feat(ai): read the importer's deck and transcript cache; studio DraftWriter fits DraftSink as is

Retrieval now reads <dataDir>/cache/decks/<courseId>/<deckId>/text.json and
<dataDir>/cache/transcripts/<courseId>/<lessonId>.json (skipped when the lesson
has a transcript file). DraftSink returns { location?, branch?, commit? }, so
the studio's createDraftWriter(ctx) needs no adapter; generateDraft computes
paths itself and reports the draft branch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

packages/feature-ai/src/retrieval/course-index.ts

@@ -128,6 +128,18 @@ function lessonTitleOf(course: LoadedCourse, lessonId: string | undefined): stri
   return lessonId ? course.lessonIndex.get(lessonId)?.lesson.title : undefined;
 }
 
+async function readJson<T>(path: string): Promise<T | null> {
+  try {
+    const f = Bun.file(path);
+    if (!(await f.exists()) || f.size > MAX_FILE * 25) return null;
+    return (await f.json()) as T;
+  } catch {
+    return null;
+  }
+}
+
+const clock = (s: number) => `${Math.floor(s / 60)}:${String(Math.floor(s % 60)).padStart(2, "0")}`;
+
 export async function buildCourseIndex(course: LoadedCourse, dataDir: string): Promise<CourseIndex> {
   const bm = new Bm25<Chunk>();
   const seen = new Set<string>();
@@ -179,8 +191,50 @@ export async function buildCourseIndex(course: LoadedCourse, dataDir: string): P
     });
   }
 
-  // Derived text from the importer and transcription jobs: <dataDir>/cache/<courseId>/**.
   const decks = new Map((course.manifest?.decks ?? []).map((d) => [d.id, d]));
+
+  // The importer's deck cache: <dataDir>/cache/decks/<courseId>/<deckId>/text.json = [{ page, text }].
+  for (const deck of decks.values()) {
+    const pages = await readJson<{ page?: number; text?: string }[]>(join(dataDir, "cache", "decks", course.id, deck.id, "text.json"));
+    if (!Array.isArray(pages)) continue;
+    const lessonId = course.lessonIndex.has(deck.attachTo) ? deck.attachTo : undefined;
+    for (const p of pages) {
+      const text = typeof p?.text === "string" ? p.text.trim() : "";
+      if (!text) continue;
+      for (const part of splitLong(text)) {
+        add({ path: posix(deck.file), kind: "deck", text: part, heading: p.page ? `Slide ${p.page}` : deck.title, lessonId, lessonTitle: lessonTitleOf(course, lessonId) });
+      }
+    }
+  }
+
+  // The importer's transcription cache: <dataDir>/cache/transcripts/<courseId>/<lessonId>.json =
+  // { lessonId, text, segments: [{ start, end, text }] }. Skipped when the lesson has a transcript file in the
+  // course, which the transcription job writes alongside and which is indexed above.
+  const trDir = join(dataDir, "cache", "transcripts", course.id);
+  for (const name of await readdir(trDir).catch(() => [] as string[])) {
+    if (extname(name) !== ".json") continue;
+    const lessonId = name.slice(0, -5);
+    const entry = course.lessonIndex.get(lessonId);
+    if (!entry || entry.lesson.transcript) continue;
+    const tr = await readJson<{ text?: string; segments?: { start?: number; text?: string }[] }>(join(trDir, name));
+    if (!tr) continue;
+    const segs = Array.isArray(tr.segments) ? tr.segments.filter((s) => typeof s?.text === "string" && s.text.trim()) : [];
+    const blocks: { at: number; text: string }[] = [];
+    if (segs.length) {
+      for (const s of segs) {
+        const last = blocks[blocks.length - 1];
+        if (last && last.text.length + s.text!.length < MAX_CHUNK) last.text += ` ${s.text!.trim()}`;
+        else blocks.push({ at: Number(s.start) || 0, text: s.text!.trim() });
+      }
+    } else if (typeof tr.text === "string" && tr.text.trim()) {
+      blocks.push(...splitLong(tr.text.trim()).map((t) => ({ at: -1, text: t })));
+    }
+    for (const b of blocks) {
+      add({ path: entry.lesson.file ? posix(entry.lesson.file) : `transcripts/${lessonId}`, kind: "transcript", text: b.text, heading: b.at >= 0 ? `at ${clock(b.at)}` : undefined, lessonId, lessonTitle: entry.lesson.title });
+    }
+  }
+
+  // Older layout: plain text under <dataDir>/cache/<courseId>/**.
   const cacheRoot = join(dataDir, "cache", course.id);
   await walk(cacheRoot, async (abs, rel) => {
     if (!CACHE_TEXT.has(extname(rel).toLowerCase())) return true;