@lightcone-ai/daemon 0.15.32 → 0.15.34

package/mcp-servers/official/video-narration-planner/core.js

@@ -1097,110 +1097,6 @@ function resolveVoiceForPhase({ profile, role, index = 0 }) {
   };
 }
 
-function resolveSummaryValue(strategy = {}, slotKey) {
-  const summary = toSafeString(strategy?.semantic_summary?.[slotKey]);
-  if (summary) return summary;
-  const slot = strategy?.semantic_slots?.[slotKey];
-  if (!slot) return '';
-  if (slotKey === 'recruitment_type') return formatRecruitmentType(slot) || slotValueToText(slot);
-  return slotValueToText(slot);
-}
-
-function describeHighlightFocus(phase) {
-  const semanticLabel = slotLabel(toSafeString(phase?.semantic_slot));
-  const text = toSafeString(phase?.highlight?.text || phase?.highlight?.id || semanticLabel || phase?.visual_action?.target_hotspot);
-  if (!text) return '页面核心信息';
-  return text;
-}
-
-function buildPhaseSentence({ phase, strategy, phaseIndex }) {
-  const role = toSafeString(phase?.role || '').toLowerCase() || (phase?.phase_id === 'cta' ? 'cta' : 'highlight');
-  const mode = normalizeModeHint(strategy?.narration_mode) || 'info_summary';
-  const slotKey = toSafeString(phase?.semantic_slot);
-  const slotText = resolveSummaryValue(strategy, slotKey);
-  const slotTextOrFallback = slotText || describeHighlightFocus(phase);
-  const company = resolveSummaryValue(strategy, 'company') || '该公司';
-  const publishedAt = resolveSummaryValue(strategy, 'published_at') || '近期';
-  const recruitmentType = resolveSummaryValue(strategy, 'recruitment_type') || '招聘信息';
-  const cohort = resolveSummaryValue(strategy, 'cohort');
-  const entry = resolveSummaryValue(strategy, 'entry_or_cta') || '官方岗位入口';
-  const locations = resolveSummaryValue(strategy, 'locations');
-  const jobDirections = resolveSummaryValue(strategy, 'job_directions');
-  const process = resolveSummaryValue(strategy, 'process');
-  const targetOrRequirements = resolveSummaryValue(strategy, 'target_or_requirements');
-
-  if (role === 'hook') {
-    if (mode === 'job_intel_broadcast') {
-      const cohortPart = cohort ? `${cohort} ` : '';
-      return ensureSentenceLength(
-        `岗位情报先划重点：${company}在${publishedAt}发布了${cohortPart}${recruitmentType}。接下来我会按岗位方向、城市和流程三块快速过一遍，只保留能直接支持投递判断的信息。`
-      );
-    }
-
-    if (mode === 'job_alert') {
-      return ensureSentenceLength(
-        `这是一条岗位提醒：${company}在${publishedAt}发布了${recruitmentType}。当前页面细节不完整，我只说已经能确认的事实，帮你先判断要不要继续花时间深挖。`
-      );
-    }
-
-    if (mode === 'refuse_auto_broadcast') {
-      return ensureSentenceLength(
-        '当前页面缺少稳定的公司、岗位或发布时间线索，我不能直接生成岗位情报式结论。为了避免误导，下面仅给你保守处理建议。'
-      );
-    }
-
-    return ensureSentenceLength(
-      '这条链接和招聘信息相关，但更像资讯或公告汇总。下面我只提可核验线索，帮助你快速决定是否跳转到官方岗位页继续确认。'
-    );
-  }
-
-  if (role === 'cta') {
-    return null;
-  }
-
-  if (mode === 'job_intel_broadcast') {
-    if (slotKey === 'job_directions') {
-      return ensureSentenceLength(
-        `岗位方向目前可确认为${slotTextOrFallback}。这直接决定你是否值得继续投入，建议先对照自己的专业或技能栈做第一轮筛选，再看后续细节。`
-      );
-    }
-    if (slotKey === 'locations') {
-      return ensureSentenceLength(
-        `工作城市线索是${slotTextOrFallback}。这一步主要用于判断通勤与生活成本是否可接受，先把城市匹配度过一遍，能明显减少后续无效投递。`
-      );
-    }
-    if (slotKey === 'process') {
-      return ensureSentenceLength(
-        `招聘流程可读到${slotTextOrFallback}。你可以按这个顺序预留时间准备简历、测评和面试节点，避免临近截止才发现准备节奏不够。`
-      );
-    }
-    if (slotKey === 'target_or_requirements') {
-      return ensureSentenceLength(
-        `任职对象与要求显示为${slotTextOrFallback}。这一段最适合做硬性条件核对，先看是否满足门槛，再决定要不要继续投入申请材料。`
-      );
-    }
-    return ensureSentenceLength(
-      `补充一个关键信息：${slotLabel(slotKey) || '重点线索'}是${slotTextOrFallback}。这条信息能帮助你更快判断岗位匹配度，避免被页面里次要内容分散注意力。`
-    );
-  }
-
-  if (mode === 'job_alert') {
-    return ensureSentenceLength(
-      `目前页面能稳定识别的${slotLabel(slotKey) || '线索'}是${slotTextOrFallback}。其余细节暂不充足，建议你把它当作提醒信号，再回到原文逐条核验。`
-    );
-  }
-
-  if (mode === 'refuse_auto_broadcast') {
-    return ensureSentenceLength(
-      `这一段的${slotLabel(slotKey) || '线索'}置信度仍然偏低，我不会把“${slotTextOrFallback}”当成可直接投递依据。先补充权威来源，再做下一步判断更安全。`
-    );
-  }
-
-  const summaryPieces = [jobDirections, locations, process, targetOrRequirements].filter(Boolean).slice(0, 2);
-  return ensureSentenceLength(
-    `第${phaseIndex + 1}个可核验线索是${slotTextOrFallback}。目前它更适合作为招聘资讯摘要，${summaryPieces.length > 0 ? `可结合${summaryPieces.join('、')}交叉判断` : `建议与官方岗位页交叉核对`}，不要直接当成完整岗位说明。`
-  );
-}
 
 function estimateDurationMs(text, speed = 1) {
   const chars = Math.max(1, toCharLength(text));
@@ -1342,6 +1238,7 @@ function shouldTryTts({ env, workspaceId }) {
 
 export async function detailSections({
   strategy = {},
+  sentences = null,
   workspace_id = null,
   credential_id = null,
   format = 'mp3',
@@ -1358,6 +1255,21 @@ export async function detailSections({
     );
   }
 
+  if (!Array.isArray(sentences) || sentences.length === 0) {
+    throw new Error(
+      'sentences are required. Write one narration sentence per phase '
+      + '(hook, highlight_1…N, cta) based on the semantic_slots and persona from plan_video. '
+      + 'Use natural peer-voice tone for the target platform — not a news broadcast.'
+    );
+  }
+
+  const sentenceMap = {};
+  for (const item of sentences) {
+    const phaseId = toSafeString(item?.phase_id);
+    const text = toSafeString(item?.text);
+    if (phaseId && text) sentenceMap[phaseId] = text;
+  }
+
   const profile = getPlatformProfile(strategy?.target_platform);
   const phases = normalizePhaseList(strategy);
 
@@ -1367,15 +1279,23 @@ export async function detailSections({
 
   for (let i = 0; i < phases.length; i += 1) {
     const phase = phases[i];
-    const role = toSafeString(phase?.role).toLowerCase() || (phase?.phase_id === 'cta' ? 'cta' : (phase?.phase_id === 'hook' ? 'hook' : 'highlight'));
+    const phaseId = toSafeString(phase.phase_id) || `phase_${i + 1}`;
+    const role = toSafeString(phase?.role).toLowerCase() || (phaseId === 'cta' ? 'cta' : (phaseId === 'hook' ? 'hook' : 'highlight'));
 
-    const sentence = buildPhaseSentence({
-      phase,
-      strategy,
-      phaseIndex: i,
-    });
+    const sentence = sentenceMap[phaseId];
+    if (!sentence) {
+      throw new Error(
+        `missing sentence for phase "${phaseId}". Provide a text sentence for every phase in the plan.`
+      );
+    }
 
-    if (sentence == null) continue;
+    const charLen = toCharLength(sentence);
+    if (charLen < SENTENCE_MIN_CHARS) {
+      throw new Error(`sentence for "${phaseId}" too short (${charLen} chars, min ${SENTENCE_MIN_CHARS}).`);
+    }
+    if (charLen > SENTENCE_MAX_CHARS) {
+      throw new Error(`sentence for "${phaseId}" too long (${charLen} chars, max ${SENTENCE_MAX_CHARS}). Shorten to fit one TTS segment.`);
+    }
 
     const voice = resolveVoiceForPhase({ profile, role, index: i });
     const phaseBudgetMs = clampInt((toFiniteNumber(phase?.duration_s) ?? 6) * 1000, 1000, 180000, 6000);
@@ -1398,9 +1318,9 @@ export async function detailSections({
         usedLiveTts = true;
       } catch (error) {
         const message = toSafeString(error?.message) || 'tts_call_failed';
-        ttsErrors.push({ phase_id: phase.phase_id, error: message });
+        ttsErrors.push({ phase_id: phaseId, error: message });
         if (strict_tts) {
-          throw new Error(`detail_sections_tts_failed:${phase.phase_id}:${message}`);
+          throw new Error(`detail_sections_tts_failed:${phaseId}:${message}`);
         }
       }
     }
@@ -1408,7 +1328,7 @@ export async function detailSections({
     if (!audio) {
       const estimatedDuration = estimateDurationMs(sentence, voice.speed);
       audio = {
-        audio_id: `estimated-${phase.phase_id || `p${i + 1}`}`,
+        audio_id: `estimated-${phaseId}`,
         duration_ms: estimatedDuration,
         audio_url: null,
         audio_path: null,
@@ -1430,7 +1350,7 @@ export async function detailSections({
     );
 
     sections.push({
-      phase_id: toSafeString(phase.phase_id) || `phase_${i + 1}`,
+      phase_id: phaseId,
       visual_action: visualAction,
       sentence,
       voice_preset: voice.voice_preset,

package/mcp-servers/official/video-narration-planner/index.js

@@ -84,9 +84,19 @@ export function createVideoNarrationPlannerServer({
 
   server.tool(
     'detail_sections',
-    'Stage 3: expand each phase into sentence+voice settings, call TTS voiceover, and fill duration/dwell.',
+    'Stage 3: take agent-written sentences, call TTS voiceover for each phase, and fill duration/dwell. You MUST write the sentences yourself before calling this tool.',
     {
       strategy: z.record(z.any()).describe('Output of plan_video (video strategy with phase_plan).'),
+      sentences: z.array(z.object({
+        phase_id: z.string().describe('Phase ID from plan_video phase_plan (e.g. hook, highlight_1, cta).'),
+        text: z.string().describe('Narration sentence you wrote for this phase.'),
+      })).describe(
+        'Narration sentences YOU write, one per phase in the plan. '
+        + 'Use semantic_slots, persona, and platform tone from plan_video output. '
+        + 'Sound like a peer sharing useful info — not a news anchor. '
+        + 'Each sentence: 8–60 chars, spoken Chinese, ends with punctuation. '
+        + 'Do NOT use "信息来源为", "发布时间为", "凭截图判断" or similar metadata/warning phrasing.'
+      ),
       workspace_id: z.string().optional().describe('Workspace id for TTS generation. Defaults to WORKSPACE_ID env if provided.'),
       credential_id: z.string().optional().describe('Optional explicit tts_provider credential id.'),
       format: z.enum(['mp3', 'wav', 'flac']).optional().describe('Desired audio format for generated voiceover.'),

package/mcp-servers/official/video-narration-planner/planner-config.json

@@ -1,8 +1,8 @@
 {
   "default_outro_video_id": "outro-default-zh",
   "sentence_char_range": {
-    "min": 100,
-    "max": 150
+    "min": 14,
+    "max": 60
   },
   "platform_profiles": {
     "xiaohongshu": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lightcone-ai/daemon",
-  "version": "0.15.32",
+  "version": "0.15.34",
   "type": "module",
   "main": "src/index.js",
   "bin": {

package/src/_vendor/video/composer/index.js

@@ -301,13 +301,30 @@ function msToAssTimestamp(ms) {
   return `${hr}:${String(min).padStart(2, '0')}:${String(sec).padStart(2, '0')}.${String(cs).padStart(2, '0')}`;
 }
 
+// Hard-wrap CJK subtitle text so it never overflows the video frame.
+// libass WrapStyle:0 doesn't handle Chinese text reliably (no word boundaries),
+// so we insert explicit \N breaks every maxChars characters.
+function wrapSubtitleText(text, maxChars = 14) {
+  const chars = Array.from(String(text ?? ''));
+  if (chars.length <= maxChars) return chars.join('');
+  const lines = [];
+  for (let i = 0; i < chars.length; i += maxChars) {
+    lines.push(chars.slice(i, i + maxChars).join(''));
+  }
+  return lines.join('\\N');
+}
+
 export function buildAssContent(subtitles = [], { playResX = 1080, playResY = 1920 } = {}) {
+  // Max chars per line: (playResX - marginL - marginR) / fontSizePx
+  // 1080 - 30 - 30 = 1020px, fontsize 72 ≈ 72px/char → 14 chars
+  const maxCharsPerLine = Math.floor((playResX - 60) / SUBTITLE_FONT_SIZE);
+
   const header = [
     '[Script Info]',
     'ScriptType: v4.00+',
     `PlayResX: ${playResX}`,
     `PlayResY: ${playResY}`,
-    'WrapStyle: 0',
+    'WrapStyle: 2',
     '',
     '[V4+ Styles]',
     'Format: Name, Fontname, Fontsize, PrimaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding',
@@ -318,7 +335,8 @@ export function buildAssContent(subtitles = [], { playResX = 1080, playResY = 19
   ].join('\n');
 
   const events = subtitles.map(({ text, start_ms, end_ms }) => {
-    const safeText = String(text ?? '').replace(/\r?\n/g, '\\N').replace(/,/g, '{\\,}');
+    const wrapped = wrapSubtitleText(text, maxCharsPerLine);
+    const safeText = wrapped.replace(/\r?\n/g, '\\N').replace(/,/g, '{\\,}');
     return `Dialogue: 0,${msToAssTimestamp(start_ms)},${msToAssTimestamp(end_ms)},Default,,0,0,0,,${safeText}`;
   });