Teach it back: the AI becomes the curious student

A new mode where the user explains a topic in their own words and the AI
plays a curious classmate: short reactions, exactly one follow-up question
at a time, never "correct"/"wrong", never lecturing, and everything it knows
comes from the user's own decks and notes.

- Entry points: a sidebar row next to Exam mode and a "Teach it back" action
  in every deck's menu (which prefills the topic and scope).
- Start screen: deck or typed topic, friendly/skeptical student, the same
  context and model chips as Chat (/teach?deck=<id> pre-selects the deck).
- Session: the student prompt is a mode of the existing /api/chat route, so
  model resolution, allowlists, rate limits and the streaming client are the
  same code; the personality lives in lib/prompts.ts (TEACH_*). The deck's
  cards travel as a numbered checklist and the student is told to ask about
  the ideas that were skipped; it must say so rather than guess when the
  notes cover nothing. One constant (TEACH_MAX_EXCHANGES = 10) ends the
  session, or the Finish button does.
- Recap: a teach_summary call (JSON, one repair retry) yields well / gaps
  with card references / next steps; each gap links to the card it means.
  "Add missed cards to review" schedules those activities due now; "Turn
  gaps into cards" generates one flashcard per gap from the deck's notes and
  shows an editable review list before saving.
- Sessions are chat conversations with a mode:"teach" marker + session
  record: same IndexedDB history, backup and sync, filtered out of the Chat
  screens; the recent list reopens a recap or continues a session.
- Reused pieces: MessageRow, Composer (now optional attach/scope chips and a
  mic), Markdown/KaTeX, AiDisclaimer, ContextPicker/ModelPicker, plus a
  shared useChatModels hook and useKeyboardInset extracted from Chat.
- Dictation: Web Speech API recognition (lib/speech-input.ts); the mic only
  renders where the browser supports it, and dictated phrases land in the
  draft.
- All strings in en + cs. teach-test: 44 checks (API prompts and guards,
  both entry points, the session, recap actions, history, exchange limit,
  360/390px layouts, mic on/off). Version 0.8.1, SW v103.

app/api/chat/route.ts

@@ -8,7 +8,7 @@ import { rejectCrossOrigin, rejectOversizedBody, requestLocale } from "@/lib/htt
 import { clientKey, rateLimit, rateLimitResponse } from "@/lib/rate-limit";
 import { acquireModelGate } from "@/lib/model-guard";
 import { translate } from "@/lib/i18n-dict";
-import { buildChatSystemPrompt } from "@/lib/prompts";
+import { buildChatSystemPrompt, buildTeachSummarySystemPrompt, buildTeachSystemPrompt } from "@/lib/prompts";
 import { estimateTokens, trimHistory } from "@/lib/chat-trim";
 
 export const runtime = "nodejs";
@@ -43,6 +43,17 @@ const requestSchema = z.object({
     )
     .max(40)
     .optional(),
+  /** Teach it back: "teach" is one student turn, "teach_summary" the recap. */
+  mode: z.enum(["teach", "teach_summary"]).optional(),
+  teach: z
+    .object({
+      topic: z.string().min(1).max(200),
+      difficulty: z.enum(["friendly", "skeptical"]),
+      exchange: z.number().int().min(0).max(60),
+      limit: z.number().int().min(1).max(60),
+      cards: z.array(z.string().max(1000)).max(80),
+    })
+    .optional(),
 });
 
 type StreamEvent =
@@ -97,7 +108,13 @@ export async function POST(request: Request) {
     );
   }
 
-  const { messages, context } = parsed.data;
+  const { messages, context, mode, teach } = parsed.data;
+  if (mode && !teach) {
+    return Response.json(
+      { error: translate(locale, "server.invalidRequest", { detail: "teach mode needs the teach payload" }) },
+      { status: 400 },
+    );
+  }
 
   // The model must be one the server actually offers — the configured
   // favourites, or anything the user's Ollama lists. The picker shows exactly
@@ -142,13 +159,29 @@ export async function POST(request: Request) {
   }
   const history = trimHistory(messages);
 
-  const system = buildChatSystemPrompt({
-    context: keptContext.map((chunk) => ({
-      deckTitle: chunk.deckTitle,
-      section: chunk.section,
-      text: chunk.text,
-    })),
-  });
+  const notesContext = keptContext.map((chunk) => ({
+    deckTitle: chunk.deckTitle,
+    section: chunk.section,
+    text: chunk.text,
+  }));
+  const system =
+    mode === "teach" && teach
+      ? buildTeachSystemPrompt({
+          topic: teach.topic,
+          difficulty: teach.difficulty,
+          cards: teach.cards,
+          context: notesContext,
+          exchange: teach.exchange,
+          limit: teach.limit,
+        })
+      : mode === "teach_summary" && teach
+        ? buildTeachSummarySystemPrompt({
+            topic: teach.topic,
+            difficulty: teach.difficulty,
+            cards: teach.cards,
+            context: notesContext,
+          })
+        : buildChatSystemPrompt({ context: notesContext });
 
   // Concurrency + the account-wide model budget (lib/model-guard), taken
   // after every cheap rejection so a 400 can never leak a slot.