@half-built/astro 0.7.0 → 0.9.0

package/README.md

@@ -72,6 +72,27 @@ component that injects `getCollection` and its own gate, the way the
 blog does. The resolvers live in `lib/drafts.ts` for consumers that
 want the logic without the components.
 
+## Derived excerpts
+
+Posts carry no excerpt frontmatter. `lib/excerpt.ts` derives the card
+text from the post body at build time: the opening prose, consecutive
+paragraphs joined by a space, cut at a word boundary within
+`EXCERPT_LIMIT` (200 characters) and always ended with an ellipsis,
+the reader's cue that the post continues. The run stops at the first
+heading, list, blockquote, code fence, or image line, so a card never
+crosses into a later section; components and the lead-break are
+invisible to it. `content/ExcerptStart.astro` on its own line moves
+the start to the paragraph after it, for a post that opens on a TL;DR
+or an aside. `postExcerpt(entry)` is the policy: a published post
+with no prose fails the build naming the slug, a draft warns once per
+build and renders blank. It takes any entry shaped
+`{ body?, data: { slug, draft? } }`, so a `CollectionEntry` passes
+with no cast. A consumer calls `postExcerpt` from every place a
+summary renders (cards, meta description, feed, search index) and
+never stores the result. The blog is the reference consumer and
+derives all four from it. The rule was settled on the blog's 54 posts
+on 2026-09-13 and moved here on 2026-09-14.
+
 ## Live code colors
 
 `shiki/code-theme` bakes its amber values into every highlighted span

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@half-built/astro",
-  "version": "0.7.0",
+  "version": "0.9.0",
   "description": "Astro components, islands, and pure helpers for the half-built design system.",
   "license": "MIT",
   "repository": {

package/src/components/Footer.astro

@@ -3,7 +3,6 @@ import type { SitemapGroup, EcosystemEntry } from "./models";
 import { BP_PHONE_MAX } from "../scripts/core/breakpoints";
 
 interface Props {
-  siteName: string;
   legalHolder: string;
   sitemap: SitemapGroup[];
   ecosystem: EcosystemEntry[];
@@ -12,7 +11,6 @@ interface Props {
 }
 
 const {
-  siteName,
   legalHolder,
   sitemap,
   ecosystem,
@@ -28,17 +26,15 @@ const year = new Date().getFullYear();
     <div class="shell">
       <div class="bottom-footer-info">
         <div class="site-info">
-          {/* Two lines at every width (owner call 2026-09-13): the
-              copyright over the site name, a real break rather than the
-              old pipe that retired on phones and left the wrap to do
-              the separating. The copyright's segments stay unbreakable
-              so a narrow screen wraps that line only after the year. */}
+          {/* The copyright line alone (owner call 2026-09-14). The
+              site-name home link that used to sit under it retired.
+              The masthead and the Ecosystem group's bold entry already
+              say where you are, and the notice is complete without it.
+              The segments stay unbreakable so a narrow screen wraps
+              only after the year. */}
           <slot name="site-info">
-            <span class="site-info-line">
-              <span class="site-info-seg">Copyright © {year}</span>{" "}
-              <span class="site-info-seg">{legalHolder}</span>
-            </span>
-            <span class="site-info-line"><a href="/">{siteName}</a></span>
+            <span class="site-info-seg">Copyright © {year}</span>{" "}
+            <span class="site-info-seg">{legalHolder}</span>
           </slot>
         </div>
       </div>
@@ -128,18 +124,9 @@ const year = new Date().getFullYear();
     font-size: var(--font-size-sm);
   }
 
-  /* The two lines center under each other like the sitemap below
-     (owner report 2026-08-25). */
+  /* Centered like the sitemap below (owner report 2026-08-25). */
   .site-info { text-align: center; }
-  .site-info-line { display: block; }
-
-  /* Underlined, unlike the sitemap links below: this one sits inside
-     running text (the copyright block), where the same ink at 0.8
-     opacity is not a visible link mark (WCAG 1.4.1, axe
-     link-in-text-block; the line break is all that frames it). */
-  .site-info a { text-decoration: underline; opacity: 0.8; transition: var(--transition); color: var(--ink); }
   .site-info-seg { white-space: nowrap; }
-  .site-info a:hover { opacity: 1; }
 
   /* Sitemap columns under the copyright line, centered like it (owner
      feedback 2026-08-01). position: relative lifts the content above the

package/src/components/content/ExcerptStart.astro

@@ -0,0 +1,12 @@
+---
+---
+
+{/* Renders nothing. Marks where the derived card text starts (README,
+     "Derived excerpts"; owner call 2026-09-13): lib/excerpt.ts takes
+     the opening prose from the paragraph after this line instead of
+     from the top of the post. Use it when a post opens with a TL;DR or
+     an aside the card should jump. A real import rather than a comment
+     so the marker is typed, visible in the source, and cannot be
+     reflowed away. The note lives in the template because Prettier's
+     Astro plugin drops a comment-only frontmatter. Moved from the blog
+     2026-09-14 (owner call). */}

package/src/lib/excerpt.ts

@@ -0,0 +1,198 @@
+/* Card text derived from a post body (owner call 2026-09-13; the rule
+   is recorded in the blog's docs/superpowers/specs/2026-09-13-derived-excerpts-design.md
+   and summarized in this package's README, "Derived excerpts"). The
+   card is the opening prose, consecutive paragraphs joined, cut at a
+   word boundary within EXCERPT_LIMIT and always ended with an
+   ellipsis, so every listing samples the post's own opening and the
+   reader can see it continues. Consumers keep no excerpt frontmatter.
+
+   Built site-side in the blog first and proved on its 54 posts, then
+   moved here on 2026-09-14 (owner call) so BEADZ and later sites get
+   the same rule. The blog is the reference consumer. */
+
+export const EXCERPT_LIMIT = 200;
+
+const MARKER = /^\s*<ExcerptStart\s*\/>\s*$/m;
+
+/* Block-level things that are not prose, removed before the body is
+   split into paragraphs. Order matters only for the code fence, which
+   must go before anything that would look inside it. */
+const IMPORT_LINE = /^import .*$/gm;
+const CODE_FENCE = /```[\s\S]*?```/g;
+const LEAD_BREAK = /<div class="lead-break">[\s\S]*?<\/div>/g;
+
+/* A capitalized tag is a component. It is a block only when it
+   occupies whole lines: opening tag starts a line, closing tag ends
+   one. Non-greedy to the same close tag; the posts do not nest a
+   component inside itself. A component inline in a sentence (PostLink,
+   WhenPublished around a link) is not matched here and keeps its text
+   through flatten; the built-output suite caught the Robot Migrates
+   card as the word "In" on 2026-09-13 when this was not so. */
+const COMPONENT_BLOCK =
+  /^[ \t]*<([A-Z][A-Za-z0-9]*)\b[^>]*>[\s\S]*?<\/\1>[ \t]*$/gm;
+
+const SELF_CLOSING = /^[ \t]*<[A-Za-z][A-Za-z0-9-]*\b[^>]*\/>[ \t]*$/gm;
+/* A lowercase tag opening a line is an html block (figure, div,
+   details); it runs to its close. */
+const HTML_BLOCK = /^<([a-z][a-z0-9-]*)\b[^>]*>[\s\S]*?<\/\1>/gm;
+const MDX_EXPRESSION_LINE = /^\s*\{[\s\S]*?\}\s*$/gm;
+
+/* Paragraph test: the first non-space character says what the block
+   is. A block that starts with a tag is not rejected here; after the
+   block-level removals above, a leading tag is an inline component or
+   element opening a sentence, and flatten keeps its text. A bare
+   triple backtick is the sentinel a code fence leaves behind. */
+const NOT_PROSE = /^(?:#|>|-\s|\*\s|\||!|```|\d+[.)]\s)/;
+
+/* A code fence is emptied rather than removed: the sentinel keeps it
+   a non-prose block, so it still ends the opening prose (spec rule 1)
+   instead of vanishing and letting the run continue past it. */
+function stripBlocks(body: string): string {
+  return body
+    .replace(CODE_FENCE, "\n\n```\n\n")
+    .replace(IMPORT_LINE, "")
+    .replace(LEAD_BREAK, "\n\n")
+    .replace(COMPONENT_BLOCK, "\n\n")
+    .replace(HTML_BLOCK, "\n\n")
+    .replace(SELF_CLOSING, "\n\n")
+    .replace(MDX_EXPRESSION_LINE, "\n\n");
+}
+
+/* The opening prose: consecutive prose blocks from the top, joined by
+   a space, until the limit is reached (owner call 2026-09-13, after
+   RoverBot's one-line opener made a one-line card; the same afternoon
+   the card had been the first paragraph alone). Non-prose blocks
+   before any prose are skipped, as before; the first non-prose block
+   after prose has started (a heading, a list, a quote, a code fence,
+   an image line) ends the run, so a card never crosses into a later
+   section. Components and the lead-break were stripped above, so a
+   gallery between two intro paragraphs does not end it. A block made
+   only of tags (a stray inline component on its own line) flattens to
+   nothing and is passed over. */
+function openingProse(body: string): string | null {
+  const blocks = stripBlocks(body)
+    .split(/\n\s*\n/)
+    .map((b) => b.trim())
+    .filter((b) => b.length > 0);
+
+  let taken = "";
+
+  for (const b of blocks) {
+    if (NOT_PROSE.test(b)) {
+      if (taken) break;
+      continue;
+    }
+
+    const text = flatten(b);
+    if (!text) continue;
+    taken = taken ? `${taken} ${text}` : text;
+    if (taken.length >= EXCERPT_LIMIT) break;
+  }
+
+  return taken || null;
+}
+
+/* Inline markdown to plain text. Links keep their text (inline and
+   reference style), mdx comments and expressions go, emphasis and code
+   markers go, inline html tags go and their text stays, escapes lose
+   the backslash, and the block's own line breaks collapse to one
+   space. */
+function flatten(block: string): string {
+  return block
+    .replace(/!\[([^\]]*)\]\([^)]*\)/g, "$1")
+    .replace(/\[([^\]]+)\]\([^)]*\)/g, "$1")
+    .replace(/\[([^\]]+)\]\[[^\]]*\]/g, "$1")
+    .replace(/\{\/\*[\s\S]*?\*\/\}/g, " ")
+    .replace(/\{[^{}\n]*\}/g, " ")
+    .replace(/<[^>]+>/g, "")
+    .replace(/`([^`]*)`/g, "$1")
+    .replace(/(\*\*|__)(.+?)\1/g, "$2")
+    .replace(/(^|[^\w\\])[*_](.+?)[*_](?=[^\w]|$)/g, "$1$2")
+    .replace(/\\([*_`[\]\\])/g, "$1")
+    .replace(/\s+/g, " ")
+    .trim();
+}
+
+/* The cut (owner call 2026-09-13, replacing the sentence rule the same
+   day): whole words while the text stays within the limit, then the
+   ellipsis, always. The ellipsis is the reader's cue that the post
+   continues, so it goes on even when the whole paragraph fits. A
+   period, comma, colon, or semicolon right before it comes off, so the
+   card ends "onto Cloudflare Pages…" and not "Pages.…"; a question
+   mark or exclamation stays. */
+function cutAtWords(text: string): string {
+  let out = text;
+
+  if (out.length > EXCERPT_LIMIT) {
+    /* One past the limit so a word that ends exactly there survives:
+       the space after it is what the search finds. */
+    const head = out.slice(0, EXCERPT_LIMIT + 1);
+    const lastSpace = head.lastIndexOf(" ");
+
+    out =
+      lastSpace > 0 ? head.slice(0, lastSpace) : out.slice(0, EXCERPT_LIMIT);
+  }
+
+  return `${out.replace(/[.,;:]+$/, "").trimEnd()}…`;
+}
+
+export function deriveExcerpt(body: string): string | null {
+  const marker = MARKER.exec(body);
+  const source = marker ? body.slice(marker.index + marker[0].length) : body;
+  const text = openingProse(source);
+  return text === null ? null : cutAtWords(text);
+}
+
+/* Each warning prints once per build. postExcerpt runs once per page
+   that lists a post (home, archives, categories, feed, search index),
+   so without this a drafts build repeated seven stub warnings fifty
+   times and buried anything else in the log (2026-09-13). */
+const warned = new Set<string>();
+
+function warnOnce(message: string): void {
+  if (warned.has(message)) return;
+  warned.add(message);
+  console.warn(message);
+}
+
+/* Test hook: the set outlives a test, so a suite that asserts a
+   warning clears it between cases. */
+export function resetExcerptWarnings(): void {
+  warned.clear();
+}
+
+/* The consumer's collection entry, structurally: body is the raw MDX
+   minus frontmatter as Astro's content layer hands it over, slug names
+   the post in a thrown error or a warning, and draft selects the
+   policy. A CollectionEntry<"posts"> satisfies this with no cast; the
+   same shape lib/drafts.ts uses for Draftable. */
+export interface Excerptable {
+  body?: string;
+  data: { slug: string; draft?: boolean };
+}
+
+/* The one place the publish policy lives (spec, "Consumers"): every
+   consumer calls this, never deriveExcerpt directly. A published post
+   with no usable opening fails the build; a draft with none warns and
+   renders blank, so a SHOW_DRAFTS=1 build still runs and the stub is
+   loud without being a wall. */
+export function postExcerpt(post: Excerptable): string {
+  const slug = post.data.slug;
+  const derived = deriveExcerpt(post.body ?? "");
+
+  if (derived === null) {
+    if (post.data.draft) {
+      warnOnce(
+        `[excerpt] draft "${slug}" has no prose paragraph; card is blank`,
+      );
+
+      return "";
+    }
+
+    throw new Error(
+      `[excerpt] published post "${slug}" has no prose paragraph to derive a card from`,
+    );
+  }
+
+  return derived;
+}