passthrough `.astro` source to its computed output path.
1414
+ * Aborts with a clear error if the destination is already in
1415
+ * `generatedPaths` (a generated page from `emitAstroPages` would
1416
+ * silently overwrite the hand-authored file otherwise).
1417
+ */
1418
+ export function emitPassthroughAstroPages(copies, outputDir, generatedPaths) {
1419
+ if (copies.length === 0)
1420
+ return { copied: 0 };
1421
+ // Collision detection — a generated page and a passthrough page
1422
+ // would write to the same file. Refuse to overwrite; tell the
1423
+ // author exactly which two files conflict.
1424
+ const collisions = [];
1425
+ for (const copy of copies) {
1426
+ if (generatedPaths.has(copy.outputRelPath)) {
1427
+ collisions.push(copy.outputRelPath);
1428
+ }
1429
+ }
1430
+ if (collisions.length > 0) {
1431
+ throw new Error(`Passthrough Astro page collides with a generated page:\n` +
1432
+ collisions.map((c) => ` - ${c}`).join("\n") + "\n" +
1433
+ `Rename the .astro source or remove the colliding entry from nav.yml.`);
1434
+ }
1435
+ let copied = 0;
1436
+ for (const copy of copies) {
1437
+ const dest = join(outputDir, copy.outputRelPath);
1438
+ mkdirSync(dirname(dest), { recursive: true });
1439
+ copyFileSync(copy.sourceAbs, dest);
1440
+ copied++;
1441
+ }
1442
+ return { copied };
844
1443
  }
845
1444
  // ─── Tier 1: config-derived ─────────────────────────────────────────────
846
1445
  // Files driven entirely by config + flags. Always regenerated; site
@@ -856,6 +1455,54 @@ export function emitConfigDerivedFiles(outputDir, options) {
856
1455
  const hasSiteUrl = Boolean(options.siteUrl && /^https?:\/\//.test(options.siteUrl));
857
1456
  writeFileSync(join(outputDir, "public", "robots.txt"), buildRobotsTxt(options, hasSiteUrl));
858
1457
  }
1458
+ /**
1459
+ * Per-deploy-target artifact emission.
1460
+ *
1461
+ * Called from `emitSiteScaffold` (with `forceOverwrite=writeScaffold`
1462
+ * so `--force` regenerates from template) and from `dogsbay site
1463
+ * build` (with `forceOverwrite=false` so an existing site can adopt
1464
+ * a deploy target by editing config and rebuilding — the missing
1465
+ * artifact gets created on the next build).
1466
+ *
1467
+ * Emit policy is the union: write when forced OR when the file is
1468
+ * missing. Author edits to e.g. the workflow YAML survive every
1469
+ * regular build.
1470
+ *
1471
+ * Currently handles `github-pages` (workflow + .nojekyll). The
1472
+ * existing `cloudflare-workers` artifacts (wrangler.jsonc + package
1473
+ * scripts) stay in the scaffold-only path because they overlap with
1474
+ * scaffold-only files (package.json scripts, devDependencies). A
1475
+ * future refactor could fold them in here too.
1476
+ */
1477
+ export function emitDeployArtifacts(outputDir, options, opts = { forceOverwrite: false }) {
1478
+ if (options.deploy === "github-pages") {
1479
+ // GitHub reads workflows from <repo-root>/.github/workflows/, NOT
1480
+ // from inside subdirectories. Use projectDir (the repo root) for
1481
+ // the workflow file; fall back to outputDir when unset (flat
1482
+ // `dogsbay convert` flows where the Astro project IS the repo).
1483
+ const projectDir = options.projectDir ?? outputDir;
1484
+ // Path of the Astro output relative to the project root. Used by
1485
+ // the workflow's working-directory + cache-dependency-path so
1486
+ // pnpm install / pnpm run build target the right place. Empty
1487
+ // string when outputDir === projectDir (flat layout).
1488
+ const astroDirRel = relative(projectDir, outputDir).replace(/\\/g, "/");
1489
+ const workflowPath = join(projectDir, ".github", "workflows", "deploy.yml");
1490
+ if (opts.forceOverwrite || !existsSync(workflowPath)) {
1491
+ mkdirSync(dirname(workflowPath), { recursive: true });
1492
+ writeFileSync(workflowPath, buildGitHubPagesWorkflow(astroDirRel));
1493
+ }
1494
+ // .nojekyll — must exist in the deployed artifact root so GH
1495
+ // Pages skips Jekyll's `_underscored-paths` filter (Astro's
1496
+ // `_astro/` chunk dir gets eaten otherwise). Lives inside the
1497
+ // Astro project's `public/` so it's copied into `dist/` at
1498
+ // build time.
1499
+ const nojekyllPath = join(outputDir, "public", ".nojekyll");
1500
+ mkdirSync(dirname(nojekyllPath), { recursive: true });
1501
+ if (opts.forceOverwrite || !existsSync(nojekyllPath)) {
1502
+ writeFileSync(nojekyllPath, "");
1503
+ }
1504
+ }
1505
+ }
859
1506
  /**
860
1507
  * Emit `src/data/switcherMap.json` describing per-page
861
1508
  * version + locale equivalents. Always writes the file —
@@ -872,7 +1519,10 @@ export function emitConfigDerivedFiles(outputDir, options) {
872
1519
  * baseline page in a multi-version site).
873
1520
  */
874
1521
  export function emitSwitcherMap(pages, outputDir, options) {
875
- const basePath = normalizeBasePath(options.basePath);
1522
+ // Switcher URLs use combined so the link the dropdown emits
1523
+ // resolves under the host's served subpath (GH Pages project
1524
+ // pages, multi-mount Cloudflare).
1525
+ const combined = combinedPrefix(options);
876
1526
  const dataDir = join(outputDir, "src", "data");
877
1527
  const outPath = join(dataDir, "switcherMap.json");
878
1528
  // Detect axis activation by inspecting the data the loader
@@ -911,7 +1561,7 @@ export function emitSwitcherMap(pages, outputDir, options) {
911
1561
  const variant = {
912
1562
  ...(ms.locale !== undefined ? { locale: ms.locale } : {}),
913
1563
  ...(ms.version !== undefined ? { version: ms.version } : {}),
914
- url: `${basePath}/${page.slug}`,
1564
+ url: `${combined}/${page.slug}`,
915
1565
  };
916
1566
  if (!byLogicalKey[key])
917
1567
  byLogicalKey[key] = [];
@@ -989,6 +1639,10 @@ export function emitMissingTranslationStubs(pages, outputDir, options) {
989
1639
  return;
990
1640
  const basePath = normalizeBasePath(options.basePath);
991
1641
  const baseSegments = basePathSegments(basePath);
1642
+ // combined drives the redirect URL (the user-facing path they
1643
+ // get bounced to); basePath stays the filesystem path under
1644
+ // src/pages/ where the stub lives.
1645
+ const combined = combinedPrefix(options);
992
1646
  // Index existing pages by (slug after locale segment) so we
993
1647
  // can detect missing translations cheaply. Key shape:
994
1648
  // `<other-axis-prefix>/<originalSlug>` where other-axis-prefix
@@ -1022,7 +1676,7 @@ export function emitMissingTranslationStubs(pages, outputDir, options) {
1022
1676
  const targetUrl = `${basePath}/${targetSlug}`;
1023
1677
  if (existingByUrl.has(targetUrl))
1024
1678
  continue; // already translated
1025
- const defaultUrl = `${basePath}/${defaultPage.slug}`;
1679
+ const defaultUrl = `${combined}/${defaultPage.slug}`;
1026
1680
  const filePath = join(outputDir, "src", "pages", ...baseSegments, ...targetSlug.split("/"));
1027
1681
  // Ensure parent dir exists; write a redirect-stub Astro
1028
1682
  // file. Adding `.astro` to the leaf turns it into a
@@ -1064,10 +1718,20 @@ export function emitAgentReadinessFiles(pages, outputNav, outputDir, siteName, o
1064
1718
  if (options.llmsTxt !== false) {
1065
1719
  emitLlmsTxtFiles(outputDir, siteName, options, outputNav, pages);
1066
1720
  // public/_headers — Cloudflare Workers / Pages convention. Adds an
1067
- // RFC 8288 Link header pointing agents at /llms.txt without parsing
1068
- // HTML. Emitted alongside llms.txt so the two files travel together.
1721
+ // RFC 8288 Link header pointing agents at this mount's llms.txt
1722
+ // (basePath-prefixed) without parsing HTML. Emitted alongside
1723
+ // llms.txt so the two files travel together.
1069
1724
  mkdirSync(join(outputDir, "public"), { recursive: true });
1070
- writeFileSync(join(outputDir, "public", "_headers"), buildHeadersFile());
1725
+ // _headers Link header points at the per-mount llms.txt at
1726
+ // <combined>/llms.txt — the URL agents would actually fetch.
1727
+ writeFileSync(join(outputDir, "public", "_headers"), buildHeadersFile(combinedPrefix(options)));
1728
+ }
1729
+ // Sitemap — emitted by Dogsbay (not @astrojs/sitemap) into
1730
+ // public/<basePath>/sitemap-{index,0}.xml so multi-mount deploys
1731
+ // don't collide at host root. Gated on a valid http(s) siteUrl
1732
+ // because <loc> entries must be absolute.
1733
+ if (options.siteUrl && /^https?:\/\//.test(options.siteUrl)) {
1734
+ emitSitemapFiles(outputDir, options, pages);
1071
1735
  }
1072
1736
  // src/middleware.ts — Tier 1 (always update). Drives both the
1073
1737
  // `Accept: text/markdown` content-negotiation rewrite (via
@@ -1086,12 +1750,29 @@ export function emitAgentReadinessFiles(pages, outputNav, outputDir, siteName, o
1086
1750
  const localeRedirectOn = options.defaultLocale !== undefined && knownLocales.length >= 2;
1087
1751
  const axisRedirectOn = versionRedirectOn || localeRedirectOn;
1088
1752
  if (mdMirrorOn || axisRedirectOn) {
1753
+ // Taxonomy index paths share a single global namespace across
1754
+ // locales / versions (one `/tags/` for the whole site, not one
1755
+ // per locale). The redirect helper has to know to skip them or
1756
+ // it will 302 chip hrefs to non-existent locale-prefixed routes.
1757
+ // Strip leading `/` and pull just the first segment so a config
1758
+ // like `/tags` becomes the global-prefix entry `tags`.
1759
+ const globalPrefixes = [];
1760
+ if (options.taxonomyIndexPaths) {
1761
+ for (const raw of Object.values(options.taxonomyIndexPaths)) {
1762
+ const first = raw.replace(/^\/+/, "").split("/")[0];
1763
+ if (first)
1764
+ globalPrefixes.push(first);
1765
+ }
1766
+ }
1089
1767
  mkdirSync(join(outputDir, "src"), { recursive: true });
1090
1768
  writeFileSync(join(outputDir, "src", "middleware.ts"), buildMiddlewareSource({
1091
1769
  mdMirror: mdMirrorOn,
1092
1770
  axisRedirect: axisRedirectOn
1093
1771
  ? {
1094
- basePath: normalizeBasePath(options.basePath),
1772
+ // Middleware compares paths against the request URL,
1773
+ // which carries the host's served subpath — so use the
1774
+ // combined prefix here.
1775
+ basePath: combinedPrefix(options),
1095
1776
  ...(versionRedirectOn
1096
1777
  ? {
1097
1778
  defaultVersion: options.defaultVersion,
@@ -1104,6 +1785,7 @@ export function emitAgentReadinessFiles(pages, outputNav, outputDir, siteName, o
1104
1785
  knownLocales,
1105
1786
  }
1106
1787
  : {}),
1788
+ ...(globalPrefixes.length > 0 ? { globalPrefixes } : {}),
1107
1789
  }
1108
1790
  : undefined,
1109
1791
  }));
@@ -1164,21 +1846,49 @@ function buildRobotsTxt(options, hasSiteUrl) {
1164
1846
  const aiInput = options.aiInput ?? "yes";
1165
1847
  const aiTrain = options.aiTrain ?? "no";
1166
1848
  const contentSignal = `Content-Signal: search=${search}, ai-input=${aiInput}, ai-train=${aiTrain}\n`;
1167
- const sitemap = hasSiteUrl
1168
- ? `Sitemap: ${options.siteUrl.replace(/\/$/, "")}/sitemap-index.xml\n`
1849
+ // Per-mount sitemap path: each Dogsbay site emits its sitemap
1850
+ // index under <basePath>/, so robots.txt must point there too.
1851
+ // (Multi-mount deploys end up with one robots.txt per site at
1852
+ // their respective hosts / paths; each correctly references its
1853
+ // own mount's sitemap-index.)
1854
+ // Sitemap URL = origin + combined + /sitemap-index.xml. Use the
1855
+ // origin (no path) from site.url and the combined prefix (urlBase
1856
+ // + basePath); siteUrl could itself include a path component when
1857
+ // hosting on a subpath (GH Pages project page), so we strip it
1858
+ // here to avoid double-counting.
1859
+ const { origin } = parseSiteUrl(options.siteUrl);
1860
+ const combined = combinedPrefix(options);
1861
+ const sitemap = hasSiteUrl && origin
1862
+ ? `Sitemap: ${origin}${withBasePath(combined, "/sitemap-index.xml")}\n`
1169
1863
  : "";
1170
- return `User-agent: *\nAllow: /\n${contentSignal}${sitemap}`;
1864
+ // Llms-Txt: line — non-standard but follows the same shape as
1865
+ // `Sitemap:`. Crawlers and agents that scan robots.txt before
1866
+ // fetching pages get a direct pointer at the per-mount llms.txt.
1867
+ // RFC 9309 explicitly permits unknown directives ("intentionally
1868
+ // permissive of such future extensions") so this is harmless to
1869
+ // standards-compliant parsers. Emitted alongside Sitemap when
1870
+ // siteUrl is set; absolute URLs only (relative paths would be
1871
+ // ambiguous without a base).
1872
+ const llmsTxt = options.llmsTxt !== false && hasSiteUrl && origin
1873
+ ? `Llms-Txt: ${origin}${withBasePath(combined, "/llms.txt")}\n`
1874
+ : "";
1875
+ return `User-agent: *\nAllow: /\n${contentSignal}${sitemap}${llmsTxt}`;
1171
1876
  }
1172
1877
  /**
1173
1878
  * Build the contents of `public/_headers` (Cloudflare Pages / Workers
1174
1879
  * Static Assets convention). Emits a global RFC 8288 Link header
1175
- * pointing at the site's llms.txt index, so agents don't need to
1880
+ * pointing at this mount's llms.txt index, so agents don't need to
1176
1881
  * parse HTML to discover the LLM-friendly content listing.
1882
+ *
1883
+ * The Link target is basePath-prefixed (`</docs/llms.txt>` for a
1884
+ * `/docs` mount) — matches where the platform actually emits
1885
+ * llms.txt under the per-mount layout.
1177
1886
  */
1178
- function buildHeadersFile() {
1887
+ function buildHeadersFile(basePath) {
1888
+ const llmsHref = withBasePath(basePath, "/llms.txt");
1179
1889
  return [
1180
1890
  "/*",
1181
- ' Link: </llms.txt>; rel="describedby"; type="text/plain"',
1891
+ ` Link: <${llmsHref}>; rel="describedby"; type="text/plain"`,
1182
1892
  "",
1183
1893
  ].join("\n");
1184
1894
  }
@@ -1187,20 +1897,28 @@ function buildMiddlewareSource(config) {
1187
1897
  "// AUTO-GENERATED by `dogsbay site build` — do not edit.",
1188
1898
  "// Composes the docs-layout middleware helpers.",
1189
1899
  "//",
1190
- "// Markdown content negotiation:",
1191
- "// This middleware fires on every request, but in Astro's static",
1192
- "// prerender mode (output: \"static\") request headers are NOT",
1193
- "// forwarded Astro warns about \"Astro.request.headers was used",
1194
- "// when rendering...\" and serves a prerendered HTML response.",
1195
- "// That means `Accept: text/markdown` negotiation only kicks in",
1196
- "// under SSR (output: \"server\") or via an edge function on the",
1197
- "// deployment layer (Cloudflare Worker, Netlify Edge, etc.).",
1198
- "// For pure-static deploys, agents should follow the page's",
1199
- "// <link rel=\"alternate\" type=\"text/markdown\"> href to fetch",
1200
- "// the .md mirror directly (e.g. /docs.md).",
1900
+ "// Static-prerender guard:",
1901
+ "// In Astro's static output mode, this middleware is invoked",
1902
+ "// for every prerendered route at build time. Reading",
1903
+ "// `context.request.headers` there triggers an Astro warning",
1904
+ "// per page (\"Astro.request.headers was used during static",
1905
+ "// render\"), which floods `dogsbay site build` / `site preview`",
1906
+ "// output. Worse, the negotiation can't actually happen at",
1907
+ "// build time there's no runtime client whose Accept header",
1908
+ "// we'd be honoring.",
1201
1909
  "//",
1202
- "// The Cloudflare-Worker-driven full fix is tracked in",
1203
- "// plans/cloudflare-deploy-content-negotiation.md.",
1910
+ "// We guard with `context.isPrerendered` so prerendered routes",
1911
+ "// short-circuit to `next()` immediately. At runtime in static",
1912
+ "// deploys, middleware doesn't fire at all (no server); at",
1913
+ "// runtime in SSR / hybrid deploys, only dynamic routes fire,",
1914
+ "// which is exactly when negotiation makes sense.",
1915
+ "//",
1916
+ "// Markdown content negotiation:",
1917
+ "// For pure-static deploys, `Accept: text/markdown` is honored",
1918
+ "// by the platform (Cloudflare _headers + Worker, Netlify Edge",
1919
+ "// functions). Agents that can't send Accept headers should",
1920
+ "// follow the page's <link rel=\"alternate\" type=\"text/markdown\">",
1921
+ "// to fetch the .md mirror directly (e.g. /docs.md).",
1204
1922
  'import { defineMiddleware } from "astro:middleware";',
1205
1923
  ];
1206
1924
  if (config.mdMirror) {
@@ -1214,6 +1932,11 @@ function buildMiddlewareSource(config) {
1214
1932
  lines.push(`const AXIS_REDIRECT_CONFIG = ${JSON.stringify(config.axisRedirect, null, 2)};`, "");
1215
1933
  }
1216
1934
  lines.push("export const onRequest = defineMiddleware((context, next) => {");
1935
+ // Skip prerendered routes — see file-top comment for the rationale.
1936
+ // Avoids per-page Astro.request.headers warnings during build, and
1937
+ // matches runtime semantics (middleware doesn't fire on prerendered
1938
+ // routes when deployed).
1939
+ lines.push(" if (context.isPrerendered) return next();");
1217
1940
  lines.push(" const url = new URL(context.request.url);");
1218
1941
  if (config.mdMirror) {
1219
1942
  lines.push(' const accept = context.request.headers.get("accept");', " const mdTarget = shouldRewriteToMarkdown(accept, url.pathname);", " if (mdTarget) return context.rewrite(mdTarget);");
@@ -1245,8 +1968,18 @@ function buildMdEndpoint(page, sourceRel) {
1245
1968
  ].join("\n");
1246
1969
  }
1247
1970
  /**
1248
- * Emit `public/llms.txt`, `public/llms-full.txt`, and per-section
1249
- * `public/<dir>/llms.txt` files for the site.
1971
+ * Emit per-mount llms.txt + llms-full.txt + per-section indexes.
1972
+ *
1973
+ * Files live under `public/<basePath>/...` so multiple Dogsbay sites
1974
+ * can mount on the same host (`/docs/llms.txt` + `/api/llms.txt` +
1975
+ * `/handbook/llms.txt`) without colliding at the root. When basePath
1976
+ * is empty, this collapses to `public/llms.txt` — the single-site
1977
+ * llmstxt.org-spec layout.
1978
+ *
1979
+ * The host root `/llms.txt` is intentionally NOT emitted by the
1980
+ * platform: it's the user's umbrella file, analogous to
1981
+ * `sitemap-index.xml`. Multi-mount deploys hand-write a top-level
1982
+ * `/llms.txt` that links to each per-mount index.
1250
1983
  *
1251
1984
  * Per-section files are written for every top-level nav group that
1252
1985
  * resolves to a site directory (either via `group.href` or via the
@@ -1258,26 +1991,92 @@ function emitLlmsTxtFiles(outputDir, siteName, options, nav, pages) {
1258
1991
  description: options.description,
1259
1992
  siteUrl: options.siteUrl,
1260
1993
  };
1261
- const publicDir = join(outputDir, "public");
1262
- mkdirSync(publicDir, { recursive: true });
1263
- const hrefPrefix = normalizeBasePath(options.basePath);
1264
- writeFileSync(join(publicDir, "llms.txt"), buildLlmsTxt(siteConfig, nav, pages, { hrefPrefix }));
1265
- writeFileSync(join(publicDir, "llms-full.txt"), buildLlmsFullTxt(siteConfig, nav, pages, {
1994
+ // hrefPrefix is the COMBINED prefix — used for the URL paths that
1995
+ // appear inside the llms.txt body (so agents fetch the correct
1996
+ // host-relative URLs). Filesystem layout uses basePath alone:
1997
+ // `public/<basePath>/llms.txt` matches the existing per-mount
1998
+ // delivery shape.
1999
+ const hrefPrefix = combinedPrefix(options);
2000
+ const basePath = normalizeBasePath(options.basePath);
2001
+ const baseSegments = basePathSegments(basePath);
2002
+ const mountDir = join(outputDir, "public", ...baseSegments);
2003
+ mkdirSync(mountDir, { recursive: true });
2004
+ writeFileSync(join(mountDir, "llms.txt"), buildLlmsTxt(siteConfig, nav, pages, { hrefPrefix }));
2005
+ writeFileSync(join(mountDir, "llms-full.txt"), buildLlmsFullTxt(siteConfig, nav, pages, {
1266
2006
  summary: "body",
1267
2007
  serializePage: serializePageMd,
1268
2008
  hrefPrefix,
1269
2009
  }));
2010
+ // Per-section files. `deriveSectionDir` returns a host-absolute
2011
+ // path derived from nav hrefs, which since the combined-prefix
2012
+ // refactor (commit 132891e) include urlBase + basePath — NOT just
2013
+ // basePath. So joining its return onto public/ directly would
2014
+ // double-prefix into `public/<urlBase>/<basePath>/<section>/llms.txt`,
2015
+ // which then serves at `<urlBase>/<urlBase>/<basePath>/<section>/...`
2016
+ // once Astro's base prefix is applied at request time.
2017
+ //
2018
+ // Strip the combined prefix off the section dir to get just the
2019
+ // section tail, then re-prepend basePath via mountDir. Result:
2020
+ // `public/<basePath>/<section>/llms.txt`, served under the deploy's
2021
+ // base mount as `<urlBase>/<basePath>/<section>/llms.txt`.
2022
+ const combinedSegs = hrefPrefix.replace(/^\//, "");
1270
2023
  for (const group of nav) {
1271
2024
  if (!group.children || group.children.length === 0)
1272
2025
  continue;
1273
2026
  const dir = deriveSectionDir(group);
1274
2027
  if (!dir)
1275
2028
  continue;
1276
- const sectionPath = join(publicDir, dir, "llms.txt");
2029
+ let relDir;
2030
+ if (combinedSegs && dir === combinedSegs) {
2031
+ relDir = "";
2032
+ }
2033
+ else if (combinedSegs && dir.startsWith(`${combinedSegs}/`)) {
2034
+ relDir = dir.slice(combinedSegs.length + 1);
2035
+ }
2036
+ else {
2037
+ // Defensive: if for some reason the dir doesn't carry the
2038
+ // combined prefix (older importer, manual nav.yml, etc.), fall
2039
+ // back to the raw value rather than rooting at /.
2040
+ relDir = dir;
2041
+ }
2042
+ const sectionPath = relDir
2043
+ ? join(mountDir, relDir, "llms.txt")
2044
+ : join(mountDir, "llms.txt");
1277
2045
  mkdirSync(dirname(sectionPath), { recursive: true });
1278
2046
  writeFileSync(sectionPath, buildSectionLlmsTxt(siteConfig, group, pages, { hrefPrefix }));
1279
2047
  }
1280
2048
  }
2049
+ /**
2050
+ * Emit per-mount sitemap files.
2051
+ *
2052
+ * Writes `public/<basePath>/sitemap-index.xml` + `sitemap-0.xml`.
2053
+ * The index lists the single sub-sitemap today; future splits add
2054
+ * more sub-sitemap entries as the page count grows past
2055
+ * sitemaps.org's 50K-URL recommendation.
2056
+ *
2057
+ * Caller has already guarded on a valid http(s) `siteUrl` — without
2058
+ * one, `<loc>` entries can't be absolute and crawlers reject the
2059
+ * file. Skip emission rather than write a broken sitemap.
2060
+ */
2061
+ function emitSitemapFiles(outputDir, options, pages) {
2062
+ // Filesystem path uses basePath (sitemap files live in
2063
+ // public/<basePath>/sitemap-*.xml). The URL prefix encoded into
2064
+ // each <loc> uses combined so the absolute URLs resolve under the
2065
+ // host's served subpath. buildSitemap strips path off siteUrl
2066
+ // internally, so passing siteUrl + combined as basePath gives
2067
+ // origin + combined as the final URL.
2068
+ const basePath = normalizeBasePath(options.basePath);
2069
+ const combined = combinedPrefix(options);
2070
+ const baseSegments = basePathSegments(basePath);
2071
+ const mountDir = join(outputDir, "public", ...baseSegments);
2072
+ mkdirSync(mountDir, { recursive: true });
2073
+ writeFileSync(join(mountDir, "sitemap-0.xml"), buildSitemap(pages, {
2074
+ siteUrl: options.siteUrl,
2075
+ basePath: combined,
2076
+ siteNoindex: options.noindex === true,
2077
+ }));
2078
+ writeFileSync(join(mountDir, "sitemap-index.xml"), buildSitemapIndex({ siteUrl: options.siteUrl, basePath: combined }));
2079
+ }
1281
2080
  /**
1282
2081
  * Pick a directory under `public/` for a top-level nav group. Prefers
1283
2082
  * the group's own href (already a `/docs/x/y` path); otherwise falls
@@ -1355,6 +2154,11 @@ function copyComponents(outputDir) {
1355
2154
  "response-tabs", "schema-viewer", "code-samples", "copy-button",
1356
2155
  "markdown-example",
1357
2156
  "accordion", "link-card", "avatar", "math",
2157
+ // Icon resolves @ui/icon/Icon.astro → built-time SVG inlining
2158
+ // via @dogsbay/icons. Used by `:::cards` `{icon=...}` and the
2159
+ // inline `:icon[name]` directive. Without this entry every page
2160
+ // emitting the icon import 500s with "module not found".
2161
+ "icon",
1358
2162
  ];
1359
2163
  for (const name of needed) {
1360
2164
  const src = join(componentsSource, name);
@@ -1409,6 +2213,32 @@ function copyAssets(sourceDir, outputDir, imageOptimization) {
1409
2213
  catch { /* source may not exist */ }
1410
2214
  }
1411
2215
  // ── CSS generation (ported from import-mkdocs.ts) ───────
2216
+ /**
2217
+ * Build the `@source inline("...")` directive that pins the
2218
+ * grid-tone palette into the generated stylesheet.
2219
+ *
2220
+ * Why we need it: tone classes like `bg-primary/10` only appear in
2221
+ * `.astro` pages emitted by Dogsbay's grid-item serializer. When
2222
+ * Tailwind's content scanner doesn't pick them up — because the
2223
+ * page lives outside the default scan globs, or because a class
2224
+ * is composed at the boundary of an interpolation — they get
2225
+ * purged. Result observed in dogsbay-docs-markdown audit: half
2226
+ * the grid demo cells render with no background. Pinning forces
2227
+ * generation regardless of scanner reach.
2228
+ *
2229
+ * Single source of truth: derived from TONE_CLASSES so any new
2230
+ * tone added to the palette is automatically safelisted.
2231
+ */
2232
+ function buildToneSafelist() {
2233
+ const seen = new Set();
2234
+ for (const classes of Object.values(TONE_CLASSES)) {
2235
+ for (const cls of classes.split(/\s+/)) {
2236
+ if (cls)
2237
+ seen.add(cls);
2238
+ }
2239
+ }
2240
+ return [...seen].sort().join(" ");
2241
+ }
1412
2242
  function generateGlobalCss() {
1413
2243
  return `@import "tailwindcss";
1414
2244
  @import "./theme.css";
@@ -1417,6 +2247,14 @@ function generateGlobalCss() {
1417
2247
  @source "../../node_modules/@dogsbay/ui/src";
1418
2248
  @source "../../node_modules/@dogsbay/docs-layout/src";
1419
2249
 
2250
+ /* Pin the grid-tone palette. These classes are emitted into
2251
+ markdown-generated .astro pages by the grid-item serializer
2252
+ (TONE_CLASSES in @dogsbay/format-astro). Without inlining,
2253
+ opacity-modified utilities like bg-primary/10 get purged when
2254
+ Tailwind doesn't see them in the scanned globs, leaving grid
2255
+ demo cells with no visible background. */
2256
+ @source inline("${buildToneSafelist()}");
2257
+
1420
2258
  /* Prose typography for rendered content */
1421
2259
  .docs-prose {
1422
2260
  line-height: 1.7;
@@ -1463,6 +2301,33 @@ function generateGlobalCss() {
1463
2301
  & dl { margin: 1rem 0; }
1464
2302
  & dt { font-weight: 600; margin-top: 0.75rem; }
1465
2303
  & dd { margin-left: 1.5rem; color: var(--muted-foreground); }
2304
+
2305
+ /* AsciiDoc block roles ([role="x"] / [.x]) preserved as classes (#384).
2306
+ _abstract marks a module's lead/intro paragraph; .lead/.small are
2307
+ Asciidoctor built-ins; _additional-resources marks the trailing
2308
+ "Additional resources" section heading. */
2309
+ & ._abstract, & .abstract, & .lead {
2310
+ font-size: 1.05rem;
2311
+ line-height: 1.6;
2312
+ color: var(--muted-foreground);
2313
+ }
2314
+ & .small { font-size: 0.875rem; }
2315
+ & h2._additional-resources, & h3._additional-resources {
2316
+ font-size: 1.15rem;
2317
+ margin-top: 2rem;
2318
+ }
2319
+
2320
+ /* Related resources (#384 R3): [role="_additional-resources"] + .Title +
2321
+ list folds to a <section class="related-resources"> (→ DITA related-links). */
2322
+ & .related-resources {
2323
+ margin-top: 2rem;
2324
+ padding-top: 1rem;
2325
+ border-top: 1px solid var(--border);
2326
+ }
2327
+ & .related-resources-title {
2328
+ font-weight: 600;
2329
+ margin: 0 0 0.5rem;
2330
+ }
1466
2331
  }
1467
2332
 
1468
2333
  @utility scrollbar-none {