@vanillaskyai/video 0.10.13 → 0.10.15

package/CHANGELOG.md

@@ -4,6 +4,14 @@ VanillaSky follows semantic versioning. This changelog begins with the 0.1 beta.
 
 ## Unreleased
 
+## 0.10.15
+
+- Recover a mislabeled first chat brief only when its complete authored content validates, preserving its shots and ending. Keep malformed JSON and incomplete or later records rejected.
+
+## 0.10.14
+
+- Attach content-free record-shape diagnostics to rejected chat plans without accepting alternate formats or exposing model output.
+
 ## 0.10.13
 
 - Plan concise AI answers around the configured video-attempt budget, including the ending, while retaining complete authored narration and chapter recovery.

package/dist/server.js

@@ -892,6 +892,24 @@ function continueAfterOpening(narration, earlier) {
 
 // src/server/chat-shot-planner.ts
 var object = (value) => value && typeof value === "object" && !Array.isArray(value) ? value : void 0;
+function planShapeError(value) {
+  const part = object(value);
+  const has = (key) => Boolean(part && Object.hasOwn(part, key));
+  const shape = value === null ? "null" : Array.isArray(value) ? "array" : typeof value;
+  const discriminator = !has("type") ? "missing" : part.type === "answer" || part.type === "shot" ? part.type : typeof part.type === "string" ? "other-string" : "non-string";
+  return new Error("Chat plan requires an answer brief followed by shots", { cause: {
+    code: "chat_plan_shape",
+    shape,
+    discriminator,
+    fields: {
+      opening: has("opening"),
+      subject: has("subject"),
+      development: has("development"),
+      visualDirection: has("visualDirection"),
+      ending: has("ending")
+    }
+  } });
+}
 function text(value, maximum) {
   return typeof value === "string" && value.trim().length <= maximum ? value.trim() : "";
 }
@@ -934,6 +952,21 @@ function readShot(value, clipDurationSec, answerSubject = "") {
     continuity: item?.continuity === "continue" ? "continue" : "cut"
   };
 }
+function recoverFirstBrief(part, clipDurationSec) {
+  if (!part || typeof part.type !== "string" || !part.type.trim() || part.type === "answer" || part.type === "shot") return;
+  const bounded = (value, maximum, allowEmpty = false) => typeof value === "string" && value.trim().length <= maximum && (allowEmpty || Boolean(value.trim()));
+  if (!["opening", "subject", "development", "visualDirection", "ending"].every((key) => Object.hasOwn(part, key)) || !bounded(part.opening, 300) || !bounded(part.subject, 80) || !bounded(part.development, 2e3, true) || !bounded(part.visualDirection, 600)) return;
+  const ending = object(part.ending);
+  if (!ending || !bounded(ending.narration, 2e3) || !bounded(ending.subject, 80, true) || !bounded(ending.action, 600, true) || Object.hasOwn(ending, "title") && !bounded(ending.title, 65) || typeof ending.durationSec !== "number" || !Number.isFinite(ending.durationSec) || ending.continuity !== "cut" && ending.continuity !== "continue") return;
+  const subject = text(part.subject, 80);
+  return {
+    opening: text(part.opening, 300),
+    subject,
+    development: text(part.development, 2e3),
+    visualDirection: text(part.visualDirection, 600),
+    ending: readShot(ending, clipDurationSec, subject)
+  };
+}
 function replaceStream(source, textStream) {
   if (!(typeof source === "object" && source != null && "textStream" in source)) return textStream;
   return new Proxy({ textStream }, {
@@ -966,7 +999,7 @@ function createChatShotPlanner(options) {
       const upstream = typeof source === "object" && source != null && "textStream" in source ? source.textStream : source;
       const translated = (async function* () {
         let brief, buffer = "", index = 0, bodyDuration = 0, lastNarration = "";
-        let firstBody = true;
+        let firstBody = true, recordsSeen = 0;
         const reject = (cause) => {
           incomplete.add(context);
           const error = cause instanceof Error ? cause : new Error(String(cause));
@@ -993,7 +1026,15 @@ function createChatShotPlanner(options) {
         const line = (raw) => {
           const trimmed = raw.trim();
           if (!trimmed || /^```(?:json|ndjson)?$/i.test(trimmed)) return;
-          const part = object(JSON.parse(trimmed));
+          const firstRecord = recordsSeen++ === 0;
+          const value = JSON.parse(trimmed);
+          const part = object(value);
+          const recovered = firstRecord && !brief && index === 0 ? recoverFirstBrief(part, clipDurationSec) : void 0;
+          if (recovered) {
+            brief = recovered;
+            options.publishOpening({ line: brief.opening, keyword: brief.subject });
+            return;
+          }
           if (part?.type === "answer") {
             if (brief) throw new Error("Chat answer brief was emitted more than once");
             brief = { opening: text(part.opening, 300), subject: text(part.subject, 80), visualDirection: text(part.visualDirection, 600), development: text(part.development, 2e3) };
@@ -1007,7 +1048,7 @@ function createChatShotPlanner(options) {
             options.publishOpening(brief.opening ? { line: brief.opening, keyword: brief.subject } : void 0);
             return;
           }
-          if (part?.type !== "shot") throw new Error("Chat plan requires an answer brief followed by shots");
+          if (part?.type !== "shot") throw planShapeError(value);
           if (!brief) throw new Error("Chat shot arrived before its answer brief");
           const shot = readShot(part, clipDurationSec, brief.subject);
           if (shot.narration === brief.ending?.narration) return;
@@ -1174,6 +1215,7 @@ async function* resolveShots(parts, context, options) {
 // src/server/video-chat-prompts.ts
 function createVideoChatResponseInstructions(generatedVideoAvailable, openingAlreadyProvided = false, maxGeneratedVideos = 5, clipDurationSec = 5, mode = "cinematic") {
   return [
+    'Use the exact record type "answer" for the first brief and "shot" for developing beats. Output JSON records only, with no prose outside them, including when explaining a limitation.',
     "Write a complete, intentful video answer as newline-delimited JSON. Match the user's form and tone; mixed intents can combine directions.",
     `First write one brief: {"type":"answer","intent":"explanation|story|comedy|imagination|practical","opening":"a short inviting spoken introduction of 6\u20139 words","subject":"literal visual subject","development":"the essential development of this answer","visualDirection":"consistent subjects, appearance and visual approach","ending":{"title":"short meaningful chapter title, at most 65 characters","narration":"the authored payoff","subject":"literal subject","action":"visible action or change","durationSec":${clipDurationSec},"continuity":"cut|continue"}}.`,
     `Then stream each developing shot on its own line: {"type":"shot","title":"short meaningful chapter title, at most 65 characters","narration":"the exact spoken beat","subject":"2\u20138 literal filmable words, at most 80 characters","action":"concrete subject, action or visible change and useful framing","durationSec":${clipDurationSec},"continuity":"cut|continue"}.`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vanillaskyai/video",
-  "version": "0.10.13",
+  "version": "0.10.15",
   "description": "Open-source voice-and-video chat SDK for AI applications.",
   "keywords": [
     "video-chat",

package/starters/video-chat/package.json

@@ -9,7 +9,7 @@
     "preview": "vite preview"
   },
   "dependencies": {
-    "@vanillaskyai/video": "0.10.13",
+    "@vanillaskyai/video": "0.10.15",
     "react": "^19.2.8",
     "react-dom": "^19.2.8",
     "@ai-sdk/anthropic": "^3.0.0",