fix: 关闭新闻放宽凑数并剥离放宽窗口文案

Co-authored-by: Cursor <cursoragent@cursor.com>

daily/news/fetch.py

@@ -8,17 +8,19 @@ import time
 import html
 import xml.etree.ElementTree as ET
 from concurrent.futures import ThreadPoolExecutor, as_completed
-from datetime import datetime, timezone, timedelta
+from datetime import datetime, time as dt_time, timezone, timedelta
 from email.utils import parsedate_to_datetime
 from typing import Any
 from urllib.parse import parse_qs, urlencode, urlparse, urlunparse
+from zoneinfo import ZoneInfo
 
 import certifi
 import httpx
 
-from daily.config import env, env_int, news_summary_limit
+from daily.config import env, env_int, news_summary_limit, wecom_news_desc_limit
 from daily.news.feeds import NEWS_CATEGORIES, NewsCategory, NewsFeed
 from daily.news.feeds_cn import CN_AI_TITLE_KEYWORDS, CN_NEWS_CATEGORIES
+from daily.text_utils import trim_brief
 
 logger = logging.getLogger(__name__)
 
@@ -49,12 +51,37 @@ def _hours_window() -> int:
     return max(1, env_int("DAILY_AI_NEWS_HOURS", 24))
 
 
+def _news_tz_name() -> str:
+    return (env("DAILY_AI_NEWS_TZ") or env("DAILY_SCHEDULE_TZ") or "Asia/Shanghai").strip()
+
+
+def _floor_today_enabled() -> bool:
+    raw = env("DAILY_AI_NEWS_FLOOR_TODAY")
+    if raw is None:
+        return True
+    return raw.strip().lower() not in {"0", "false", "no", "off"}
+
+
+def _cutoff_datetime(*, floor_today: bool) -> datetime:
+    """滚动 N 小时窗口；国际新闻可叠加「不早于今日 0 点（本地时区）」。"""
+    now = _now_utc()
+    rolling = now - timedelta(hours=_hours_window())
+    if not floor_today:
+        return rolling
+    tz = ZoneInfo(_news_tz_name())
+    local = now.astimezone(tz)
+    start_today = local.replace(hour=0, minute=0, second=0, microsecond=0).astimezone(timezone.utc)
+    return max(rolling, start_today)
+
+
 def _per_feed_limit() -> int:
-    return max(1, env_int("DAILY_AI_NEWS_PER_FEED", 3))
+    want = max(_wecom_limit(), _wecom_cn_limit())
+    return max(want // 2, env_int("DAILY_AI_NEWS_PER_FEED", 5))
 
 
 def _per_category_limit() -> int:
-    return max(1, env_int("DAILY_AI_NEWS_PER_CATEGORY", 5))
+    want = max(_wecom_limit(), _wecom_cn_limit())
+    return max(want, env_int("DAILY_AI_NEWS_PER_CATEGORY", 10))
 
 
 def _wecom_limit() -> int:
@@ -62,7 +89,7 @@ def _wecom_limit() -> int:
 
 
 def _wecom_cn_limit() -> int:
-    return max(1, env_int("DAILY_WECOM_CN_AI_NEWS", 8))
+    return max(1, env_int("DAILY_WECOM_CN_AI_NEWS", 10))
 
 
 def _matches_cn_ai_title(title: str) -> bool:
@@ -102,7 +129,6 @@ def _parse_datetime(value: str | None) -> datetime | None:
     for fmt in (
         "%Y-%m-%dT%H:%M:%SZ",
         "%Y-%m-%dT%H:%M:%S%z",
-        "%Y-%m-%d",
     ):
         try:
             dt = datetime.strptime(text[: len(fmt.replace("%z", "+0000"))], fmt.replace("%z", ""))
@@ -111,17 +137,103 @@ def _parse_datetime(value: str | None) -> datetime | None:
             return dt.astimezone(timezone.utc)
         except ValueError:
             continue
+    if re.match(r"^\d{4}-\d{2}-\d{2}$", text):
+        try:
+            tz = ZoneInfo(_news_tz_name())
+            day = datetime.strptime(text, "%Y-%m-%d").date()
+            # 仅日期时按本地中午估算，避免 UTC 0 点误判为「前一天」
+            dt = datetime.combine(day, dt_time(12, 0), tzinfo=tz)
+            return dt.astimezone(timezone.utc)
+        except ValueError:
+            pass
     return None
 
 
 def _clean_text(text: str | None, limit: int = 200) -> str:
     if not text:
         return ""
-    plain = STRIP_HTML.sub(" ", html.unescape(text))
-    plain = WS.sub(" ", plain).strip()
+    plain = _strip_summary_plain(text)
     if limit <= 0 or len(plain) <= limit:
         return plain
-    return plain[: limit - 3] + "..."
+    return trim_brief(plain, limit)
+
+
+def _strip_summary_plain(text: str | None) -> str:
+    if not text:
+        return ""
+    plain = STRIP_HTML.sub(" ", html.unescape(text))
+    return WS.sub(" ", plain).strip()
+
+
+_JUNK_SUMMARY_RE = re.compile(
+    r"^(点击查看原文|article url:|comments url:|discussion on hn|read more)",
+    re.IGNORECASE,
+)
+
+
+def _is_junk_news_summary(text: str) -> bool:
+    if not text:
+        return True
+    if _JUNK_SUMMARY_RE.match(text.strip()):
+        return True
+    if text.strip().endswith(">") and "点击" in text:
+        return True
+    return False
+
+
+def brief_news_summary(text: str | None, limit: int | None = None) -> str:
+    """企微新闻一句摘要：去 HTML、过滤占位文案、句读处截断。"""
+    plain = _strip_summary_plain(text)
+    if _is_junk_news_summary(plain):
+        return ""
+    lim = wecom_news_desc_limit() if limit is None else limit
+    return trim_brief(plain, lim)
+
+
+def sync_wecom_news_rows(items: list[dict[str, Any]], flat: list[dict[str, Any]]) -> None:
+    """中文化后，用 flat 最新 summary 刷新企微 desc_short。"""
+    by_link = {_normalize_link(str(i.get("link") or "")): i for i in flat if i.get("link")}
+    for row in items:
+        link = _normalize_link(str(row.get("link") or ""))
+        src = by_link.get(link)
+        if src:
+            row["desc_short"] = brief_news_summary(src.get("summary"))
+
+
+def finalize_wecom_news_items(
+    items: list[dict[str, Any]],
+    *,
+    force_chinese: bool = False,
+) -> None:
+    """企微新闻摘要：确保 desc_short 为中文（国际源 force_chinese=True）。"""
+    from daily.localize import LocalizeJob, localize_brief_descriptions, needs_chinese
+    from daily.news.sanitize import strip_relax_window_prefix
+
+    limit = wecom_news_desc_limit()
+    jobs: list[LocalizeJob] = []
+    keyed: list[tuple[str, dict[str, Any]]] = []
+    for idx, item in enumerate(items):
+        text = strip_relax_window_prefix(
+            (item.get("desc_short") or item.get("summary_plain") or "").strip()
+        )
+        if text:
+            item["desc_short"] = text
+        if not text or _is_junk_news_summary(text):
+            item["desc_short"] = ""
+            continue
+        if force_chinese or needs_chinese(text):
+            key = f"wecom-news:{item.get('link') or idx}"
+            jobs.append(LocalizeJob(key, text, limit))
+            keyed.append((key, item))
+        elif not item.get("desc_short"):
+            item["desc_short"] = brief_news_summary(text, limit)
+
+    if not jobs:
+        return
+    zh_map = localize_brief_descriptions(jobs, archive=True)
+    for key, item in keyed:
+        if key in zh_map:
+            item["desc_short"] = strip_relax_window_prefix(zh_map[key])
 
 
 def _normalize_link(link: str) -> str:
@@ -317,10 +429,119 @@ def _dedupe_items(items: list[dict[str, Any]]) -> list[dict[str, Any]]:
     return result
 
 
+def _filter_flat_in_window(
+    flat: list[dict[str, Any]],
+    *,
+    floor_today: bool,
+) -> list[dict[str, Any]]:
+    cutoff = _cutoff_datetime(floor_today=floor_today)
+    items = [i for i in _dedupe_items(flat) if _within_window(i, cutoff)]
+    items.sort(key=_sort_key, reverse=True)
+    return items
+
+
+def _pick_news_items(
+    flat: list[dict[str, Any]],
+    limit: int,
+    preferred: tuple[str, ...],
+    *,
+    one_per_source: bool = False,
+) -> list[dict[str, Any]]:
+    picked: list[dict[str, Any]] = []
+    seen_links: set[str] = set()
+    seen_sources: set[str] = set()
+
+    def _try_take(item: dict[str, Any]) -> bool:
+        link = _normalize_link(item.get("link", ""))
+        if not link or link in seen_links:
+            return False
+        if one_per_source:
+            source = item.get("source_name", "?")
+            if source in seen_sources:
+                return False
+            seen_sources.add(source)
+        seen_links.add(link)
+        picked.append(item)
+        return True
+
+    for cat in preferred:
+        for item in flat:
+            if item.get("category_id") != cat:
+                continue
+            if _try_take(item) and len(picked) >= limit:
+                return picked[:limit]
+
+    for item in flat:
+        if _try_take(item) and len(picked) >= limit:
+            break
+    return picked[:limit]
+
+
+def _fill_picked_to_limit(
+    picked: list[dict[str, Any]],
+    pools: list[list[dict[str, Any]]],
+    limit: int,
+) -> list[dict[str, Any]]:
+    seen_links = {_normalize_link(i.get("link", "")) for i in picked}
+    for pool in pools:
+        for item in pool:
+            if len(picked) >= limit:
+                return picked[:limit]
+            link = _normalize_link(item.get("link", ""))
+            if not link or link in seen_links:
+                continue
+            picked.append(item)
+            seen_links.add(link)
+    return picked[:limit]
+
+
+def _to_wecom_news_row(item: dict[str, Any]) -> dict[str, Any]:
+    plain = _strip_summary_plain(item.get("summary", ""))
+    return {
+        "title": item.get("title", "?"),
+        "link": item.get("link", ""),
+        "source_name": item.get("source_name", "?"),
+        "published_fmt": item.get("published_fmt", ""),
+        "desc_short": brief_news_summary(plain),
+        "summary_plain": plain,
+    }
+
+
+def _apply_pushed_dedup_with_backfill(
+    items: list[dict[str, Any]],
+    picked: list[dict[str, Any]],
+    *,
+    date_str: str | None,
+    limit: int,
+) -> list[dict[str, Any]]:
+    if not date_str:
+        return items[:limit]
+    from daily.config import news_backfill_enabled
+    from daily.news.pushed_links import filter_unpushed_items
+
+    fresh = filter_unpushed_items(items, date_str=date_str)
+    if len(fresh) >= limit:
+        return fresh[:limit]
+    if not news_backfill_enabled():
+        if len(fresh) < limit:
+            logger.info("news_short:%s", len(fresh))
+        return fresh[:limit]
+    seen = {_normalize_link(i.get("link", "")) for i in fresh if i.get("link")}
+    for item in picked:
+        if len(fresh) >= limit:
+            break
+        link = _normalize_link(item.get("link", ""))
+        if not link or link in seen:
+            continue
+        fresh.append(_to_wecom_news_row(item))
+        seen.add(link)
+    return fresh[:limit]
+
+
 def _within_window(item: dict[str, Any], cutoff: datetime) -> bool:
     dt = _entry_datetime(item)
     if dt is None:
-        return True
+        return False
     return dt >= cutoff
 
 
@@ -331,11 +552,11 @@ def _sort_key(item: dict[str, Any]) -> tuple[int, datetime]:
     return (0, dt)
 
 
-def _fetch_news(categories: tuple[NewsCategory, ...]) -> dict[str, Any]:
+def _fetch_news(categories: tuple[NewsCategory, ...], *, floor_today: bool = False) -> dict[str, Any]:
     hours = _hours_window()
     per_feed = _per_feed_limit()
     per_category = _per_category_limit()
-    cutoff = _now_utc() - timedelta(hours=hours)
+    cutoff = _cutoff_datetime(floor_today=floor_today)
 
     headers = {"User-Agent": USER_AGENT, "Accept": "application/rss+xml, application/atom+xml, application/xml, text/xml, */*"}
     tasks: list[tuple[NewsCategory, NewsFeed]] = []
@@ -406,6 +627,7 @@ def _fetch_news(categories: tuple[NewsCategory, ...]) -> dict[str, Any]:
     return {
         "enabled": True,
         "hours": hours,
+        "floor_today": floor_today,
         "categories": categories_out,
         "flat": flat,
         "stats": stats,
@@ -416,7 +638,7 @@ def fetch_ai_news() -> dict[str, Any]:
     """按类别抓取国际 AI 时讯，返回 {enabled, hours, categories, flat, stats}。"""
     if not _enabled():
         return {"enabled": False, "categories": [], "flat": [], "stats": {}}
-    return _fetch_news(NEWS_CATEGORIES)
+    return _fetch_news(NEWS_CATEGORIES, floor_today=_floor_today_enabled())
 
 
 def fetch_cn_ai_news() -> dict[str, Any]:
@@ -459,12 +681,13 @@ def _format_news_section(
 
     categories = news.get("categories") or []
     hours = news.get("hours", 72)
+    floor_note = " · 仅今日" if news.get("floor_today") else ""
     lines = [
         "---",
         "",
         f"## {section_no}、{title}",
         "",
-        f"> 近 **{hours}h** · {news.get('stats', {}).get('feeds_ok', 0)}/{news.get('stats', {}).get('feeds_total', 0)} 源可用",
+        f"> 近 **{hours}h**{floor_note} · {news.get('stats', {}).get('feeds_ok', 0)}/{news.get('stats', {}).get('feeds_total', 0)} 源可用",
         "",
     ]
 
@@ -501,88 +724,23 @@ def prepare_wecom_news_items(news: dict[str, Any], *, date_str: str | None = Non
     if not news.get("enabled"):
         return []
     limit = _wecom_limit()
-    flat = _dedupe_items(news.get("flat") or [])
-    flat.sort(key=_sort_key, reverse=True)
+    floor = bool(news.get("floor_today", _floor_today_enabled()))
+    flat_strict = _filter_flat_in_window(news.get("flat") or [], floor_today=floor)
+    flat_relaxed = _filter_flat_in_window(news.get("flat") or [], floor_today=False)
     preferred = ("media", "newsletter", "official", "community", "research", "developer")
-    picked: list[dict[str, Any]] = []
-    seen: set[str] = set()
-    for cat in preferred:
-        for item in flat:
-            link = _normalize_link(item.get("link", ""))
-            if item.get("category_id") != cat or link in seen:
-                continue
-            picked.append(item)
-            seen.add(link)
-            if len(picked) >= limit:
-                break
-        if len(picked) >= limit:
-            break
-    items: list[dict[str, Any]] = []
-    for item in picked[:limit]:
-        items.append(
-            {
-                "title": item.get("title", "?"),
-                "link": item.get("link", ""),
-                "source_name": item.get("source_name", "?"),
-                "published_fmt": item.get("published_fmt", ""),
-                "desc_short": _clean_text(item.get("summary", ""), 36),
-            }
-        )
-    if date_str:
-        from daily.news.pushed_links import filter_unpushed_items
-
-        items = filter_unpushed_items(items, date_str=date_str)
-    return items
+    picked = _pick_news_items(flat_strict, limit, preferred)
+    picked = _fill_picked_to_limit(picked, [flat_relaxed, news.get("flat") or []], limit)
+    items = [_to_wecom_news_row(item) for item in picked[:limit]]
+    return _apply_pushed_dedup_with_backfill(items, picked, date_str=date_str, limit=limit)
 
 
 def prepare_wecom_cn_news_items(news: dict[str, Any], *, date_str: str | None = None) -> list[dict[str, Any]]:
     if not news.get("enabled"):
         return []
     limit = _wecom_cn_limit()
-    flat = _dedupe_items(news.get("flat") or [])
-    flat.sort(key=_sort_key, reverse=True)
-    preferred = ("media", "tech", "dev")
-    picked: list[dict[str, Any]] = []
-    seen_links: set[str] = set()
-    seen_sources: set[str] = set()
-
-    for cat in preferred:
-        for item in flat:
-            link = _normalize_link(item.get("link", ""))
-            source = item.get("source_name", "?")
-            if item.get("category_id") != cat or not link or link in seen_links or source in seen_sources:
-                continue
-            picked.append(item)
-            seen_links.add(link)
-            seen_sources.add(source)
-            if len(picked) >= limit:
-                break
-        if len(picked) >= limit:
-            break
-
-    if len(picked) < limit:
-        for item in flat:
-            link = _normalize_link(item.get("link", ""))
-            if not link or link in seen_links:
-                continue
-            picked.append(item)
-            seen_links.add(link)
-            if len(picked) >= limit:
-                break
-
-    items: list[dict[str, Any]] = []
-    for item in picked[:limit]:
-        items.append(
-            {
-                "title": item.get("title", "?"),
-                "link": item.get("link", ""),
-                "source_name": item.get("source_name", "?"),
-                "published_fmt": item.get("published_fmt", ""),
-                "desc_short": _clean_text(item.get("summary", ""), 36),
-            }
-        )
-    if date_str:
-        from daily.news.pushed_links import filter_unpushed_items
-
-        items = filter_unpushed_items(items, date_str=date_str)
-    return items
+    flat = _filter_flat_in_window(news.get("flat") or [], floor_today=False)
+    preferred = ("media", "tech")
+    picked = _pick_news_items(flat, limit, preferred, one_per_source=True)
+    picked = _fill_picked_to_limit(picked, [news.get("flat") or []], limit)
+    items = [_to_wecom_news_row(item) for item in picked[:limit]]
+    return _apply_pushed_dedup_with_backfill(items, picked, date_str=date_str, limit=limit)

daily/news/research.py

@@ -0,0 +1,324 @@
+"""Cursor SDK + deep-research 工作流：采集 AI 时讯（方案 A，内置 WebSearch）。"""
+
+from __future__ import annotations
+
+import json
+import logging
+from datetime import datetime, timezone, timedelta
+from pathlib import Path
+from typing import Any
+from urllib.parse import urlparse
+
+from daily.config import OUTPUT_DIR, ROOT, env, env_int, wecom_ai_news_tech_limit
+from daily.llm_client import cursor_agent_prompt, extract_json_object, has_cursor_configured
+from daily.news.fetch import brief_news_summary, _normalize_link
+from daily.news.pushed_links import filter_unpushed_items
+
+logger = logging.getLogger(__name__)
+
+_SKILL_DIR = ROOT / "skills" / "daily-ai-news-research"
+_DEEP_RESEARCH_CANDIDATES = (
+    ROOT / "skills" / "deep-research" / "SKILL.md",
+    Path.home() / ".agents" / "skills" / "deep-research" / "SKILL.md",
+    Path.home() / ".cursor" / "skills" / "deep-research" / "SKILL.md",
+)
+
+
+def ai_news_mode() -> str:
+    return (env("DAILY_AI_NEWS_MODE") or "rss").strip().lower()
+
+
+def is_research_mode() -> bool:
+    return ai_news_mode() == "research"
+
+
+def research_hours() -> int:
+    return max(1, env_int("DAILY_AI_NEWS_HOURS", 24))
+
+
+def research_limit() -> int:
+    return max(1, env_int("DAILY_WECOM_AI_NEWS", 10))
+
+
+def research_json_path(date_str: str) -> Path:
+    return OUTPUT_DIR / f"{date_str}.ai-news-research.json"
+
+
+def _save_research_json(path: Path, data: dict[str, Any]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+
+
+def _load_skill() -> str:
+    parts: list[str] = []
+    for path in _DEEP_RESEARCH_CANDIDATES:
+        if path.exists():
+            parts.append(path.read_text(encoding="utf-8").strip())
+            break
+    local = _SKILL_DIR / "SKILL.md"
+    if local.exists():
+        parts.append(local.read_text(encoding="utf-8").strip())
+    if not parts:
+        return "你是 AI 时讯调研员，只输出 JSON。"
+    return "\n\n---\n\n".join(parts)
+
+
+def _guess_source_name(link: str, explicit: str) -> str:
+    name = (explicit or "").strip()
+    if name:
+        return name
+    host = urlparse(link).netloc.lower().removeprefix("www.")
+    mapping = {
+        "techcrunch.com": "TechCrunch",
+        "theverge.com": "The Verge",
+        "openai.com": "OpenAI",
+        "anthropic.com": "Anthropic",
+        "arxiv.org": "arXiv",
+        "qbitai.com": "量子位",
+        "36kr.com": "36氪",
+        "leiphone.com": "雷锋网",
+    }
+    for key, label in mapping.items():
+        if host.endswith(key) or key in host:
+            return label
+    return host.split(".")[0].capitalize() if host else "?"
+
+
+def _normalize_research_item(raw: dict[str, Any]) -> dict[str, Any] | None:
+    title = str(raw.get("title") or "").strip()
+    link = _normalize_link(str(raw.get("link") or ""))
+    if not title or not link or not link.startswith("http"):
+        return None
+    desc = brief_news_summary(str(raw.get("desc_short") or raw.get("summary") or ""))
+    return {
+        "title": title,
+        "link": link,
+        "source_name": _guess_source_name(link, str(raw.get("source_name") or "")),
+        "published_fmt": str(raw.get("published_fmt") or "").strip(),
+        "desc_short": desc,
+        "summary_plain": desc,
+    }
+
+
+def research_tech_limit() -> int:
+    return wecom_ai_news_tech_limit()
+
+
+def _parse_items_array(
+    items_raw: Any,
+    *,
+    limit: int,
+    seen: set[str],
+) -> list[dict[str, Any]]:
+    if not isinstance(items_raw, list):
+        return []
+    out: list[dict[str, Any]] = []
+    for row in items_raw:
+        if not isinstance(row, dict):
+            continue
+        item = _normalize_research_item(row)
+        if not item:
+            continue
+        if item["link"] in seen:
+            continue
+        seen.add(item["link"])
+        out.append(item)
+        if len(out) >= limit:
+            break
+    return out
+
+
+def parse_research_response(
+    raw: str,
+    *,
+    limit: int,
+    tech_limit: int = 0,
+) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
+    parsed = extract_json_object(raw)
+    seen: set[str] = set()
+    items = _parse_items_array(parsed.get("items"), limit=limit, seen=seen)
+    tech_items = _parse_items_array(parsed.get("tech_items"), limit=tech_limit, seen=seen) if tech_limit else []
+    return items, tech_items
+
+
+def _apply_pushed_dedup(items: list[dict[str, Any]], *, date_str: str, limit: int) -> list[dict[str, Any]]:
+    from daily.config import news_backfill_enabled
+    from daily.news.sanitize import strip_relax_window_prefix
+
+    for item in items:
+        if item.get("desc_short"):
+            item["desc_short"] = strip_relax_window_prefix(str(item.get("desc_short") or ""))
+    fresh = filter_unpushed_items(items, date_str=date_str)
+    if len(fresh) >= limit:
+        return fresh[:limit]
+    if not news_backfill_enabled():
+        if len(fresh) < limit:
+            logger.info("news_short:%s", len(fresh))
+        return fresh[:limit]
+    seen = {i.get("link") for i in fresh}
+    for item in items:
+        if len(fresh) >= limit:
+            break
+        if item.get("link") not in seen:
+            fresh.append(item)
+            seen.add(item.get("link"))
+    return fresh[:limit]
+
+
+def fetch_ai_news_research(
+    *,
+    date_str: str,
+    hours: int | None = None,
+    limit: int | None = None,
+) -> dict[str, Any]:
+    """Cursor Agent 调研 AI 时讯；返回 {enabled, mode, hours, items, flat, stats}。"""
+    h = hours if hours is not None else research_hours()
+    lim = limit if limit is not None else research_limit()
+    tech_lim = research_tech_limit()
+
+    if not has_cursor_configured():
+        logger.warning("DAILY_AI_NEWS_MODE=research 但未配置 CURSOR_API_KEY")
+        return {
+            "enabled": False,
+            "mode": "research",
+            "items": [],
+            "tech_items": [],
+            "flat": [],
+            "stats": {"error": "no_cursor_key"},
+        }
+
+    skill = _load_skill()
+    now_cst = datetime.now(timezone(timedelta(hours=8)))
+    tech_clause = ""
+    if tech_lim:
+        tech_clause = (
+            f"\n另输出 **tech_items 恰好 {tech_lim} 条**，聚焦工程技术："
+            "模型/框架发布、开源项目、芯片算力、开发者工具、推理与工程实践。"
+            "与 items 不得重复 link。"
+        )
+    system = (
+        f"{skill}\n\n"
+        "当前执行 **早报 AI 时讯调研**。\n"
+        f"时间窗口：近 **{h}** 小时（截至 {now_cst.strftime('%Y-%m-%d %H:%M')} UTC+8）。\n"
+        f"输出 **恰好 {lim} 条** items，按重要性排序。{tech_clause}\n"
+        "使用 WebSearch 检索；不要读取本项目文档或 RSS 配置。"
+    )
+    user = (
+        f"/deep-research 获取近 {h} 小时的 AI 人工智能新闻资讯，"
+        "不区分国内国外，合并精选。"
+        f"只输出 JSON，items 长度={lim}"
+        + (f"，tech_items 长度={tech_lim}" if tech_lim else "")
+        + "。"
+    )
+
+    try:
+        raw = cursor_agent_prompt(system, user)
+    except Exception as exc:
+        logger.warning("AI 时讯 research 失败：%s", exc)
+        return {
+            "enabled": False,
+            "mode": "research",
+            "items": [],
+            "tech_items": [],
+            "flat": [],
+            "stats": {"error": str(exc)},
+        }
+
+    if not raw:
+        return {
+            "enabled": False,
+            "mode": "research",
+            "items": [],
+            "tech_items": [],
+            "flat": [],
+            "stats": {"error": "empty_response"},
+        }
+
+    items, tech_items = parse_research_response(raw, limit=lim, tech_limit=tech_lim)
+    payload = extract_json_object(raw)
+    if payload:
+        _save_research_json(research_json_path(date_str), payload)
+
+    if not items and not tech_items:
+        logger.warning("AI 时讯 research JSON 无效或无条目")
+        return {
+            "enabled": False,
+            "mode": "research",
+            "items": [],
+            "tech_items": [],
+            "flat": [],
+            "stats": {"error": "invalid_json"},
+        }
+
+    items = _apply_pushed_dedup(items, date_str=date_str, limit=lim)
+    if tech_items:
+        tech_items = _apply_pushed_dedup(tech_items, date_str=date_str, limit=tech_lim)
+    logger.info("AI 时讯 research 完成：%d 条 + %d 技术", len(items), len(tech_items))
+
+    flat = [
+        {
+            "title": i["title"],
+            "link": i["link"],
+            "summary": i.get("summary_plain") or i.get("desc_short") or "",
+            "source_name": i["source_name"],
+            "published_fmt": i.get("published_fmt") or "",
+            "category_id": "research",
+            "category_name": "Deep Research",
+            "category_icon": "🔍",
+        }
+        for i in items + tech_items
+    ]
+
+    return {
+        "enabled": True,
+        "mode": "research",
+        "hours": h,
+        "items": items,
+        "tech_items": tech_items,
+        "flat": flat,
+        "stats": {"source": "cursor_research", "items": len(items), "tech_items": len(tech_items)},
+    }
+
+
+def format_research_news_section(
+    research: dict[str, Any],
+    *,
+    section_no: int,
+    wecom_limit: int | None = None,
+) -> list[str]:
+    if not research.get("enabled"):
+        hint = research.get("stats", {}).get("error", "调研失败或未配置 CURSOR_API_KEY")
+        return [
+            "---",
+            "",
+            f"## {section_no}、AI 时讯精选（Deep Research）",
+            "",
+            f"*不可用：{hint}*",
+            "",
+        ]
+
+    hours = research.get("hours", 24)
+    items = (research.get("flat") or [])[: wecom_limit or research_limit()]
+    lines = [
+        "---",
+        "",
+        f"## {section_no}、AI 时讯精选（Deep Research）",
+        "",
+        f"> 近 **{hours}h** · Cursor Agent WebSearch · {len(items)} 条",
+        "",
+    ]
+    if not items:
+        lines.append("*暂无可用条目。*")
+        lines.append("")
+        return lines
+
+    for i, item in enumerate(items, 1):
+        pub = f" · {item['published_fmt']}" if item.get("published_fmt") else ""
+        lines.append(
+            f"{i}. **[{item['title']}]({item['link']})** · `{item['source_name']}`{pub}"
+        )
+        summary = item.get("summary") or ""
+        if summary:
+            lines.append(f"   - {summary}")
+    lines.append("")
+    return lines

daily/news/sanitize.py

@@ -0,0 +1,18 @@
+"""新闻文案清洗：剥离「放宽窗口」类凑数前缀。"""
+
+from __future__ import annotations
+
+import re
+
+_RELAX_PREFIX = re.compile(
+    r"^(?:放宽窗口|放宽至[^：:]*)\s*[:：]\s*",
+    re.UNICODE,
+)
+
+
+def strip_relax_window_prefix(text: str) -> str:
+    """去掉开头的「放宽窗口：」/「放宽至…：」前缀。"""
+    raw = (text or "").strip()
+    if not raw:
+        return ""
+    return _RELAX_PREFIX.sub("", raw, count=1).strip()

skills/daily-ai-news-research/SKILL.md

@@ -0,0 +1,58 @@
+# AI 时讯 Deep Research（早报专用）
+
+你是 **AI 时讯调研员**。使用 **WebSearch** 与网页抓取工具，收集近 N 小时全球 AI 新闻（不区分国内/国外），输出供企微早报使用的结构化 JSON。
+
+## 工作流
+
+1. 将任务拆成 3–5 个子问题（模型发布、监管政策、大厂动态、芯片算力、研究突破等）
+2. 每个子问题用 WebSearch 检索 2–3 组关键词（中英文混合）
+3. 交叉验证：优先权威媒体 / 官方博客 / 学术来源
+4. 精选最多 **10 条**最重要、可核实的新闻（`items`）；窗口内不足则少返回，勿凑数
+5. 另精选最多 **5 条**工程技术向新闻（`tech_items`）：模型/框架发布、开源、芯片算力、开发者工具、推理与工程实践；不得与 `items` 重复 link；不足则少返回
+6. **只输出 JSON**，不要 Markdown 报告，不要代码块
+
+## 质量规则
+
+1. 每条必须有可访问的 `link`（https://）
+2. 禁止编造未在搜索结果中出现的事实
+3. 优先近 N 小时内的新闻；若不足目标条数，**少返回**即可，禁止放宽至 48 小时凑数，禁止在 `desc_short` 标注「放宽窗口」
+4. `desc_short` 用中文一句话摘要（≤72 字）
+5. `title` 保留原文标题；中文源可用中文标题
+6. `source_name` 为媒体/站点简称（如 TechCrunch、量子位、OpenAI Blog）
+
+## 输出格式（严格 JSON）
+
+```json
+{
+  "items": [
+    {
+      "title": "Apple sues OpenAI over trade secret theft",
+      "link": "https://techcrunch.com/...",
+      "source_name": "TechCrunch",
+      "desc_short": "苹果起诉 OpenAI 涉嫌窃取硬件商业机密",
+      "published_fmt": "07-11 05:00"
+    }
+  ],
+  "tech_items": [
+    {
+      "title": "Meta Iris AI chip enters production",
+      "link": "https://example.com/...",
+      "source_name": "TechCrunch",
+      "desc_short": "Meta 自研 Iris 芯片 9 月量产",
+      "published_fmt": ""
+    }
+  ],
+  "methodology": "检索 6 组 query，分析 12 源，子问题：诉讼、模型安全、监管"
+}
+```
+
+- `items` 数组长度 **必须等于** 请求的 limit（默认 10）
+- `tech_items` 数组长度 **必须等于** 请求的 tech limit（默认 5）；聚焦工程技术，可与 `items` 主题重叠但 link 不得重复
+- `published_fmt` 格式 `MM-DD HH:MM`（UTC+8），无法确定则留空字符串
+- 不要输出 `items` 以外的长文；`methodology` 可选，一行即可
+
+## 禁止
+
+- 不要输出 ```json 代码块包裹（直接输出 JSON 对象）
+- 不要输出 Executive Summary / Key Takeaways 等报告章节
+- 不要使用本项目 RSS 或本地文档作为来源

tests/test_news_relax.py

@@ -0,0 +1,62 @@
+# tests/test_news_relax.py
+from __future__ import annotations
+
+import os
+import unittest
+from unittest.mock import patch
+
+
+class NewsRelaxTests(unittest.TestCase):
+    def test_strip_relax_prefix(self):
+        from daily.news.sanitize import strip_relax_window_prefix
+
+        self.assertEqual(
+            strip_relax_window_prefix("放宽窗口：苹果起诉 OpenAI"),
+            "苹果起诉 OpenAI",
+        )
+        self.assertEqual(
+            strip_relax_window_prefix("放宽至48小时：某新闻"),
+            "某新闻",
+        )
+        self.assertEqual(
+            strip_relax_window_prefix("正常摘要无前缀"),
+            "正常摘要无前缀",
+        )
+
+    def test_backfill_disabled_does_not_reinsert_pushed(self):
+        from daily.news.fetch import _apply_pushed_dedup_with_backfill
+
+        fresh_only = [
+            {
+                "link": "https://example.com/fresh",
+                "title": "fresh",
+                "source_name": "S",
+                "published_fmt": "07-14",
+                "desc_short": "新",
+                "summary_plain": "新",
+            }
+        ]
+        picked = [
+            {
+                "link": "https://example.com/old",
+                "title": "old",
+                "source_name": "S",
+                "published": "2026-07-13T10:00:00+00:00",
+                "summary": "旧闻",
+            }
+        ]
+        with patch("daily.news.pushed_links.filter_unpushed_items", return_value=list(fresh_only)):
+            with patch.dict(os.environ, {"DAILY_NEWS_BACKFILL": "0"}, clear=False):
+                out = _apply_pushed_dedup_with_backfill(
+                    fresh_only + [{"link": "https://example.com/old", "title": "old"}],
+                    picked,
+                    date_str="2026-07-14",
+                    limit=5,
+                )
+        links = [x.get("link") for x in out]
+        self.assertIn("https://example.com/fresh", links)
+        self.assertNotIn("https://example.com/old", links)
+
+
+if __name__ == "__main__":
+    unittest.main()