@lightcone-ai/daemon 0.15.53 → 0.15.55

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

@@ -1,11 +0,0 @@
-export { analyzePage, analyzePageFromHtmlFixture } from './analyze-page.js';
-export { SITE_CORE_SELECTORS, SEMANTIC_CORE_SELECTORS, resolveSiteSelectors } from './site-selectors.js';
-export {
-  RECRUITMENT_SLOT_KEYS,
-  MIN_BIN_CHAR_COUNT,
-  MIN_TEXT_BIN_COUNT,
-  chooseUnderstandingPath,
-  hasEnoughTextBins,
-  normalizePageUnderstanding,
-  validatePageUnderstanding,
-} from './schema.js';

package/src/_vendor/video/understanding/llm-client.js

@@ -1,261 +0,0 @@
-const ANTHROPIC_API_URL = 'https://api.anthropic.com/v1/messages';
-
-function clampInt(value, min, max, fallback) {
-  const n = Number(value);
-  if (!Number.isFinite(n)) return fallback;
-  return Math.max(min, Math.min(max, Math.round(n)));
-}
-
-function parseJsonMaybe(text) {
-  try {
-    return JSON.parse(text);
-  } catch {
-    return null;
-  }
-}
-
-function extractFirstJsonObject(text) {
-  const input = String(text ?? '').trim();
-  if (!input) return null;
-  const direct = parseJsonMaybe(input);
-  if (direct && typeof direct === 'object') return direct;
-
-  const first = input.indexOf('{');
-  if (first < 0) return null;
-
-  let depth = 0;
-  for (let i = first; i < input.length; i += 1) {
-    const ch = input[i];
-    if (ch === '{') depth += 1;
-    if (ch === '}') depth -= 1;
-    if (depth !== 0) continue;
-
-    const slice = input.slice(first, i + 1);
-    const parsed = parseJsonMaybe(slice);
-    if (parsed && typeof parsed === 'object') return parsed;
-    break;
-  }
-
-  return null;
-}
-
-export function resolveAnthropicConfig(overrides = {}) {
-  const apiKey = String(
-    overrides.apiKey
-    ?? process.env.PAGE_UNDERSTANDING_ANTHROPIC_API_KEY
-    ?? process.env.ANTHROPIC_API_KEY
-    ?? ''
-  ).trim();
-
-  return {
-    apiKey,
-    model: String(
-      overrides.model
-      ?? process.env.PAGE_UNDERSTANDING_ANTHROPIC_MODEL
-      ?? 'claude-3-5-sonnet-20241022'
-    ).trim(),
-    maxTokens: clampInt(
-      overrides.maxTokens ?? process.env.PAGE_UNDERSTANDING_ANTHROPIC_MAX_TOKENS,
-      128,
-      4096,
-      1024
-    ),
-    temperature: Number.isFinite(Number(overrides.temperature))
-      ? Number(overrides.temperature)
-      : 0.1,
-    timeoutMs: clampInt(
-      overrides.timeoutMs ?? process.env.PAGE_UNDERSTANDING_ANTHROPIC_TIMEOUT_MS,
-      5000,
-      120000,
-      45000
-    ),
-  };
-}
-
-export function isAnthropicConfigured(config = resolveAnthropicConfig()) {
-  return Boolean(config.apiKey);
-}
-
-async function callAnthropic({
-  system,
-  userContent,
-  config = resolveAnthropicConfig(),
-}) {
-  if (!config.apiKey) {
-    throw new Error('anthropic_api_key_missing');
-  }
-
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(), config.timeoutMs);
-  try {
-    const response = await fetch(ANTHROPIC_API_URL, {
-      method: 'POST',
-      headers: {
-        'x-api-key': config.apiKey,
-        'anthropic-version': '2023-06-01',
-        'content-type': 'application/json',
-      },
-      body: JSON.stringify({
-        model: config.model,
-        max_tokens: config.maxTokens,
-        temperature: config.temperature,
-        system,
-        messages: [
-          {
-            role: 'user',
-            content: userContent,
-          },
-        ],
-      }),
-      signal: controller.signal,
-    });
-
-    const text = await response.text();
-    const payload = parseJsonMaybe(text) ?? { raw: text.slice(0, 1000) };
-    if (!response.ok) {
-      const details = payload?.error?.message ?? text.slice(0, 300);
-      throw new Error(`anthropic_http_${response.status}:${details}`);
-    }
-
-    const blocks = Array.isArray(payload.content) ? payload.content : [];
-    const answerText = blocks
-      .filter((item) => item?.type === 'text')
-      .map((item) => String(item.text ?? ''))
-      .join('\n')
-      .trim();
-
-    const json = extractFirstJsonObject(answerText);
-    if (!json) {
-      throw new Error('anthropic_json_parse_failed');
-    }
-
-    return {
-      model: payload.model ?? config.model,
-      data: json,
-      rawText: answerText,
-    };
-  } finally {
-    clearTimeout(timer);
-  }
-}
-
-function sanitizeHighlights(input, { fallbackChunk = 'unknown' } = {}) {
-  const list = Array.isArray(input) ? input : [];
-  return list
-    .map((row) => ({
-      y: Number.isFinite(Number(row?.y)) ? Math.round(Number(row.y)) : NaN,
-      from_chunk: String(row?.from_chunk ?? row?.fromChunk ?? fallbackChunk).trim() || fallbackChunk,
-      reason: String(row?.reason ?? '').trim(),
-    }))
-    .filter((row) => Number.isFinite(row.y))
-    .slice(0, 3);
-}
-
-export async function runTextLlm({
-  persona,
-  structure,
-  configOverrides = {},
-}) {
-  const config = resolveAnthropicConfig(configOverrides);
-  const bins = Array.isArray(structure?.bins) ? structure.bins : [];
-  const compactBins = bins.slice(0, 24).map((bin) => ({
-    y_center: Math.round(Number(bin?.y_center ?? 0)),
-    text: String(bin?.text ?? '').slice(0, 220),
-  }));
-
-  const userPrompt = JSON.stringify({
-    task: '阅读页面结构化文本 bin，输出 1-3 个 highlight',
-    persona: String(persona ?? '').trim() || '通用用户',
-    url: structure?.url ?? '',
-    hostname: structure?.hostname ?? '',
-    core_y_range: structure?.core_y_range ?? [],
-    focus_y_range: structure?.focus_y_range ?? [],
-    meta: structure?.meta ?? {},
-    bins: compactBins,
-    output_schema: {
-      core_message: 'string',
-      highlights: [{ y: 1200, from_chunk: 'bin_3', reason: 'string' }],
-    },
-  }, null, 2);
-
-  const result = await callAnthropic({
-    system: '你是页面理解助手。必须返回 JSON，不要返回额外解释。',
-    userContent: [{ type: 'text', text: userPrompt }],
-    config,
-  });
-
-  return {
-    core_message: String(result.data?.core_message ?? '').trim(),
-    highlights: sanitizeHighlights(result.data?.highlights, { fallbackChunk: 'bin' }),
-    model_used: result.model,
-  };
-}
-
-export async function runVisionLlm({
-  persona,
-  structure,
-  chunks,
-  configOverrides = {},
-}) {
-  const config = resolveAnthropicConfig(configOverrides);
-  const content = [];
-
-  for (const chunk of chunks) {
-    content.push({
-      type: 'image',
-      source: {
-        type: 'base64',
-        media_type: 'image/png',
-        data: chunk.image_base64,
-      },
-    });
-  }
-
-  const prompt = JSON.stringify({
-    task: '从页面截图中为短视频挑选 1-3 个 highlights',
-    persona: String(persona ?? '').trim() || '通用用户',
-    url: structure?.url ?? '',
-    hostname: structure?.hostname ?? '',
-    core_y_range: structure?.core_y_range ?? [],
-    focus_y_range: structure?.focus_y_range ?? [],
-    candidate_pool: Array.isArray(structure?.hotspots)
-      ? structure.hotspots.slice(0, 24).map((item) => ({
-        y: item.y,
-        type: item.type,
-        reason: item.reason,
-      }))
-      : [],
-    chunks: chunks.map((chunk) => ({
-      chunk_id: chunk.id,
-      y_start: chunk.y_start,
-      y_end: chunk.y_end,
-    })),
-    constraints: [
-      '只返回 JSON',
-      'highlights 数量 1-3',
-      'y 必须在 focus_y_range 内',
-      '优先 candidate_pool',
-    ],
-    output_schema: {
-      core_message: 'string',
-      highlights: [{ y: 1200, from_chunk: 'chunk_1', reason: 'string' }],
-    },
-  }, null, 2);
-
-  content.push({
-    type: 'text',
-    text: prompt,
-  });
-
-  const result = await callAnthropic({
-    system: '你是页面视觉理解助手。必须输出 JSON。',
-    userContent: content,
-    config,
-  });
-
-  return {
-    core_message: String(result.data?.core_message ?? '').trim(),
-    highlights: sanitizeHighlights(result.data?.highlights, { fallbackChunk: 'chunk' }),
-    model_used: result.model,
-  };
-}

package/src/_vendor/video/understanding/schema.js

@@ -1,254 +0,0 @@
-export const MIN_TEXT_BIN_COUNT = 10;
-export const MIN_BIN_CHAR_COUNT = 30;
-export const RECRUITMENT_SLOT_KEYS = Object.freeze([
-  'company',
-  'published_at',
-  'recruitment_type',
-  'cohort',
-  'job_directions',
-  'locations',
-  'target_or_requirements',
-  'process',
-  'entry_or_cta',
-]);
-
-const SLOT_SOURCE_TYPE_SET = new Set(['title', 'meta', 'body_text', 'image_ocr', 'heuristic']);
-const SLOT_STATUS_SET = new Set(['present', 'missing']);
-const MODE_HINT_SET = new Set([
-  'job_intel_broadcast',
-  'job_alert',
-  'info_summary',
-  'refuse_auto_broadcast',
-]);
-
-function clampInt(value, min, max, fallback) {
-  const n = Number(value);
-  if (!Number.isFinite(n)) return fallback;
-  return Math.max(min, Math.min(max, Math.round(n)));
-}
-
-function normalizeRange(input, { floor = 0, ceil = Number.MAX_SAFE_INTEGER, fallback = [0, 0] } = {}) {
-  const raw = Array.isArray(input) ? input : fallback;
-  const start = clampInt(raw[0], floor, ceil, fallback[0]);
-  const end = clampInt(raw[1], floor, ceil, fallback[1]);
-  if (end < start) return [start, start];
-  return [start, end];
-}
-
-function normalizeDurationRange(input, fallback = [35, 55]) {
-  const range = normalizeRange(input, { floor: 5, ceil: 240, fallback });
-  if (range[1] <= range[0]) return [range[0], range[0] + 5];
-  return range;
-}
-
-function normalizeConfidence(value, fallback = 0) {
-  const n = Number(value);
-  if (!Number.isFinite(n)) return fallback;
-  return Number(Math.max(0, Math.min(1, n)).toFixed(2));
-}
-
-function normalizeSlotValue(value) {
-  if (Array.isArray(value)) {
-    const seen = new Set();
-    const output = [];
-    for (const row of value) {
-      const text = String(row ?? '').trim();
-      if (!text) continue;
-      const key = text.toLowerCase();
-      if (seen.has(key)) continue;
-      seen.add(key);
-      output.push(text);
-    }
-    return output.length > 0 ? output : null;
-  }
-
-  const text = String(value ?? '').trim();
-  return text || null;
-}
-
-function normalizeSourceType(value) {
-  const sourceType = String(value ?? '').trim();
-  if (SLOT_SOURCE_TYPE_SET.has(sourceType)) return sourceType;
-  return 'heuristic';
-}
-
-function normalizeModeHint(value) {
-  const modeHint = String(value ?? '').trim();
-  if (MODE_HINT_SET.has(modeHint)) return modeHint;
-  return 'refuse_auto_broadcast';
-}
-
-function normalizeSlot(row, { totalHeight, fallbackSourceType = 'heuristic' } = {}) {
-  const sourceType = normalizeSourceType(row?.source_type ?? fallbackSourceType);
-  const value = normalizeSlotValue(row?.value);
-  const focusRegion = row?.focus_region == null
-    ? null
-    : normalizeRange(row?.focus_region, { floor: 0, ceil: totalHeight, fallback: [0, 0] });
-  const hasFocus = Array.isArray(focusRegion) && focusRegion[1] > focusRegion[0];
-  const status = SLOT_STATUS_SET.has(String(row?.status ?? '').trim())
-    ? String(row?.status ?? '').trim()
-    : (value == null ? 'missing' : 'present');
-
-  return {
-    value,
-    source_type: sourceType,
-    confidence: normalizeConfidence(row?.confidence, 0),
-    focus_region: hasFocus ? focusRegion : null,
-    status,
-  };
-}
-
-function normalizeSemanticSlots(input, { totalHeight }) {
-  const source = input && typeof input === 'object' && !Array.isArray(input) ? input : {};
-  const slots = {};
-
-  for (const key of RECRUITMENT_SLOT_KEYS) {
-    slots[key] = normalizeSlot(source[key], { totalHeight, fallbackSourceType: 'heuristic' });
-  }
-
-  return slots;
-}
-
-export function hasEnoughTextBins(
-  bins,
-  {
-    minCount = MIN_TEXT_BIN_COUNT,
-    minChars = MIN_BIN_CHAR_COUNT,
-  } = {}
-) {
-  if (!Array.isArray(bins)) return false;
-  const qualified = bins.filter((bin) => String(bin?.text ?? '').trim().length > minChars);
-  return qualified.length >= minCount;
-}
-
-export function chooseUnderstandingPath(structure, options = {}) {
-  return hasEnoughTextBins(structure?.bins, options) ? 'text' : 'vision';
-}
-
-function normalizeHighlights(input, { focusRange, fallbackReason = 'heuristic' } = {}) {
-  const list = Array.isArray(input) ? input : [];
-  const [focusStart, focusEnd] = normalizeRange(focusRange, { fallback: [0, 0] });
-  const output = [];
-
-  for (const row of list) {
-    const y = clampInt(row?.y, focusStart, focusEnd, NaN);
-    if (!Number.isFinite(y)) continue;
-    output.push({
-      y,
-      from_chunk: String(row?.from_chunk ?? row?.fromChunk ?? '').trim() || 'unknown',
-      reason: String(row?.reason ?? '').trim() || fallbackReason,
-    });
-    if (output.length >= 3) break;
-  }
-
-  return output;
-}
-
-function normalizeSkipZones(input, { totalHeight, coreRange }) {
-  const provided = Array.isArray(input) ? input : [];
-  const [coreStart, coreEnd] = normalizeRange(coreRange, { floor: 0, ceil: totalHeight, fallback: [0, totalHeight] });
-
-  const fallback = [];
-  if (coreStart > 0) {
-    fallback.push({ y_range: [0, coreStart], reason: 'non_core_top' });
-  }
-  if (coreEnd < totalHeight) {
-    fallback.push({ y_range: [coreEnd, totalHeight], reason: 'non_core_bottom' });
-  }
-
-  const merged = provided.length > 0 ? provided : fallback;
-  return merged
-    .map((zone) => ({
-      y_range: normalizeRange(zone?.y_range, { floor: 0, ceil: totalHeight, fallback: [0, 0] }),
-      reason: String(zone?.reason ?? '').trim() || 'non_core',
-    }))
-    .filter((zone) => zone.y_range[1] > zone.y_range[0]);
-}
-
-export function normalizePageUnderstanding(raw = {}) {
-  const totalHeight = clampInt(raw.total_height, 100, 200000, 1920);
-  const coreRange = normalizeRange(raw.core_y_range, { floor: 0, ceil: totalHeight, fallback: [0, totalHeight] });
-  const focusRange = normalizeRange(raw.focus_y_range, { floor: coreRange[0], ceil: totalHeight, fallback: coreRange });
-  const highlights = normalizeHighlights(raw.candidate_hotspots, { focusRange, fallbackReason: 'candidate' });
-  const semanticSlots = normalizeSemanticSlots(raw.semantic_slots, { totalHeight });
-  const modeHint = normalizeModeHint(raw.mode_hint);
-  const modeHintConfidence = normalizeConfidence(raw.mode_hint_confidence, 0.2);
-
-  return {
-    url: String(raw.url ?? '').trim(),
-    page_type: String(raw.page_type ?? 'generic_article').trim() || 'generic_article',
-    hostname: String(raw.hostname ?? '').trim(),
-    core_message: String(raw.core_message ?? '').trim(),
-    core_y_range: coreRange,
-    focus_y_range: focusRange,
-    total_height: totalHeight,
-    candidate_hotspots: highlights,
-    skip_zones: normalizeSkipZones(raw.skip_zones, { totalHeight, coreRange }),
-    recommended_duration_s: normalizeDurationRange(raw.recommended_duration_s),
-    semantic_slots: semanticSlots,
-    mode_hint: modeHint,
-    mode_hint_confidence: modeHintConfidence,
-    meta: raw.meta && typeof raw.meta === 'object' && !Array.isArray(raw.meta)
-      ? raw.meta
-      : {},
-  };
-}
-
-function isSlotValueValid(value) {
-  if (value == null) return true;
-  if (Array.isArray(value)) return value.every(item => typeof item === 'string');
-  return typeof value === 'string';
-}
-
-export function validatePageUnderstanding(payload) {
-  const errors = [];
-  const model = payload ?? {};
-
-  if (!model.url || typeof model.url !== 'string') errors.push('url required');
-  if (!model.hostname || typeof model.hostname !== 'string') errors.push('hostname required');
-  if (!Array.isArray(model.core_y_range) || model.core_y_range.length !== 2) errors.push('core_y_range invalid');
-  if (!Array.isArray(model.focus_y_range) || model.focus_y_range.length !== 2) errors.push('focus_y_range invalid');
-  if (!Number.isFinite(model.total_height) || model.total_height <= 0) errors.push('total_height invalid');
-  if (!Array.isArray(model.candidate_hotspots)) errors.push('candidate_hotspots invalid');
-  if (!Array.isArray(model.skip_zones)) errors.push('skip_zones invalid');
-  if (!Array.isArray(model.recommended_duration_s) || model.recommended_duration_s.length !== 2) {
-    errors.push('recommended_duration_s invalid');
-  }
-  if (!model.semantic_slots || typeof model.semantic_slots !== 'object' || Array.isArray(model.semantic_slots)) {
-    errors.push('semantic_slots invalid');
-  } else {
-    for (const key of RECRUITMENT_SLOT_KEYS) {
-      const slot = model.semantic_slots[key];
-      if (!slot || typeof slot !== 'object' || Array.isArray(slot)) {
-        errors.push(`semantic_slots.${key} invalid`);
-        continue;
-      }
-      if (!isSlotValueValid(slot.value)) errors.push(`semantic_slots.${key}.value invalid`);
-      if (!SLOT_SOURCE_TYPE_SET.has(String(slot.source_type ?? '').trim())) {
-        errors.push(`semantic_slots.${key}.source_type invalid`);
-      }
-      if (!Number.isFinite(Number(slot.confidence))) {
-        errors.push(`semantic_slots.${key}.confidence invalid`);
-      }
-      if (!SLOT_STATUS_SET.has(String(slot.status ?? '').trim())) {
-        errors.push(`semantic_slots.${key}.status invalid`);
-      }
-      if (slot.focus_region != null) {
-        if (!Array.isArray(slot.focus_region) || slot.focus_region.length !== 2) {
-          errors.push(`semantic_slots.${key}.focus_region invalid`);
-        }
-      }
-    }
-  }
-  if (!MODE_HINT_SET.has(String(model.mode_hint ?? '').trim())) {
-    errors.push('mode_hint invalid');
-  }
-  if (!Number.isFinite(Number(model.mode_hint_confidence))) {
-    errors.push('mode_hint_confidence invalid');
-  }
-
-  return {
-    ok: errors.length === 0,
-    errors,
-  };
-}

package/src/_vendor/video/understanding/site-selectors.js

@@ -1,47 +0,0 @@
-export const SITE_CORE_SELECTORS = Object.freeze({
-  'mp.weixin.qq.com': Object.freeze([
-    '#js_content',
-    '#img-content',
-    'article',
-    'main',
-  ]),
-  'zhuanlan.zhihu.com': Object.freeze([
-    '.Post-RichTextContainer',
-    '.RichText.ztext',
-    'article',
-    'main',
-  ]),
-  'jianshu.com': Object.freeze([
-    'article',
-    '.note .post',
-    '.article',
-    'main',
-  ]),
-  'juejin.cn': Object.freeze([
-    'article',
-    '.article-content',
-    '.main-area article',
-    'main',
-  ]),
-});
-
-export const SEMANTIC_CORE_SELECTORS = Object.freeze([
-  'article',
-  'main',
-  '[role="main"]',
-]);
-
-export function resolveSiteSelectors(hostname) {
-  const normalizedHost = String(hostname ?? '').trim().toLowerCase();
-  if (!normalizedHost) return null;
-
-  for (const [domain, selectors] of Object.entries(SITE_CORE_SELECTORS)) {
-    if (!normalizedHost.includes(domain)) continue;
-    return {
-      domain,
-      selectors: [...selectors],
-    };
-  }
-
-  return null;
-}