1 files changed, 116 insertions, 0 deletions
diff --git a/src/features/markdown/logic/text-file.ts b/src/features/markdown/logic/text-file.ts
new file mode 100644
index 0000000..4321c6f
--- /dev/null
+++ b/src/features/markdown/logic/text-file.ts
@@ -0,0 +1,116 @@
+import { Marked } from "marked";
+
+export interface TextFileReference {
+  readonly label: string;
+  readonly path: string;
+}
+export type MarkdownPart =
+  | { readonly kind: "markdown"; readonly text: string }
+  | { readonly kind: "text-file"; readonly reference: TextFileReference };
+
+/** Input is marked's already Markdown-unescaped href. Decode percent escapes ONCE. */
+export function textFilePath(href: string): string | null {
+  if (!href.startsWith("text-file:")) return null;
+  let path: string;
+  try {
+    path = decodeURIComponent(href.slice(10));
+  } catch {
+    return null;
+  }
+  if (
+    !path ||
+    path !== path.trim() ||
+    path.length > 4096 ||
+    path.includes("\\") ||
+    Array.from(path).some((character) => {
+      const code = character.charCodeAt(0);
+      return code < 32 || (code >= 127 && code <= 159);
+    }) ||
+    path.startsWith("~") ||
+    path.startsWith("//") ||
+    /^[a-zA-Z][a-zA-Z0-9+.-]*:/.test(path)
+  )
+    return null;
+  return path;
+}
+
+/** Only a complete, top-level, standalone inline-link paragraph opts in. No I/O. */
+export function splitTextFiles(text: string, streaming = false): readonly MarkdownPart[] {
+  if (!text.includes("text-file:")) return [{ kind: "markdown", text }];
+  const parts: MarkdownPart[] = [];
+  let prose = "";
+  let offset = 0;
+  let embeds = 0;
+  const tokens = new Marked({ gfm: true }).lexer(text);
+  for (const token of tokens) {
+    offset += token.raw.length;
+    const children = token.type === "paragraph" ? token.tokens : undefined;
+    const link = children?.length === 1 ? children[0] : undefined;
+    const path =
+      link?.type === "link" &&
+      link.title === null &&
+      // Reference-style links and multiline paragraphs are deliberately not embeds.
+      /^\s*\[[\s\S]*\]\(/.test(token.raw) &&
+      !token.raw.trim().includes("\n")
+        ? textFilePath(link.href)
+        : null;
+    // During streaming wait for a paragraph boundary, not merely a closing ')'
+    // that could still grow into adjacent prose on the next delta.
+    const settled = !streaming || /^\s*\n/.test(text.slice(offset));
+    if (path !== null && link?.type === "link" && settled && embeds < 8) {
+      embeds++;
+      if (prose) parts.push({ kind: "markdown", text: prose });
+      prose = "";
+      parts.push({ kind: "text-file", reference: { label: link.text, path } });
+    } else {
+      prose += token.raw;
+      if (path !== null && settled && embeds >= 8)
+        prose += "\n\n_Text-file embed limit reached (8 per message)._\n\n";
+    }
+  }
+  if (prose) parts.push({ kind: "markdown", text: prose });
+  // Reference definitions apply document-wide, even across embed boundaries.
+  const definitions = tokens
+    .filter((token) => token.type === "def")
+    .map((token) => token.raw)
+    .join("\n");
+  return embeds && definitions
+    ? parts.map((part) =>
+        part.kind === "markdown"
+          ? {
+              ...part,
+              text: `${part.text}\n\n${definitions}`,
+            }
+          : part,
+      )
+    : parts;
+}
+
+/** Restore top-level fence context when the tail begins inside a code fence. */
+function fenceAt(lines: readonly string[], end: number): string | null {
+  let fence: { marker: string; info: string } | null = null;
+  for (const line of lines.slice(0, end)) {
+    if (fence) {
+      const close = /^ {0,3}(`+|~+)\s*$/.exec(line)?.[1];
+      if (close && close[0] === fence.marker[0] && close.length >= fence.marker.length)
+        fence = null;
+    } else {
+      const match = /^ {0,3}(`{3,}|~{3,})(.*)$/.exec(line);
+      if (match?.[1]) fence = { marker: match[1], info: match[2] ?? "" };
+    }
+  }
+  return fence ? `${fence.marker}${fence.info}` : null;
+}
+export function textFilePreview(text: string, markdown: boolean) {
+  const lines = text.replace(/\r\n/g, "\n").replace(/\n$/, "").split("\n");
+  if (lines.length <= 12) return { head: text, tail: "", omitted: 0 };
+  const headLines = lines.slice(0, 6);
+  const tailLines = lines.slice(-6);
+  const headFence = markdown ? fenceAt(lines, 6) : null;
+  const tailFence = markdown ? fenceAt(lines, lines.length - 6) : null;
+  return {
+    head: headLines.join("\n") + (headFence ? `\n${headFence.match(/^[`~]+/)?.[0]}` : ""),
+    tail: (tailFence ? `${tailFence}\n` : "") + tailLines.join("\n"),
+    omitted: lines.length - 12,
+  };
+}