prose: yarn prose — LanguageTool lint for blog posts, noise-filtered

package.json

@@ -16,7 +16,8 @@
     "post": "node scripts/new-post.mjs",
     "post:image": "node scripts/add-image.mjs",
     "test:e2e": "playwright test",
-    "test:e2e:ui": "playwright test --ui"
+    "test:e2e:ui": "playwright test --ui",
+    "prose": "node scripts/prose.mjs"
   },
   "engines": {
     "node": ">=18.17.0"

scripts/prose.mjs

+#!/usr/bin/env node
+/**
+ * yarn prose [slug|file …] — LanguageTool lint for blog prose.
+ *
+ * No args → every post under content/blog/ (both languages). A slug arg
+ * checks that folder; a path ending in .md checks that file. The server is
+ * expected at localhost:8081 (`languagetool-server --port 8081`, brew).
+ *
+ * Noise filters, so the report stays worth reading:
+ *  - repeated whitespace → markdown hard breaks (trailing two spaces),
+ *    not typos;
+ *  - underscore emphasis confusing the tokenizer (e.g. `_ensemble_`);
+ *  - language-specific: FR "Parce que…" anaphora advice (REP_PARCE_QUE1)
+ *    and anglicism gripes on words the author uses deliberately.
+ * Everything else prints rule, message, context and top suggestion.
+ */
+
+import fs from 'fs';
+import path from 'path';
+
+const BLOG_ROOT = path.join(process.cwd(), 'content', 'blog');
+const LT_URL = 'http://localhost:8081/v2/check';
+const LANG_FOR = { fr: 'fr-FR', en: 'en-US' };
+
+const IGNORE_RULES = new Set([
+  'WHITESPACE_RULE', // markdown hard breaks read as "repeated space"
+  'FRENCH_WHITESPACE', // same, thin-space flavor (author's `?`/`!` spacing)
+  'REP_PARCE_QUE1', // deliberate anaphora is a feature of this blog's voice
+  'ENGLISH_WORD_REPEAT_BEGINNING_RULE', // same, EN flavor
+]);
+
+const NOISE_CONTEXT = [
+  // "_word_" italic underscores read as glued punctuation
+  (m) => /_[a-zà-ÿ]+_/i.test(m.context.text.slice(m.context.offset - 1, m.context.offset + m.context.length + 1)),
+];
+
+const targets = process.argv.slice(2);
+const files = [];
+
+function collect(slugOrFile) {
+  if (slugOrFile.endsWith('.md')) return files.push(path.resolve(slugOrFile));
+  const dir = path.isAbsolute(slugOrFile) ? slugOrFile : path.join(BLOG_ROOT, slugOrFile);
+  if (fs.existsSync(dir)) {
+    for (const f of fs.readdirSync(dir)) {
+      if (/^(fr|en)\.md$/.test(f)) files.push(path.join(dir, f));
+    }
+    return;
+  }
+  console.error(`prose: no post or file for "${slugOrFile}"`);
+}
+
+if (targets.length > 0) targets.forEach(collect);
+else {
+  for (const slug of fs.readdirSync(BLOG_ROOT)) {
+    const dir = path.join(BLOG_ROOT, slug);
+    if (fs.statSync(dir).isDirectory()) collect(dir);
+  }
+}
+
+function bodyOf(file) {
+  const raw = fs.readFileSync(file, 'utf8');
+  const parts = raw.split('---');
+  return parts.length >= 3 ? parts.slice(2).join('---').trim() : raw.trim();
+}
+
+async function check(file) {
+  const lang = path.basename(file, '.md');
+  const text = bodyOf(file);
+  const res = await fetch(LT_URL, {
+    method: 'POST',
+    body: new URLSearchParams({ text, language: LANG_FOR[lang] ?? lang, level: 'picky' }),
+  });
+  if (!res.ok) throw new Error(`LanguageTool ${res.status} — is it running? (languagetool-server --port 8081)`);
+  const { matches } = await res.json();
+  const worth = matches.filter((m) => {
+    if (IGNORE_RULES.has(m.rule.id)) return false;
+    return !NOISE_CONTEXT.some((fn) => {
+      try { return fn(m); } catch { return false; }
+    });
+  });
+  const label = path.relative(process.cwd(), file);
+  console.log(`\n=== ${label} — ${worth.length} finding(s) (${matches.length} raw) ===`);
+  for (const m of worth) {
+    const { text: ctx, offset, length } = m.context;
+    const snippet = ctx.slice(Math.max(0, offset - 20), offset + length + 20).replace(/\n/g, '⏎');
+    const fix = m.replacements?.[0]?.value ?? '—';
+    console.log(`[${m.rule.id}] ${m.message.slice(0, 120)}`);
+    console.log(`    …${snippet}…  → ${fix.slice(0, 60)}`);
+  }
+}
+
+let failed = 0;
+for (const file of files) {
+  try {
+    await check(file);
+  } catch (e) {
+    failed += 1;
+    console.error(`prose: ${path.basename(file)}: ${e.message}`);
+  }
+}
+process.exit(failed > 0 ? 1 : 0);