feat(scanner): source-markdown scanner + fixer (#1963)

package.json

@@ -30,6 +30,7 @@
     "generate-dark-theme": "tsx scripts/generate-dark-theme.ts",
     "validate-tutorials": "tsx scripts/validate-tutorials.ts",
     "lint:tutorial-markdown": "tsx scripts/lint-tutorial-markdown.ts",
+    "scan-tutorial-markdown": "tsx scripts/scan-markdown-source.ts",
     "publish-content": "tsx scripts/publish-content.ts",
     "backfill-images": "tsx scripts/backfill-images.ts",
     "preflight:ai-quiz-smoke": "tsx scripts/preflight-ai-quiz-smoke.ts",

scripts/parsers/images.ts

@@ -7,6 +7,29 @@ export interface ImageResolveOpts {
   rewriteImages?: boolean
 }
 
+/**
+ * Strip the authoring directive comment (`<!-- border -->`, `<!-- size:540px -->`,
+ * or any combination like `<!-- border; size:540px -->`) that prefixes an image.
+ *
+ * If this comment survives into the body, goldmark treats the leading `<!--` as
+ * the start of an HTML block and consumes the trailing `![…]` as raw HTML text,
+ * so the image never renders (#1137). The directives may appear in either order,
+ * separated by `;` and/or whitespace. Only comments composed solely of these
+ * known directives are stripped — unrelated comments are left intact.
+ *
+ * Extracted from `resolveImageURLs` so the source-markdown scanner (issue #1963)
+ * and this in-flight pre-processor share ONE detection/transform and can never
+ * disagree. Pure, idempotent, and independent of repo/branch/URL context — the
+ * only part of image handling that is a genuine *source* fix (the URL rewrite is
+ * render-time and stays here, not in the scanner).
+ */
+export function stripImageDirectiveComments(content: string): string {
+  return content.replace(
+    /<!--\s*(?:border|size:\s*\d+px)(?:\s*;?\s*(?:border|size:\s*\d+px))*\s*-->\s*(!\[)/g,
+    '$1'
+  )
+}
+
 export function resolveImageURLs(content: string, opts: ImageResolveOpts): string {
   const { repo, branch, slug, rewriteImages = true } = opts
   let result = content
@@ -24,16 +47,5 @@ export function resolveImageURLs(content: string, opts: ImageResolveOpts): strin
     )
   }
 
-  // Strip the authoring directive comment (`<!-- border -->`, `<!-- size:540px -->`,
-  // or any combination like `<!-- border; size:540px -->`) that prefixes an image.
-  // If this comment survives into the body, goldmark treats the leading `<!--` as
-  // the start of an HTML block and consumes the trailing `![…]` as raw HTML text,
-  // so the image never renders (#1137). The directives may appear in either order,
-  // separated by `;` and/or whitespace. Only comments composed solely of these
-  // known directives are stripped — unrelated comments are left intact.
-  result = result.replace(
-    /<!--\s*(?:border|size:\s*\d+px)(?:\s*;?\s*(?:border|size:\s*\d+px))*\s*-->\s*(!\[)/g,
-    '$1'
-  )
-  return result
+  return stripImageDirectiveComments(result)
 }