feat: LLM 解耦 bot、RSS 外置与内容过滤

daily 自建 Cursor bridge；DAILY_LLM_PROVIDER 控制后端；config/feeds.yaml 与 sensitive_words 可配置。

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

.env.example

@@ -62,6 +62,10 @@ DAILY_CN_AI_NEWS=1
 
 # Agent 工作流（推荐：叙事化早报）
 # DAILY_REPORT_MODE=agent
+# DAILY_LLM_PROVIDER=auto
+#   auto   — agent 模式优先 Cursor，否则 OpenAI；classic 优先 OpenAI
+#   openai — 强制 OpenAI 兼容 API
+#   cursor — 强制 Cursor SDK
 # CURSOR_API_KEY=cursor_...
 # CURSOR_MODEL=composer-2.5
 # DAILY_CURSOR_CWD=.
@@ -83,6 +87,10 @@ DAILY_AI_NEWS_PER_CATEGORY=5
 # DAILY_AGENT_NEWS_POOL=40
 # DAILY_AGENT_CN_NEWS_POOL=30
 
+# RSS 源：编辑 config/feeds.yaml（intl / cn）；缺失时回退内置默认
+# 内容过滤：config/sensitive_words.yaml + DAILY_CONTENT_FILTER=1
+# DAILY_CONTENT_FILTER=0
+
 # Reddit RSS（403/429 时在 Reddit 偏好设置 → RSS feeds 复制 user / feed 参数）
 # REDDIT_RSS_USER=your_username
 # REDDIT_RSS_FEED=your_feed_token

Dockerfile

@@ -16,7 +16,7 @@ RUN pip install --no-cache-dir -r requirements.txt
 COPY daily/ daily/
 COPY shared/ shared/
 COPY skills/ skills/
-COPY bot/ bot/
+COPY config/ config/
 COPY run-daily.sh .
 
 RUN chmod +x run-daily.sh \

README.md

@@ -116,10 +116,11 @@ Python 抓取 → Step1 趋势分析 (.trends.json) → Step2 写企微稿 (.wec
 
 ## 自定义 RSS 源
 
-国际 RSS 列表：`daily/news/feeds.py`  
+RSS 源配置在 **`config/feeds.yaml`**（国际 `intl` / 国内 `cn`），改 URL 或增删 feed 后无需改 Python。文件缺失或解析失败时，会回退 `daily/news/feeds_defaults.py` 内置列表。
-国内 RSS 列表：`daily/news/feeds_cn.py`
 
-在对应文件的 `FEEDS` 列表中增删 URL 即可。常用开关（见 `.env.example`）：
+可选 **`config/sensitive_words.yaml`** + `DAILY_CONTENT_FILTER=1`：标题/摘要命中敏感词则丢弃该条。
+
+旧版硬编码路径（仍可读作参考）：`daily/news/feeds_defaults.py`
 
 ```env
 DAILY_AI_NEWS=1

config/feeds.yaml

@@ -0,0 +1,166 @@
+version: 1
+intl:
+  categories:
+  - id: official
+    name: 厂商官方
+    icon: 🏢
+    feeds:
+    - name: Anthropic Claude 更新
+      url: https://docs.anthropic.com/en/release-notes/feed
+    - name: OpenAI
+      url: https://openai.com/news/rss.xml
+    - name: Google AI
+      url: https://blog.google/technology/ai/rss/
+    - name: DeepMind
+      url: https://deepmind.google/blog/rss.xml
+    - name: Meta Engineering
+      url: https://engineering.fb.com/feed/
+    - name: Microsoft Research
+      url: https://www.microsoft.com/en-us/research/feed/
+    - name: Microsoft Blog
+      url: https://blogs.microsoft.com/feed/
+    - name: Cohere
+      url: https://cohere.com/blog/rss.xml
+    - name: Cursor Changelog
+      url: https://cursor.com/changelog/rss.xml
+  - id: developer
+    name: Agent / LLM 开发者
+    icon: 🛠
+    feeds:
+    - name: LangChain
+      url: https://blog.langchain.dev/rss/
+    - name: Hugging Face
+      url: https://huggingface.co/blog/feed.xml
+    - name: Vercel Changelog
+      url: https://vercel.com/changelog/rss.xml
+    - name: GitHub Copilot
+      url: https://github.blog/changelog/label/copilot/feed/
+  - id: media
+    name: 综合科技媒体
+    icon: 📰
+    feeds:
+    - name: The Verge AI
+      url: https://www.theverge.com/rss/ai-artificial-intelligence/index.xml
+    - name: TechCrunch AI
+      url: https://techcrunch.com/category/artificial-intelligence/feed/
+    - name: Ars Technica AI
+      url: https://arstechnica.com/ai/feed/
+    - name: Wired AI
+      url: https://www.wired.com/feed/tag/ai/latest/rss
+    - name: MIT Tech Review
+      url: https://www.technologyreview.com/feed/
+    - name: VentureBeat AI
+      url: https://venturebeat.com/category/ai/feed/
+  - id: newsletter
+    name: Newsletter 日报
+    icon: ✉️
+    feeds:
+    - name: Ben's Bites
+      url: https://bensbites.substack.com/feed
+    - name: The Rundown AI
+      url: https://therundown.substack.com/feed
+    - name: Latent Space
+      url: https://www.latent.space/feed
+    - name: Simon Willison
+      url: https://simonwillison.net/atom/everything/
+    - name: Import AI
+      url: https://importai.substack.com/feed
+    - name: Last Week in AI
+      url: https://lastweekin.ai/feed
+    - name: The Neuron
+      url: https://www.theneuron.ai/feed
+  - id: research
+    name: 研究 / 论文
+    icon: 📚
+    feeds:
+    - name: arXiv cs.CL
+      url: https://arxiv.org/rss/cs.CL
+    - name: arXiv cs.AI
+      url: https://arxiv.org/rss/cs.AI
+    - name: arXiv cs.LG
+      url: https://arxiv.org/rss/cs.LG
+  - id: trending
+    name: 热点 / 趋势
+    icon: 🔥
+    feeds:
+    - name: Google News · AI
+      url: https://news.google.com/rss/search?q=artificial+intelligence+OR+LLM+OR+Claude+OR+GPT&hl=en-US&gl=US&ceid=US:en
+    - name: Google News · Technology
+      url: https://news.google.com/rss/headlines/section/topic/TECHNOLOGY?hl=en-US&gl=US&ceid=US:en
+    - name: Techmeme
+      url: https://www.techmeme.com/feed.xml
+    - name: HN · Front Page
+      url: https://hnrss.org/frontpage
+    - name: HN · 100+ Points
+      url: https://hnrss.org/newest?points=100
+    - name: Dev.to · AI
+      url: https://dev.to/feed/tag/ai
+    - name: Lobsters
+      url: https://lobste.rs/rss
+  - id: community
+    name: 社区讨论
+    icon: 💬
+    feeds:
+    - name: HN · AI/LLM/Agent
+      url: https://hnrss.org/newest?q=AI+OR+LLM+OR+Claude+OR+agent+OR+GPT+OR+Gemini
+    - name: Reddit · LLM/Claude/ML
+      url: https://old.reddit.com/r/LocalLLaMA+ClaudeAI+MachineLearning+OpenAI/.rss?limit=25
+      slow: true
+cn:
+  title_keywords:
+  - 人工智能
+  - 大模型
+  - 智能体
+  - 多模态
+  - AIGC
+  - LLM
+  - GPT
+  - Claude
+  - Gemini
+  - ChatGPT
+  - OpenAI
+  - Anthropic
+  - Copilot
+  - Agent
+  - 'AI '
+  - ' AI'
+  - AI·
+  - AI业务
+  - AI模型
+  - AI助手
+  - AI工具
+  - AI编程
+  - AI 编程
+  - AI版
+  - AI Agent
+  - 推理模型
+  - 深度学习
+  - 机器学习
+  - Function Calling
+  categories:
+  - id: media
+    name: AI 专业媒体
+    icon: 📰
+    feeds:
+    - name: 量子位
+      url: https://www.qbitai.com/feed
+    - name: InfoQ 中文
+      url: https://www.infoq.cn/feed/AI
+  - id: tech
+    name: 综合科技
+    icon: 📱
+    feeds:
+    - name: 36氪
+      url: https://36kr.com/feed
+      ai_filter: true
+    - name: 雷锋网
+      url: https://www.leiphone.com/feed
+    - name: Google News · AI
+      url: https://news.google.com/rss/search?q=人工智能+OR+大模型+OR+Agent+OR+LLM&hl=zh-CN&gl=CN&ceid=CN:zh-Hans
+  - id: dev
+    name: 开发者社区
+    icon: 💻
+    feeds:
+    - name: 掘金
+      url: https://juejin.cn/rss
+      ai_filter: true

config/sensitive_words.yaml

@@ -0,0 +1,10 @@
+# 标题或摘要命中任一词则丢弃（不区分大小写）
+words:
+  - 赌博
+  - 六合彩
+  - 网赌
+  - 色情
+  - 裸聊
+  - 代孕
+  - 办证
+  - 刷单兼职

daily/cursor_bridge.py

@@ -0,0 +1,142 @@
+"""Windows 兼容的 Cursor SDK bridge（daily 包自用，不依赖 bot/）。"""
+
+from __future__ import annotations
+
+import codecs
+import json
+import logging
+import os
+import subprocess
+import threading
+import time
+from pathlib import Path
+from typing import Any, Mapping
+
+from daily.config import ROOT, env
+
+logger = logging.getLogger(__name__)
+
+READY_LINE_PREFIX = "cursor-sdk-bridge ready "
+_bridge_lock = threading.Lock()
+_bridge_process: subprocess.Popen[bytes] | None = None
+
+
+def cursor_cwd() -> str:
+    return env("DAILY_CURSOR_CWD") or env("CURSOR_CWD") or str(ROOT)
+
+
+def _parse_discovery_line(line: str) -> Mapping[str, Any] | None:
+    if not line.startswith(READY_LINE_PREFIX):
+        return None
+    payload = line[len(READY_LINE_PREFIX) :].strip()
+    loaded = json.loads(payload)
+    if not isinstance(loaded, dict):
+        raise RuntimeError("Bridge discovery payload must be an object")
+    return loaded
+
+
+def _read_discovery_polling(process: subprocess.Popen[bytes], timeout: float = 60) -> Mapping[str, Any]:
+    if process.stderr is None:
+        raise RuntimeError("Bridge stderr unavailable")
+
+    fd = process.stderr.fileno()
+    was_blocking = os.get_blocking(fd)
+    os.set_blocking(fd, False)
+    decoder = codecs.getincrementaldecoder("utf-8")(errors="replace")
+    pending = ""
+    stderr_lines: list[str] = []
+    deadline = time.monotonic() + timeout
+
+    try:
+        while time.monotonic() < deadline:
+            try:
+                chunk = os.read(fd, 8192)
+            except BlockingIOError:
+                chunk = b""
+
+            if chunk:
+                pending += decoder.decode(chunk)
+                while "\n" in pending:
+                    line, pending = pending.split("\n", 1)
+                    stderr_lines.append(line)
+                    discovery = _parse_discovery_line(line)
+                    if discovery is not None:
+                        return discovery
+            else:
+                code = process.poll()
+                if code is not None:
+                    pending += decoder.decode(b"", final=True)
+                    if pending.strip():
+                        stderr_lines.append(pending.strip())
+                    joined = "\n".join(stderr_lines)[-2000:]
+                    raise RuntimeError(
+                        f"Bridge 启动失败 exit={code}: {joined or '无 stderr 输出'}"
+                    )
+                time.sleep(0.05)
+    finally:
+        os.set_blocking(fd, was_blocking)
+
+    raise RuntimeError("等待 Cursor bridge 就绪超时")
+
+
+def _auth_token_from_discovery(discovery: Mapping[str, Any]) -> str:
+    token = str(discovery.get("authToken") or "").strip()
+    if token:
+        return token
+    token_file = discovery.get("authTokenFile")
+    if token_file:
+        return Path(str(token_file)).read_text(encoding="utf-8").strip()
+    raise RuntimeError("Bridge discovery 缺少 auth token")
+
+
+def warm_cursor_bridge(force: bool = False) -> None:
+    """启动 cursor-sdk-bridge 并写入 CURSOR_SDK_BRIDGE_* 环境变量。"""
+    global _bridge_process
+
+    with _bridge_lock:
+        if (
+            not force
+            and _bridge_process is not None
+            and _bridge_process.poll() is None
+            and os.environ.get("CURSOR_SDK_BRIDGE_URL")
+            and os.environ.get("CURSOR_SDK_BRIDGE_TOKEN")
+        ):
+            return
+
+        if _bridge_process is not None and _bridge_process.poll() is None:
+            _bridge_process.terminate()
+            try:
+                _bridge_process.wait(timeout=5)
+            except subprocess.TimeoutExpired:
+                _bridge_process.kill()
+
+        from cursor_sdk._vendor import resolve_bridge_path
+
+        cwd = cursor_cwd()
+        os.environ["CURSOR_CWD"] = cwd
+        argv = [resolve_bridge_path(), "--workspace", cwd]
+        logger.info("启动 Cursor bridge workspace=%s", cwd)
+
+        process = subprocess.Popen(
+            argv,
+            stdout=subprocess.DEVNULL,
+            stderr=subprocess.PIPE,
+        )
+        try:
+            discovery = _read_discovery_polling(process)
+        except Exception:
+            process.kill()
+            process.wait(timeout=5)
+            raise
+
+        url = str(discovery.get("url") or "").strip()
+        if not url:
+            host = str(discovery.get("host") or "127.0.0.1")
+            port = discovery.get("port")
+            url = f"http://{host}:{port}"
+
+        token = _auth_token_from_discovery(discovery)
+        os.environ["CURSOR_SDK_BRIDGE_URL"] = url
+        os.environ["CURSOR_SDK_BRIDGE_TOKEN"] = token
+        _bridge_process = process
+        logger.info("Cursor bridge 就绪: %s", url)

daily/cursor_client.py

@@ -0,0 +1,33 @@
+"""Cursor SDK 调用（daily 包专用）。"""
+
+from __future__ import annotations
+
+from cursor_sdk import Agent, AgentOptions, CursorAgentError, LocalAgentOptions
+
+from daily.config import env
+from daily.cursor_bridge import cursor_cwd, warm_cursor_bridge
+
+
+def cursor_chat(system: str, user: str) -> str:
+    api_key = (env("CURSOR_API_KEY") or "").strip()
+    if not api_key:
+        return ""
+
+    warm_cursor_bridge()
+    cwd = cursor_cwd()
+    model = env("CURSOR_MODEL") or "composer-2.5"
+    prompt = f"{system}\n\n{user}"
+    try:
+        result = Agent.prompt(
+            prompt,
+            AgentOptions(
+                api_key=api_key,
+                model=model,
+                local=LocalAgentOptions(cwd=cwd),
+            ),
+        )
+    except CursorAgentError as exc:
+        raise RuntimeError(f"LLM 调用失败：{exc.message}") from exc
+    if result.status == "error":
+        raise RuntimeError(f"LLM 调用失败：{result.result or '未知错误'}")
+    return (result.result or "").strip()

daily/llm_client.py

@@ -3,17 +3,14 @@
 from __future__ import annotations
 
 import json
-import logging
-import os
 import re
 from typing import Any
 
 import certifi
 import httpx
 
-from daily.config import ROOT, env, env_int
+from daily.config import env, env_int
-
+from daily.cursor_client import cursor_chat
-logger = logging.getLogger(__name__)
 
 _JSON_BLOCK = re.compile(r"```(?:json)?\s*([\s\S]*?)```", re.IGNORECASE)
 
@@ -44,6 +41,47 @@ def extract_json_object(text: str) -> dict[str, Any]:
     return {}
 
 
+def _has_openai_configured() -> bool:
+    return bool(env("DAILY_LLM_API_KEY") or env("OPENAI_API_KEY"))
+
+
+def _has_cursor_configured() -> bool:
+    return bool(env("CURSOR_API_KEY"))
+
+
+def llm_provider() -> str:
+    raw = (env("DAILY_LLM_PROVIDER") or "auto").strip().lower()
+    if raw in {"openai", "cursor"}:
+        return raw
+    return "auto"
+
+
+def resolve_llm_backend() -> str:
+    """返回 openai | cursor | 空字符串。"""
+    provider = llm_provider()
+    has_openai = _has_openai_configured()
+    has_cursor = _has_cursor_configured()
+
+    if provider == "openai":
+        if has_openai:
+            return "openai"
+        return "cursor" if has_cursor else ""
+
+    if provider == "cursor":
+        if has_cursor:
+            return "cursor"
+        return "openai" if has_openai else ""
+
+    agent_mode = (env("DAILY_REPORT_MODE") or "").strip().lower() == "agent"
+    if agent_mode and has_cursor:
+        return "cursor"
+    if has_openai:
+        return "openai"
+    if has_cursor:
+        return "cursor"
+    return ""
+
+
 def _openai_chat(system: str, user: str) -> str:
     api_key = (env("DAILY_LLM_API_KEY") or env("OPENAI_API_KEY") or "").strip()
     if not api_key:
@@ -70,53 +108,14 @@ def _openai_chat(system: str, user: str) -> str:
     return str(data["choices"][0]["message"]["content"] or "").strip()
 
 
-def _cursor_chat(system: str, user: str) -> str:
-    api_key = (env("CURSOR_API_KEY") or "").strip()
-    if not api_key:
-        return ""
-    import sys
-
-    from daily.config import ROOT
-    from cursor_sdk import Agent, AgentOptions, CursorAgentError, LocalAgentOptions
-
-    _bot = str(ROOT / "bot")
-    if _bot not in sys.path:
-        sys.path.insert(0, _bot)
-    try:
-        from bridge_manager import warm_cursor_bridge
-    except ImportError:
-        warm_cursor_bridge = lambda: None  # noqa: E731
-
-    cwd = env("DAILY_CURSOR_CWD") or str(ROOT)
-    # bridge_manager 读 bot env_config 的 CURSOR_CWD，早报侧须先对齐工作目录
-    os.environ["CURSOR_CWD"] = cwd
-    warm_cursor_bridge()
-    model = env("CURSOR_MODEL") or "composer-2.5"
-    prompt = f"{system}\n\n{user}"
-    try:
-        result = Agent.prompt(
-            prompt,
-            AgentOptions(
-                api_key=api_key,
-                model=model,
-                local=LocalAgentOptions(cwd=cwd),
-            ),
-        )
-    except CursorAgentError as exc:
-        raise RuntimeError(f"LLM 调用失败：{exc.message}") from exc
-    if result.status == "error":
-        raise RuntimeError(f"LLM 调用失败：{result.result or '未知错误'}")
-    return (result.result or "").strip()
-
-
 def llm_chat(system: str, user: str) -> str:
-    """优先 OpenAI 兼容 API，否则 Cursor SDK。"""
+    backend = resolve_llm_backend()
-    if env("DAILY_LLM_API_KEY") or env("OPENAI_API_KEY"):
+    if backend == "openai":
         return _openai_chat(system, user)
-    if env("CURSOR_API_KEY"):
+    if backend == "cursor":
-        return _cursor_chat(system, user)
+        return cursor_chat(system, user)
     return ""
 
 
 def has_llm_configured() -> bool:
-    return bool(env("DAILY_LLM_API_KEY") or env("OPENAI_API_KEY") or env("CURSOR_API_KEY"))
+    return bool(resolve_llm_backend())

daily/news/content_filter.py

@@ -0,0 +1,77 @@
+"""敏感词内容过滤。"""
+
+from __future__ import annotations
+
+import logging
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+from daily.config import ROOT, env
+
+logger = logging.getLogger(__name__)
+
+_WORDS_FILE = ROOT / "config" / "sensitive_words.yaml"
+
+
+def content_filter_enabled() -> bool:
+    raw = (env("DAILY_CONTENT_FILTER") or "0").strip().lower()
+    return raw in {"1", "true", "yes", "on"}
+
+
+@lru_cache(maxsize=1)
+def load_sensitive_words() -> tuple[str, ...]:
+    if not _WORDS_FILE.exists():
+        return ()
+    try:
+        import yaml
+    except ImportError:
+        logger.warning("未安装 PyYAML，无法读取 %s", _WORDS_FILE)
+        return ()
+    try:
+        data = yaml.safe_load(_WORDS_FILE.read_text(encoding="utf-8"))
+    except OSError as exc:
+        logger.warning("读取敏感词配置失败: %s", exc)
+        return ()
+    except Exception as exc:
+        logger.warning("解析 sensitive_words.yaml 失败: %s", exc)
+        return ()
+    if not isinstance(data, dict):
+        return ()
+    words = data.get("words") or data.get("sensitive_words") or []
+    if not isinstance(words, list):
+        return ()
+    cleaned = tuple(str(word).strip() for word in words if str(word).strip())
+    return cleaned
+
+
+def matches_sensitive_text(text: str, words: tuple[str, ...]) -> str | None:
+    haystack = (text or "").lower()
+    if not haystack:
+        return None
+    for word in words:
+        needle = word.lower()
+        if needle and needle in haystack:
+            return word
+    return None
+
+
+def filter_news_items(items: list[dict[str, Any]]) -> tuple[list[dict[str, Any]], int]:
+    if not content_filter_enabled():
+        return items, 0
+    words = load_sensitive_words()
+    if not words:
+        return items, 0
+
+    kept: list[dict[str, Any]] = []
+    removed = 0
+    for item in items:
+        text = f"{item.get('title', '')} {item.get('summary', '')}"
+        hit = matches_sensitive_text(text, words)
+        if hit:
+            removed += 1
+            continue
+        kept.append(item)
+    if removed:
+        logger.info("内容过滤移除 %d 条（敏感词）", removed)
+    return kept, removed

daily/news/feeds.py

@@ -1,125 +1,10 @@
-"""国际 AI 时讯 RSS 源定义（按类别分组）。"""
+"""国际 AI 时讯 RSS 源（优先 config/feeds.yaml）。"""
 
 from __future__ import annotations
 
-from dataclasses import dataclass
+from daily.news.feeds_loader import load_categories
+from daily.news.feeds_types import NewsCategory, NewsFeed
 
+NEWS_CATEGORIES: tuple[NewsCategory, ...] = load_categories("intl")
 
-@dataclass(frozen=True)
+__all__ = ["NewsCategory", "NewsFeed", "NEWS_CATEGORIES"]
-class NewsFeed:
-    name: str
-    url: str
-    slow: bool = False  # 限速源（如 Reddit）串行抓取
-    ai_filter: bool = False  # 综合源仅保留标题命中 AI 关键词的条目
-
-
-@dataclass(frozen=True)
-class NewsCategory:
-    id: str
-    name: str
-    icon: str
-    feeds: tuple[NewsFeed, ...]
-
-
-NEWS_CATEGORIES: tuple[NewsCategory, ...] = (
-    NewsCategory(
-        id="official",
-        name="厂商官方",
-        icon="🏢",
-        feeds=(
-            NewsFeed("Anthropic Claude 更新", "https://docs.anthropic.com/en/release-notes/feed"),
-            NewsFeed("OpenAI", "https://openai.com/news/rss.xml"),
-            NewsFeed("Google AI", "https://blog.google/technology/ai/rss/"),
-            NewsFeed("DeepMind", "https://deepmind.google/blog/rss.xml"),
-            NewsFeed("Meta Engineering", "https://engineering.fb.com/feed/"),
-            NewsFeed("Microsoft Research", "https://www.microsoft.com/en-us/research/feed/"),
-            NewsFeed("Microsoft Blog", "https://blogs.microsoft.com/feed/"),
-            NewsFeed("Cohere", "https://cohere.com/blog/rss.xml"),
-            NewsFeed("Cursor Changelog", "https://cursor.com/changelog/rss.xml"),
-        ),
-    ),
-    NewsCategory(
-        id="developer",
-        name="Agent / LLM 开发者",
-        icon="🛠",
-        feeds=(
-            NewsFeed("LangChain", "https://blog.langchain.dev/rss/"),
-            NewsFeed("Hugging Face", "https://huggingface.co/blog/feed.xml"),
-            NewsFeed("Vercel Changelog", "https://vercel.com/changelog/rss.xml"),
-            NewsFeed("GitHub Copilot", "https://github.blog/changelog/label/copilot/feed/"),
-        ),
-    ),
-    NewsCategory(
-        id="media",
-        name="综合科技媒体",
-        icon="📰",
-        feeds=(
-            NewsFeed("The Verge AI", "https://www.theverge.com/rss/ai-artificial-intelligence/index.xml"),
-            NewsFeed("TechCrunch AI", "https://techcrunch.com/category/artificial-intelligence/feed/"),
-            NewsFeed("Ars Technica AI", "https://arstechnica.com/ai/feed/"),
-            NewsFeed("Wired AI", "https://www.wired.com/feed/tag/ai/latest/rss"),
-            NewsFeed("MIT Tech Review", "https://www.technologyreview.com/feed/"),
-            NewsFeed("VentureBeat AI", "https://venturebeat.com/category/ai/feed/"),
-        ),
-    ),
-    NewsCategory(
-        id="newsletter",
-        name="Newsletter 日报",
-        icon="✉️",
-        feeds=(
-            NewsFeed("Ben's Bites", "https://bensbites.substack.com/feed"),
-            NewsFeed("The Rundown AI", "https://therundown.substack.com/feed"),
-            NewsFeed("Latent Space", "https://www.latent.space/feed"),
-            NewsFeed("Simon Willison", "https://simonwillison.net/atom/everything/"),
-            NewsFeed("Import AI", "https://importai.substack.com/feed"),
-            NewsFeed("Last Week in AI", "https://lastweekin.ai/feed"),
-            NewsFeed("The Neuron", "https://www.theneuron.ai/feed"),
-        ),
-    ),
-    NewsCategory(
-        id="research",
-        name="研究 / 论文",
-        icon="📚",
-        feeds=(
-            NewsFeed("arXiv cs.CL", "https://arxiv.org/rss/cs.CL"),
-            NewsFeed("arXiv cs.AI", "https://arxiv.org/rss/cs.AI"),
-            NewsFeed("arXiv cs.LG", "https://arxiv.org/rss/cs.LG"),
-        ),
-    ),
-    NewsCategory(
-        id="trending",
-        name="热点 / 趋势",
-        icon="🔥",
-        feeds=(
-            NewsFeed(
-                "Google News · AI",
-                "https://news.google.com/rss/search?q=artificial+intelligence+OR+LLM+OR+Claude+OR+GPT&hl=en-US&gl=US&ceid=US:en",
-            ),
-            NewsFeed(
-                "Google News · Technology",
-                "https://news.google.com/rss/headlines/section/topic/TECHNOLOGY?hl=en-US&gl=US&ceid=US:en",
-            ),
-            NewsFeed("Techmeme", "https://www.techmeme.com/feed.xml"),
-            NewsFeed("HN · Front Page", "https://hnrss.org/frontpage"),
-            NewsFeed("HN · 100+ Points", "https://hnrss.org/newest?points=100"),
-            NewsFeed("Dev.to · AI", "https://dev.to/feed/tag/ai"),
-            NewsFeed("Lobsters", "https://lobste.rs/rss"),
-        ),
-    ),
-    NewsCategory(
-        id="community",
-        name="社区讨论",
-        icon="💬",
-        feeds=(
-            NewsFeed(
-                "HN · AI/LLM/Agent",
-                "https://hnrss.org/newest?q=AI+OR+LLM+OR+Claude+OR+agent+OR+GPT+OR+Gemini",
-            ),
-            NewsFeed(
-                "Reddit · LLM/Claude/ML",
-                "https://old.reddit.com/r/LocalLLaMA+ClaudeAI+MachineLearning+OpenAI/.rss?limit=25",
-                slow=True,
-            ),
-        ),
-    ),
-)

daily/news/feeds_cn.py

@@ -1,71 +1,11 @@
-"""国内 AI 时讯 RSS 源定义（按类别分组）。"""
+"""国内 AI 时讯 RSS 源（优先 config/feeds.yaml）。"""
 
 from __future__ import annotations
 
-from daily.news.feeds import NewsCategory, NewsFeed
+from daily.news.feeds_loader import load_categories, load_cn_title_keywords
+from daily.news.feeds_types import NewsCategory, NewsFeed
 
-# 综合源 ai_filter=True 时，仅保留标题命中以下词之一的条目
+CN_AI_TITLE_KEYWORDS: tuple[str, ...] = load_cn_title_keywords()
-CN_AI_TITLE_KEYWORDS: tuple[str, ...] = (
+CN_NEWS_CATEGORIES: tuple[NewsCategory, ...] = load_categories("cn")
-    "人工智能",
-    "大模型",
-    "智能体",
-    "多模态",
-    "AIGC",
-    "LLM",
-    "GPT",
-    "Claude",
-    "Gemini",
-    "ChatGPT",
-    "OpenAI",
-    "Anthropic",
-    "Copilot",
-    "Agent",
-    "AI ",
-    " AI",
-    "AI·",
-    "AI业务",
-    "AI模型",
-    "AI助手",
-    "AI工具",
-    "AI编程",
-    "AI 编程",
-    "AI版",
-    "AI Agent",
-    "推理模型",
-    "深度学习",
-    "机器学习",
-    "Function Calling",
-)
 
-CN_NEWS_CATEGORIES: tuple[NewsCategory, ...] = (
+__all__ = ["CN_AI_TITLE_KEYWORDS", "CN_NEWS_CATEGORIES", "NewsCategory", "NewsFeed"]
-    NewsCategory(
-        id="media",
-        name="AI 专业媒体",
-        icon="📰",
-        feeds=(
-            NewsFeed("量子位", "https://www.qbitai.com/feed"),
-            NewsFeed("InfoQ 中文", "https://www.infoq.cn/feed/AI"),
-        ),
-    ),
-    NewsCategory(
-        id="tech",
-        name="综合科技",
-        icon="📱",
-        feeds=(
-            NewsFeed("36氪", "https://36kr.com/feed", ai_filter=True),
-            NewsFeed("雷锋网", "https://www.leiphone.com/feed"),
-            NewsFeed(
-                "Google News · AI",
-                "https://news.google.com/rss/search?q=人工智能+OR+大模型+OR+Agent+OR+LLM&hl=zh-CN&gl=CN&ceid=CN:zh-Hans",
-            ),
-        ),
-    ),
-    NewsCategory(
-        id="dev",
-        name="开发者社区",
-        icon="💻",
-        feeds=(
-            NewsFeed("掘金", "https://juejin.cn/rss", ai_filter=True),
-        ),
-    ),
-)

daily/news/feeds_defaults.py

@@ -0,0 +1,11 @@
+"""Built-in RSS defaults when config/feeds.yaml is missing or invalid."""
+
+from __future__ import annotations
+
+from daily.news.feeds_types import NewsCategory, NewsFeed
+
+CN_AI_TITLE_KEYWORDS: tuple[str, ...] = ('人工智能', '大模型', '智能体', '多模态', 'AIGC', 'LLM', 'GPT', 'Claude', 'Gemini', 'ChatGPT', 'OpenAI', 'Anthropic', 'Copilot', 'Agent', 'AI ', ' AI', 'AI·', 'AI业务', 'AI模型', 'AI助手', 'AI工具', 'AI编程', 'AI 编程', 'AI版', 'AI Agent', '推理模型', '深度学习', '机器学习', 'Function Calling')
+
+CN_NEWS_CATEGORIES: tuple[NewsCategory, ...] = (NewsCategory(id='media', name='AI 专业媒体', icon='📰', feeds=(NewsFeed(name='量子位', url='https://www.qbitai.com/feed', slow=False, ai_filter=False), NewsFeed(name='InfoQ 中文', url='https://www.infoq.cn/feed/AI', slow=False, ai_filter=False))), NewsCategory(id='tech', name='综合科技', icon='📱', feeds=(NewsFeed(name='36氪', url='https://36kr.com/feed', slow=False, ai_filter=True), NewsFeed(name='雷锋网', url='https://www.leiphone.com/feed', slow=False, ai_filter=False), NewsFeed(name='Google News · AI', url='https://news.google.com/rss/search?q=人工智能+OR+大模型+OR+Agent+OR+LLM&hl=zh-CN&gl=CN&ceid=CN:zh-Hans', slow=False, ai_filter=False))), NewsCategory(id='dev', name='开发者社区', icon='💻', feeds=(NewsFeed(name='掘金', url='https://juejin.cn/rss', slow=False, ai_filter=True),)))
+
+NEWS_CATEGORIES: tuple[NewsCategory, ...] = (NewsCategory(id='official', name='厂商官方', icon='🏢', feeds=(NewsFeed(name='Anthropic Claude 更新', url='https://docs.anthropic.com/en/release-notes/feed', slow=False, ai_filter=False), NewsFeed(name='OpenAI', url='https://openai.com/news/rss.xml', slow=False, ai_filter=False), NewsFeed(name='Google AI', url='https://blog.google/technology/ai/rss/', slow=False, ai_filter=False), NewsFeed(name='DeepMind', url='https://deepmind.google/blog/rss.xml', slow=False, ai_filter=False), NewsFeed(name='Meta Engineering', url='https://engineering.fb.com/feed/', slow=False, ai_filter=False), NewsFeed(name='Microsoft Research', url='https://www.microsoft.com/en-us/research/feed/', slow=False, ai_filter=False), NewsFeed(name='Microsoft Blog', url='https://blogs.microsoft.com/feed/', slow=False, ai_filter=False), NewsFeed(name='Cohere', url='https://cohere.com/blog/rss.xml', slow=False, ai_filter=False), NewsFeed(name='Cursor Changelog', url='https://cursor.com/changelog/rss.xml', slow=False, ai_filter=False))), NewsCategory(id='developer', name='Agent / LLM 开发者', icon='🛠', feeds=(NewsFeed(name='LangChain', url='https://blog.langchain.dev/rss/', slow=False, ai_filter=False), NewsFeed(name='Hugging Face', url='https://huggingface.co/blog/feed.xml', slow=False, ai_filter=False), NewsFeed(name='Vercel Changelog', url='https://vercel.com/changelog/rss.xml', slow=False, ai_filter=False), NewsFeed(name='GitHub Copilot', url='https://github.blog/changelog/label/copilot/feed/', slow=False, ai_filter=False))), NewsCategory(id='media', name='综合科技媒体', icon='📰', feeds=(NewsFeed(name='The Verge AI', url='https://www.theverge.com/rss/ai-artificial-intelligence/index.xml', slow=False, ai_filter=False), NewsFeed(name='TechCrunch AI', url='https://techcrunch.com/category/artificial-intelligence/feed/', slow=False, ai_filter=False), NewsFeed(name='Ars Technica AI', url='https://arstechnica.com/ai/feed/', slow=False, ai_filter=False), NewsFeed(name='Wired AI', url='https://www.wired.com/feed/tag/ai/latest/rss', slow=False, ai_filter=False), NewsFeed(name='MIT Tech Review', url='https://www.technologyreview.com/feed/', slow=False, ai_filter=False), NewsFeed(name='VentureBeat AI', url='https://venturebeat.com/category/ai/feed/', slow=False, ai_filter=False))), NewsCategory(id='newsletter', name='Newsletter 日报', icon='✉️', feeds=(NewsFeed(name="Ben's Bites", url='https://bensbites.substack.com/feed', slow=False, ai_filter=False), NewsFeed(name='The Rundown AI', url='https://therundown.substack.com/feed', slow=False, ai_filter=False), NewsFeed(name='Latent Space', url='https://www.latent.space/feed', slow=False, ai_filter=False), NewsFeed(name='Simon Willison', url='https://simonwillison.net/atom/everything/', slow=False, ai_filter=False), NewsFeed(name='Import AI', url='https://importai.substack.com/feed', slow=False, ai_filter=False), NewsFeed(name='Last Week in AI', url='https://lastweekin.ai/feed', slow=False, ai_filter=False), NewsFeed(name='The Neuron', url='https://www.theneuron.ai/feed', slow=False, ai_filter=False))), NewsCategory(id='research', name='研究 / 论文', icon='📚', feeds=(NewsFeed(name='arXiv cs.CL', url='https://arxiv.org/rss/cs.CL', slow=False, ai_filter=False), NewsFeed(name='arXiv cs.AI', url='https://arxiv.org/rss/cs.AI', slow=False, ai_filter=False), NewsFeed(name='arXiv cs.LG', url='https://arxiv.org/rss/cs.LG', slow=False, ai_filter=False))), NewsCategory(id='trending', name='热点 / 趋势', icon='🔥', feeds=(NewsFeed(name='Google News · AI', url='https://news.google.com/rss/search?q=artificial+intelligence+OR+LLM+OR+Claude+OR+GPT&hl=en-US&gl=US&ceid=US:en', slow=False, ai_filter=False), NewsFeed(name='Google News · Technology', url='https://news.google.com/rss/headlines/section/topic/TECHNOLOGY?hl=en-US&gl=US&ceid=US:en', slow=False, ai_filter=False), NewsFeed(name='Techmeme', url='https://www.techmeme.com/feed.xml', slow=False, ai_filter=False), NewsFeed(name='HN · Front Page', url='https://hnrss.org/frontpage', slow=False, ai_filter=False), NewsFeed(name='HN · 100+ Points', url='https://hnrss.org/newest?points=100', slow=False, ai_filter=False), NewsFeed(name='Dev.to · AI', url='https://dev.to/feed/tag/ai', slow=False, ai_filter=False), NewsFeed(name='Lobsters', url='https://lobste.rs/rss', slow=False, ai_filter=False))), NewsCategory(id='community', name='社区讨论', icon='💬', feeds=(NewsFeed(name='HN · AI/LLM/Agent', url='https://hnrss.org/newest?q=AI+OR+LLM+OR+Claude+OR+agent+OR+GPT+OR+Gemini', slow=False, ai_filter=False), NewsFeed(name='Reddit · LLM/Claude/ML', url='https://old.reddit.com/r/LocalLLaMA+ClaudeAI+MachineLearning+OpenAI/.rss?limit=25', slow=True, ai_filter=False))))

daily/news/feeds_loader.py

@@ -0,0 +1,113 @@
+"""从 config/feeds.yaml 加载 RSS 源（失败时回退内置默认）。"""
+
+from __future__ import annotations
+
+import logging
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+from daily.config import ROOT
+from daily.news.feeds_types import NewsCategory, NewsFeed
+
+logger = logging.getLogger(__name__)
+
+_FEEDS_FILE = ROOT / "config" / "feeds.yaml"
+
+
+def _parse_feed(raw: dict[str, Any]) -> NewsFeed:
+    return NewsFeed(
+        name=str(raw.get("name") or "").strip(),
+        url=str(raw.get("url") or "").strip(),
+        slow=bool(raw.get("slow")),
+        ai_filter=bool(raw.get("ai_filter")),
+    )
+
+
+def _parse_category(raw: dict[str, Any]) -> NewsCategory | None:
+    cat_id = str(raw.get("id") or "").strip()
+    if not cat_id:
+        return None
+    feeds_raw = raw.get("feeds") or []
+    feeds: list[NewsFeed] = []
+    for item in feeds_raw:
+        if not isinstance(item, dict):
+            continue
+        feed = _parse_feed(item)
+        if feed.name and feed.url:
+            feeds.append(feed)
+    if not feeds:
+        return None
+    return NewsCategory(
+        id=cat_id,
+        name=str(raw.get("name") or cat_id),
+        icon=str(raw.get("icon") or "📰"),
+        feeds=tuple(feeds),
+    )
+
+
+def _parse_categories(items: Any) -> tuple[NewsCategory, ...]:
+    if not isinstance(items, list):
+        return ()
+    categories: list[NewsCategory] = []
+    for raw in items:
+        if not isinstance(raw, dict):
+            continue
+        cat = _parse_category(raw)
+        if cat:
+            categories.append(cat)
+    return tuple(categories)
+
+
+def _load_yaml() -> dict[str, Any] | None:
+    if not _FEEDS_FILE.exists():
+        return None
+    try:
+        import yaml
+    except ImportError:
+        logger.warning("未安装 PyYAML，无法读取 %s", _FEEDS_FILE)
+        return None
+    try:
+        data = yaml.safe_load(_FEEDS_FILE.read_text(encoding="utf-8"))
+    except OSError as exc:
+        logger.warning("读取 feeds 配置失败: %s", exc)
+        return None
+    except Exception as exc:
+        logger.warning("解析 feeds.yaml 失败: %s", exc)
+        return None
+    return data if isinstance(data, dict) else None
+
+
+@lru_cache(maxsize=1)
+def _yaml_payload() -> dict[str, Any] | None:
+    return _load_yaml()
+
+
+def load_categories(region: str) -> tuple[NewsCategory, ...]:
+    data = _yaml_payload()
+    if data:
+        block = data.get(region) or {}
+        categories = _parse_categories(block.get("categories"))
+        if categories:
+            return categories
+        logger.warning("feeds.yaml 中 %s.categories 为空，使用内置默认", region)
+
+    from daily.news import feeds_defaults as defaults
+
+    if region == "cn":
+        return defaults.CN_NEWS_CATEGORIES
+    return defaults.NEWS_CATEGORIES
+
+
+def load_cn_title_keywords() -> tuple[str, ...]:
+    data = _yaml_payload()
+    if data:
+        block = data.get("cn") or {}
+        keywords = block.get("title_keywords")
+        if isinstance(keywords, list):
+            cleaned = tuple(str(x).strip() for x in keywords if str(x).strip())
+            if cleaned:
+                return cleaned
+    from daily.news.feeds_defaults import CN_AI_TITLE_KEYWORDS
+
+    return CN_AI_TITLE_KEYWORDS

daily/news/feeds_types.py

@@ -0,0 +1,21 @@
+"""RSS 源数据结构。"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class NewsFeed:
+    name: str
+    url: str
+    slow: bool = False
+    ai_filter: bool = False
+
+
+@dataclass(frozen=True)
+class NewsCategory:
+    id: str
+    name: str
+    icon: str
+    feeds: tuple[NewsFeed, ...]

daily/news/fetch.py

@@ -17,6 +17,7 @@ import certifi
 import httpx
 
 from daily.config import env, env_int, news_summary_limit
+from daily.news.content_filter import filter_news_items
 from daily.news.feeds import NEWS_CATEGORIES, NewsCategory, NewsFeed
 from daily.news.feeds_cn import CN_AI_TITLE_KEYWORDS, CN_NEWS_CATEGORIES
 
@@ -349,6 +350,7 @@ def _fetch_news(categories: tuple[NewsCategory, ...]) -> dict[str, Any]:
         "feeds_ok": 0,
         "items_raw": 0,
         "feeds_failed": [],
+        "content_filtered": 0,
     }
 
     with httpx.Client(timeout=15.0, verify=certifi.where(), follow_redirects=True, headers=headers) as client:
@@ -394,6 +396,9 @@ def _fetch_news(categories: tuple[NewsCategory, ...]) -> dict[str, Any]:
     for category in categories:
         items = raw_by_category[category.id]
         items = [i for i in items if _within_window(i, cutoff)]
+        items, filtered_count = filter_news_items(items)
+        if filtered_count:
+            stats["content_filtered"] = stats.get("content_filtered", 0) + filtered_count
         items.sort(key=_sort_key, reverse=True)
         items = _dedupe_items(items)[:per_category]
         for item in items:

requirements.txt

@@ -1,5 +1,6 @@
 python-dotenv>=1.0.0
 httpx>=0.27.0
 certifi>=2024.0.0
+PyYAML>=6.0.0
 # 英文描述转中文（使用 CURSOR_API_KEY 时需安装）
 cursor-sdk>=0.1.0

scripts/gen_feeds_defaults.py

@@ -0,0 +1,47 @@
+from pathlib import Path
+
+import yaml
+
+from daily.news.feeds_types import NewsCategory, NewsFeed
+
+data = yaml.safe_load(Path("config/feeds.yaml").read_text(encoding="utf-8"))
+
+
+def parse_feed(raw: dict) -> NewsFeed:
+    return NewsFeed(
+        name=raw["name"],
+        url=raw["url"],
+        slow=bool(raw.get("slow")),
+        ai_filter=bool(raw.get("ai_filter")),
+    )
+
+
+def parse_cat(raw: dict) -> NewsCategory:
+    return NewsCategory(
+        id=raw["id"],
+        name=raw["name"],
+        icon=raw["icon"],
+        feeds=tuple(parse_feed(item) for item in raw["feeds"]),
+    )
+
+
+intl = tuple(parse_cat(item) for item in data["intl"]["categories"])
+cn = tuple(parse_cat(item) for item in data["cn"]["categories"])
+keywords = tuple(data["cn"]["title_keywords"])
+
+lines = [
+    '"""Built-in RSS defaults when config/feeds.yaml is missing or invalid."""',
+    "",
+    "from __future__ import annotations",
+    "",
+    "from daily.news.feeds_types import NewsCategory, NewsFeed",
+    "",
+    f"CN_AI_TITLE_KEYWORDS: tuple[str, ...] = {keywords!r}",
+    "",
+    f"CN_NEWS_CATEGORIES: tuple[NewsCategory, ...] = {cn!r}",
+    "",
+    f"NEWS_CATEGORIES: tuple[NewsCategory, ...] = {intl!r}",
+    "",
+]
+Path("daily/news/feeds_defaults.py").write_text("\n".join(lines), encoding="utf-8")
+print("written")

tests/test_feeds_and_filter.py

@@ -0,0 +1,40 @@
+from __future__ import annotations
+
+import pytest
+
+
+def test_load_categories_from_yaml():
+    from daily.news.feeds_loader import load_categories
+
+    categories = load_categories("intl")
+    assert categories
+    assert any(cat.id == "official" for cat in categories)
+    assert categories[0].feeds
+
+
+def test_load_categories_fallback_when_yaml_missing(monkeypatch, tmp_path):
+    import daily.config as config
+    import daily.news.feeds_loader as loader
+
+    missing = tmp_path / "missing" / "feeds.yaml"
+    monkeypatch.setattr(loader, "_FEEDS_FILE", missing)
+    loader._yaml_payload.cache_clear()
+
+    categories = loader.load_categories("intl")
+    assert categories
+    assert any(cat.id == "official" for cat in categories)
+
+
+def test_content_filter_removes_matching_items(monkeypatch):
+    import daily.news.content_filter as cf
+
+    monkeypatch.setenv("DAILY_CONTENT_FILTER", "1")
+    cf.load_sensitive_words.cache_clear()
+    items = [
+        {"title": "正常 AI 新闻", "summary": "OpenAI 发布新模型"},
+        {"title": "违规推广", "summary": "六合彩内幕消息"},
+    ]
+    kept, removed = cf.filter_news_items(items)
+    assert removed == 1
+    assert len(kept) == 1
+    assert kept[0]["title"].startswith("正常")

tests/test_llm_client.py

@@ -0,0 +1,33 @@
+from __future__ import annotations
+
+import pytest
+
+
+def test_resolve_llm_backend_prefers_cursor_in_agent_mode(monkeypatch):
+    import daily.llm_client as llm
+
+    monkeypatch.setenv("DAILY_REPORT_MODE", "agent")
+    monkeypatch.setenv("CURSOR_API_KEY", "cursor_test")
+    monkeypatch.setenv("DAILY_LLM_API_KEY", "openai_test")
+    monkeypatch.delenv("DAILY_LLM_PROVIDER", raising=False)
+    assert llm.resolve_llm_backend() == "cursor"
+
+
+def test_resolve_llm_backend_openai_override(monkeypatch):
+    import daily.llm_client as llm
+
+    monkeypatch.setenv("DAILY_REPORT_MODE", "agent")
+    monkeypatch.setenv("DAILY_LLM_PROVIDER", "openai")
+    monkeypatch.setenv("CURSOR_API_KEY", "cursor_test")
+    monkeypatch.setenv("DAILY_LLM_API_KEY", "openai_test")
+    assert llm.resolve_llm_backend() == "openai"
+
+
+def test_resolve_llm_backend_classic_defaults_openai(monkeypatch):
+    import daily.llm_client as llm
+
+    monkeypatch.setenv("DAILY_REPORT_MODE", "classic")
+    monkeypatch.setenv("CURSOR_API_KEY", "cursor_test")
+    monkeypatch.setenv("DAILY_LLM_API_KEY", "openai_test")
+    monkeypatch.delenv("DAILY_LLM_PROVIDER", raising=False)
+    assert llm.resolve_llm_backend() == "openai"