feat: tag-aware daily digest - tags in AI prompt, grouping, frontmatter, cross-day continuity

analyzers/daily_digest.py

@@ -5,6 +5,7 @@ import logging
 import os
 import sys
 import subprocess
+from collections import Counter
 from datetime import datetime, timezone, timedelta
 
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
@@ -33,7 +34,6 @@ def utc_to_beijing(ts_str):
 def read_previous_digest(date):
     """Read the most recent previous daily digest (if exists) for continuity."""
     daily_root = get_output_dir("daily")
-    # Try yesterday first, then go back up to 7 days
     for days_back in range(1, 8):
         prev_date = date - timedelta(days=days_back)
         prev_dir = os.path.join(daily_root, prev_date.strftime("%Y-%m-%d"))
@@ -45,10 +45,8 @@ def read_previous_digest(date):
         prev_file = os.path.join(prev_dir, files[0])
         with open(prev_file, "r", encoding="utf-8") as f:
             content = f.read()
-            # Extract just the AI analysis section, skip frontmatter
             ai_section_start = content.find("## AI 分析")
             if ai_section_start != -1:
-                # Find the "我的批注" section boundary
                 annotation_start = content.find("## 我的批注", ai_section_start)
                 if annotation_start != -1:
                     ai_section = content[ai_section_start:annotation_start].strip()
@@ -60,7 +58,6 @@ def read_previous_digest(date):
                 )
                 return prev_date, ai_section
             else:
-                # No AI section found, return whole file
                 logger.info(
                     "Found previous digest from %s (no AI section found, using full)",
                     prev_date.strftime("%Y-%m-%d")
@@ -71,19 +68,28 @@ def read_previous_digest(date):
 
 
 # ============================================================
-# System prompt for daily analysis
+# System prompt for daily analysis (v2 - tag-aware)
 # ============================================================
 DAILY_SYSTEM_PROMPT = """你是一个私人思考伙伴。你的任务是认真阅读用户今天的每一条灵感记录，结合之前的分析，写一篇有深度的分析文章。
 
 注意：所有时间戳都是**北京时间**（UTC+8），不是 UTC。用户在中国，记录和活动时间均以北京时间为准。
 
+## 关于标签
+
+用户的每条记录可能带有标签（如 #培训 #交易 #生活 等）。标签是用户自己对记录的分类和上下文标注，对理解记录非常重要：
+
+1. **标签揭示场景**：同一条灵感，标注 #培训 和标注 #交易 的解读角度完全不同。标签告诉你用户在什么情境下产生的这个想法。
+2. **标签揭示延续性**：如果多天都有 #90后四级副职培训班 标签，说明这是一个持续性的活动/项目，分析时应体现跨日的连续思考。
+3. **同标签分组**：当多条记录共享同一标签时，它们通常是同一主题下的不同侧面，应该融会贯通地分析，而不是孤立看待。
+4. **不要忽略无标签记录**：没有标签的记录可能是随想、零散灵感，但也有其价值。
+
 核心原则：不做缩写，不做分类表，不写空话。
 
 具体要求：
 
 1. **引用原文**：分析每一条灵感时，必须先引用（或精炼复述）用户的原文，让用户一眼就知道"哦我在说这个"。引用要压缩但不失原意，不能断章取义。
 
-2. **融会贯通，而非逐条罗列**：不要"第一条...第二条...第三条..."机械堆砌。要把所有灵感当作一个整体来思考——哪几条在说同一个主题？哪几条看似无关实则互补？把它们串起来写。文章是流淌的整体，不是并列的零件。
+2. **融会贯通，而非逐条罗列**：不要"第一条...第二条...第三条..."机械堆砌。要把所有灵感当作一个整体来思考——哪几条在说同一个主题？哪几条看似无关实则互补？把它们串起来写。文章是流淌的整体，不是并列的零件。**注意利用标签来识别主题聚类**。
 
 3. **承接延续之前的分析**：你会看到上一次的分析内容。今天的分析不能只写"今天的事"，要把今天的灵感和昨天的分析结合起来——昨天讨论了什么？今天有什么进展？哪些问题有了答案？哪些问题还在延续？要让文章有跨日的时间纵深感。
 
@@ -110,13 +116,61 @@ DAILY_SYSTEM_PROMPT = """你是一个私人思考伙伴。你的任务是认真
 
 ---
 
-注意：即便只有1条灵感，也要写出深度。宁可写长，不可简略。"""
+注意：即便只有1条灵感，也要写出深度。篇幅不设上限，内容完整即可。"""
+
+
+def build_user_prompt(memos, date):
+    """Build user prompt for AI, with tag-aware formatting and grouping.
+
+    Memos are grouped by shared tags so the AI can see thematic clusters.
+    Each memo includes its tags in the text block.
+    """
+    # Step 1: Collect all tags across today's memos
+    tag_freq = MemosClient.aggregate_tags(memos)
+    groups = MemosClient.group_memos_by_tag(memos)
+
+    prompt_parts = []
+
+    # Header: date + tag overview
+    date_str = date.strftime("%Y-%m-%d")
+    prompt_parts.append(f"以下是我 {date_str} 的灵感记录（共 {len(memos)} 条）。")
+
+    if tag_freq:
+        tag_summary = "、".join(f"#{t}（{c}条）" for t, c in tag_freq.items())
+        prompt_parts.append(f"\n今天的标签分布：{tag_summary}")
+        prompt_parts.append("\n标签说明：用户通过标签自行分类记录。同标签的记录通常属于同一主题或场景，分析时应关注标签揭示的上下文。")
+
+    # Step 2: Group memos by tags for thematic presentation
+    prompt_parts.append("\n---\n")
+
+    # Priority: present tagged memos first (grouped by tag), then untagged
+    sorted_tags = sorted(groups.keys(), key=lambda t: (t == "无标签", -len(groups[t])))
+
+    for tag in sorted_tags:
+        group_memos = groups[tag]
+        if tag == "无标签":
+            prompt_parts.append(f"\n## 无标签记录（{len(group_memos)}条）")
+        else:
+            prompt_parts.append(f"\n## 标签 #{tag}（{len(group_memos)}条）")
+
+        for m in group_memos:
+            time_bj = utc_to_beijing(m["created_at"])
+            content = m["content"].strip()
+            memo_tags = m.get("tags", [])
+
+            # Show all tags on this memo (for cross-tagged memos)
+            tag_line = ""
+            if memo_tags:
+                tag_line = " [" + ", ".join("#" + t for t in memo_tags) + "]"
+
+            prompt_parts.append(f"\n- **[{time_bj}]**{tag_line}\n{content}")
+
+    return "\n".join(prompt_parts)
 
 
 def git_push(digest_file):
     """Commit and push the digest file to Gitea."""
     try:
-        # git add
         result = subprocess.run(
             ["git", "add", digest_file],
             cwd=PROJECT_DIR,
@@ -126,7 +180,6 @@ def git_push(digest_file):
             logger.warning("git add failed: %s", result.stderr.strip())
             return False
 
-        # git commit (check for changes first)
         result = subprocess.run(
             ["git", "status", "--porcelain", "ai-insights/"],
             cwd=PROJECT_DIR,
@@ -146,7 +199,6 @@ def git_push(digest_file):
             logger.warning("git commit failed: %s", result.stderr.strip())
             return False
 
-        # git push
         result = subprocess.run(
             ["git", "push", "origin", "main"],
             cwd=PROJECT_DIR,
@@ -156,7 +208,6 @@ def git_push(digest_file):
             logger.warning("git push failed: %s", result.stderr.strip())
             return False
 
-        # Extract remote result line
         for line in result.stdout.split("\n"):
             if "->" in line or "remote:" in line:
                 logger.info("Gitea push: %s", line.strip())
@@ -192,17 +243,11 @@ def run(memos_client, llm_client, date=None):
         with open(filepath, "w", encoding="utf-8") as f:
             f.write(content)
         logger.info("Empty digest written to %s", filepath)
-        # Still push
         git_push(filepath)
         return filepath, 0
 
-    # Step 2: Prepare user prompt
+    # Step 2: Build tag-aware user prompt
-    memo_texts = []
+    user_prompt = build_user_prompt(memos, date)
-    for m in memos:
-        time_bj = utc_to_beijing(m["created_at"])
-        memo_texts.append("- **[" + time_bj + "]** " + m["content"].strip())
-
-    user_prompt = "以下是我今天的灵感记录：\n\n" + "\n\n".join(memo_texts)
 
     # Step 3: Include previous analysis for continuity
     prev_date, prev_analysis = read_previous_digest(date)
@@ -215,17 +260,58 @@ def run(memos_client, llm_client, date=None):
             + prev_analysis
         )
 
+    # Step 4: Check for cross-day tag continuity
+    # If today has tags that appeared in previous digest, note this
+    tag_freq = MemosClient.aggregate_tags(memos)
+    today_tags = set(tag_freq.keys())
+    if prev_date and today_tags:
+        try:
+            prev_tags_line = None
+            # Extract tags from previous digest frontmatter
+            daily_root = get_output_dir("daily")
+            prev_dir = os.path.join(daily_root, prev_date.strftime("%Y-%m-%d"))
+            if os.path.isdir(prev_dir):
+                prev_files = sorted([f for f in os.listdir(prev_dir) if f.endswith(".md")], reverse=True)
+                if prev_files:
+                    with open(os.path.join(prev_dir, prev_files[0]), "r", encoding="utf-8") as f:
+                        prev_content = f.read(500)
+                    if "tags:" in prev_content:
+                        prev_tags_line = prev_content.split("tags:")[1].split("\n")[0]
+
+            if prev_tags_line:
+                import re
+                prev_tags = set(re.findall(r"'([^']+)'", prev_tags_line))
+                cross_tags = today_tags & prev_tags
+                if cross_tags:
+                    user_prompt += (
+                        "\n\n---\n\n"
+                        "跨标签延续提示：以下标签与上一次分析（"
+                        + prev_date.strftime("%Y-%m-%d")
+                        + "）共享："
+                        + "、".join("#" + t for t in sorted(cross_tags))
+                        + "。这些标签代表持续性的主题，请特别关注跨日的思维连续性。"
+                    )
+        except Exception as e:
+            logger.debug("Cross-tag check failed: %s", e)
+
     user_prompt += "\n\n---\n\n请基于以上所有素材，写一篇有深度的分析文章。"
 
-    # Step 4: Call DeepSeek API
+    # Step 5: Call DeepSeek API
     raw_response = llm_client.ask(
         system_prompt=DAILY_SYSTEM_PROMPT,
         user_prompt=user_prompt,
         temperature=0.5
     )
 
-    # Step 5: Wrap with frontmatter and write
+    # Step 6: Build frontmatter tags (default tags + today's user tags)
-    content = format_daily_digest(date, ai_body=raw_response, tags=["灵感收集器", "每日总结", "AI分析"], doc_type="daily-digest")
+    default_tags = ["灵感收集器", "每日总结", "AI分析"]
+    user_tags = sorted(tag_freq.keys())
+    all_tags = default_tags + user_tags
+
+    content = format_daily_digest(
+        date, ai_body=raw_response,
+        tags=all_tags, doc_type="daily-digest"
+    )
 
     date_str = date.strftime("%Y-%m-%d")
     daily_dir = os.path.join(get_output_dir("daily"), date_str)
@@ -238,11 +324,11 @@ def run(memos_client, llm_client, date=None):
         f.write(content)
 
     logger.info(
-        "Daily digest written to %s | %d memos | %d chars analysis",
+        "Daily digest written to %s | %d memos | %d chars analysis | tags: %s",
-        filepath, len(memos), len(raw_response)
+        filepath, len(memos), len(raw_response), ", ".join(user_tags) if user_tags else "(none)"
     )
 
-    # Step 6: Auto-push to Gitea
+    # Step 7: Auto-push to Gitea
     git_push(filepath)
 
     return filepath, len(memos)

tools/memos_client.py

@@ -1,12 +1,14 @@
 """Memos API client - fetch memos using Connect RPC protocol."""
 
 import logging
+from collections import Counter
 from datetime import datetime, timezone, timedelta
-
+from re import findall
-TZ_BEIJING = timezone(timedelta(hours=8))
 
 import requests
 
+TZ_BEIJING = timezone(timedelta(hours=8))
+
 logger = logging.getLogger(__name__)
 
 
@@ -45,13 +47,11 @@ class MemosClient:
     def list_memos(self, days=1, page_size=100):
         """Fetch memos from the last N days (Beijing time reference) via Connect RPC."""
         user = self.get_user_id()
-        # Use Beijing time as reference for "last N days"
         now_bj = datetime.now(TZ_BEIJING)
         since_bj = now_bj - timedelta(days=days)
         since_utc = since_bj.astimezone(timezone.utc)
         since = since_utc.isoformat()
 
-        # Connect RPC uses POST with JSON body
         payload = {
             "pageSize": page_size,
             "filter": f"creator == '{user}'",
@@ -68,7 +68,6 @@ class MemosClient:
         memos = data.get("memos", [])
         logger.info("Memos API OK | fetched %d memos", len(memos))
 
-        # Filter by time client-side (Beijing time reference)
         results = []
         for m in memos:
             created = m.get("createTime", "")
@@ -88,9 +87,7 @@ class MemosClient:
                 "visibility": m.get("visibility", ""),
             })
 
-        logger.info(
+        logger.info("Filtered to %d memos since %s", len(results), since[:10])
-            "Filtered to %d memos since %s", len(results), since[:10]
-        )
         return results
 
     def list_all_memos_from_range(self, start_date, end_date, page_size=200):
@@ -110,7 +107,6 @@ class MemosClient:
         memos = data.get("memos", [])
         logger.info("Memos API OK | fetched %d memos total", len(memos))
 
-        # Client-side date range filtering
         results = []
         for m in memos:
             created = m.get("createTime", "")
@@ -136,8 +132,76 @@ class MemosClient:
         )
         return results
 
+    def list_memos_by_tag(self, tag, page_size=200):
+        """Fetch all memos containing a specific tag.
+
+        Useful for retrieving memos across multiple days related to a specific
+        topic (e.g. a training course that spans a full week).
+        """
+        payload = {"pageSize": page_size}
+
+        resp = self.session.post(
+            f"{self.base_url}/memos.api.v1.MemoService/ListMemos",
+            json=payload,
+            timeout=15
+        )
+        resp.raise_for_status()
+        data = resp.json()
+        memos = data.get("memos", [])
+
+        results = []
+        for m in memos:
+            content_text = m.get("content", "")
+            tags = self._extract_tags(content_text)
+            if tag not in tags:
+                continue
+
+            results.append({
+                "id": m.get("name", "").split("/")[-1],
+                "content": content_text,
+                "created_at": m.get("createTime", ""),
+                "tags": tags,
+                "visibility": m.get("visibility", ""),
+            })
+
+        logger.info("Found %d memos with tag #%s", len(results), tag)
+        return results
+
+    @staticmethod
+    def aggregate_tags(memos):
+        """Count tag frequency across a list of memos.
+
+        Returns dict of {tag: count}, sorted by frequency descending.
+        """
+        counter = Counter()
+        for m in memos:
+            for tag in m.get("tags", []):
+                counter[tag] += 1
+        return dict(counter.most_common())
+
+    @staticmethod
+    def group_memos_by_tag(memos):
+        """Group memos by their tags.
+
+        Returns dict of {tag: [memo_list]}, with "无标签" key for memos
+        without any tags. A memo can appear under multiple tags.
+        """
+        groups = {}
+        untagged = []
+        for m in memos:
+            tags = m.get("tags", [])
+            if not tags:
+                untagged.append(m)
+                continue
+            for tag in tags:
+                if tag not in groups:
+                    groups[tag] = []
+                groups[tag].append(m)
+        if untagged:
+            groups["无标签"] = untagged
+        return groups
+
     @staticmethod
     def _extract_tags(content):
         """Extract #tags from memo content."""
-        import re
+        return findall(r"#(\w[\w\-]*)", content)
-        return re.findall(r"#(\w[\w\-]*)", content)