daily folder structure: each date gets its own directory

2026-06-13 09:48:49 +08:00
parent 75dbbee7cc
commit a921c40dba
8 changed files with 581 additions and 131 deletions
--- a/analyzers/daily_digest.py
+++ b/analyzers/daily_digest.py
@ -1,12 +1,12 @@
-"""Daily digest analyzer - fetches today's memos and generates AI summary."""
+"""Daily digest analyzer - fetches today's memos and generates AI analysis."""

 import json
 import logging
 import os
 import sys
-from datetime import datetime, timezone
+import subprocess
+from datetime import datetime, timezone, timedelta

-# Add parent dir to path for local imports
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

 from tools.config import load_secrets, get_output_dir
@ -16,40 +16,157 @@ from tools.formatter import format_daily_digest

 logger = logging.getLogger(__name__)

-# System prompt for daily digest
-DAILY_SYSTEM_PROMPT = """你是一个灵感整理助手。你的任务是将用户零散的灵感记录进行智能整理。
-
-请严格按照要求整理，输出纯 JSON（不要代码块标记，不要额外说明）：
-
-{
-  "summary": "一段简短的今日灵感总结，50字以内",
-  "categories": {
-    "分类名1": ["具体灵感项1", "具体灵感项2"],
-    "分类名2": ["具体灵感项3"]
-  },
-  "connections": ["跨主题关联发现1", "2-3条"],
-  "todos": ["待办事项1", "2-3条"]
-}
-
-分类原则：根据内容自然归类，分类名不超过4个字。
-关联发现：识别不同灵感之间的关联、冲突或可串联的主题。
-待办：从灵感中提取可执行的事项。"""
+TZ_BEIJING = timezone(timedelta(hours=8))
+PROJECT_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))


-def parse_ai_response(text):
-    """Parse AI response, handling both pure JSON and code-block wrapped JSON."""
-    text = text.strip()
-    # Remove markdown code block wrappers if present
-    if text.startswith("```"):
-        lines = text.split("\n")
-        # Remove first and last ``` lines
-        if lines[0].startswith("```"):
-            lines = lines[1:]
-        if lines and lines[-1].strip() == "```":
-            lines = lines[:-1]
-        text = "\n".join(lines).strip()
+def utc_to_beijing(ts_str):
+    """Convert UTC timestamp string (ISO 8601 with Z) to Beijing time string."""
+    try:
+        dt = datetime.fromisoformat(ts_str.replace("Z", "+00:00"))
+        dt_bj = dt.astimezone(TZ_BEIJING)
+        return dt_bj.strftime("%Y-%m-%d %H:%M")
+    except (ValueError, AttributeError):
+        return ts_str[:16].replace("T", " ") + " (UTC?)"

-    return json.loads(text)
+
+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"))
+        if not os.path.isdir(prev_dir):
+            continue
+        files = sorted([f for f in os.listdir(prev_dir) if f.endswith(".md")], reverse=True)
+        if not files:
+            continue
+        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()
+                else:
+                    ai_section = content[ai_section_start:].strip()
+                logger.info(
+                    "Found previous digest from %s (%d chars)",
+                    prev_date.strftime("%Y-%m-%d"), len(ai_section)
+                )
+                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")
+                )
+                return prev_date, content
+    logger.info("No previous digest found in the last 7 days")
+    return None, None
+
+
+# ============================================================
+# System prompt for daily analysis
+# ============================================================
+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。
+
+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。
+
+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。
+
+4. **逐渐深入，有节奏感**：从表面现象往下挖。先指出用户说了什么，再追问为什么，再展开你的洞察。每一段都要比上一段深一层。
+
+5. **有实质内容**：不要写"这是一个有价值的思考"这种废话。要说清楚为什么有价值，值在哪里，用户可以从这个方向挖到什么。
+
+6. **关联要真实**：如果多条灵感确实指向同一个主题，就写一段贯通的分析。不要硬凑关联。
+
+7. **待办要锋利**：待办事项不是"调研X""优化Y"这种万金油。要写到"具体做什么、什么时机做、做到什么程度"的颗粒度。
+
+输出格式：纯 Markdown，不要代码块包裹，不要 JSON。
+
+文章结构参考（不是模板，不必严格遵守）：
+
+### 1. 今日概览
+一两句话点出今天思考的主旋律，并与之前的内容形成呼应。
+
+### 2. 深入分析
+融会贯通地写。引用原文要自然嵌入行文。注意与上次分析的承接关系。
+
+### 3. 待办事项
+具体、可执行的事项列表。
+
+---
+
+注意：即便只有1条灵感，也要写出深度。宁可写长，不可简略。"""
+
+
+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,
+            capture_output=True, text=True, timeout=15
+        )
+        if result.returncode != 0:
+            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,
+            capture_output=True, text=True, timeout=10
+        )
+        if not result.stdout.strip():
+            logger.info("No changes to commit")
+            return True
+
+        date_str = datetime.now(TZ_BEIJING).strftime("%Y-%m-%d %H:%M")
+        result = subprocess.run(
+            ["git", "commit", "-m", "daily digest " + date_str],
+            cwd=PROJECT_DIR,
+            capture_output=True, text=True, timeout=15
+        )
+        if result.returncode != 0 and "nothing to commit" not in result.stdout:
+            logger.warning("git commit failed: %s", result.stderr.strip())
+            return False
+
+        # git push
+        result = subprocess.run(
+            ["git", "push", "origin", "main"],
+            cwd=PROJECT_DIR,
+            capture_output=True, text=True, timeout=30
+        )
+        if result.returncode != 0:
+            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())
+        logger.info("Pushed to Gitea successfully")
+        return True
+
+    except subprocess.TimeoutExpired:
+        logger.error("git push timed out")
+        return False
+    except Exception as e:
+        logger.error("git push error: %s", e)
+        return False


 def run(memos_client, llm_client, date=None):
@ -63,82 +180,73 @@ def run(memos_client, llm_client, date=None):

    if not memos:
        logger.info("No memos today, skipping")
-        # Still generate a minimal file
-        content = format_daily_digest(
-            date=date,
-            categories={},
-            summary="今天没有记录灵感。",
-            connections=[],
-            todos=[]
-        )
-        output_dir = get_output_dir("daily")
-        filename = f"{date.strftime('%Y-%m-%d')}_digest.md"
-        filepath = os.path.join(output_dir, filename)
-
+        content = format_daily_digest(date, ai_body="今天没有记录灵感。")
+        date_str = date.strftime("%Y-%m-%d")
+        daily_dir = os.path.join(get_output_dir("daily"), date_str)
+        os.makedirs(daily_dir, exist_ok=True)
+        now_str = datetime.now(TZ_BEIJING).strftime("%H%M%S")
+        filename = date_str + "_digest_" + now_str + ".md"
+        filepath = os.path.join(daily_dir, filename)
        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 prompt
+    # Step 2: Prepare user prompt
    memo_texts = []
    for m in memos:
-        time_str = m["created_at"][:16].replace("T", " ")
-        memo_texts.append(f"[{time_str}] {m['content']}")
+        time_bj = utc_to_beijing(m["created_at"])
+        memo_texts.append("- **[" + time_bj + "]** " + m["content"].strip())

-    user_prompt = f"请分析以下灵感记录：\n\n" + "\n".join(memo_texts)
+    user_prompt = "以下是我今天的灵感记录：\n\n" + "\n\n".join(memo_texts)

-    # Step 3: Call DeepSeek API
+    # Step 3: Include previous analysis for continuity
+    prev_date, prev_analysis = read_previous_digest(date)
+    if prev_analysis:
+        user_prompt += (
+            "\n\n---\n\n"
+            "以下是我上一次的分析内容（"
+            + prev_date.strftime("%Y-%m-%d") + "），"
+            "请结合今天的灵感一起思考，保持连续性：\n\n"
+            + prev_analysis
+        )
+
+    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
    raw_response = llm_client.ask(
        system_prompt=DAILY_SYSTEM_PROMPT,
        user_prompt=user_prompt,
-        temperature=0.3
+        temperature=0.5
    )

-    # Step 4: Parse and format
-    try:
-        data = parse_ai_response(raw_response)
-    except (json.JSONDecodeError, KeyError) as e:
-        logger.error("Failed to parse AI response: %s", e)
-        logger.error("Raw response: %s", raw_response[:200])
-        data = {
-            "summary": f"AI 解析失败，请查看原始 Memos。",
-            "categories": {"未分类": [m["content"] for m in memos]},
-            "connections": [],
-            "todos": []
-        }
+    # Step 5: Wrap with frontmatter and write
+    content = format_daily_digest(date, ai_body=raw_response)

-    categories = data.get("categories", {})
-    connections = data.get("connections", [])
-    todos = data.get("todos", [])
-    summary = data.get("summary", "")
-
-    # Step 5: Format and write output
-    content = format_daily_digest(
-        date=date,
-        categories=categories,
-        summary=summary,
-        connections=connections,
-        todos=todos
-    )
-
-    output_dir = get_output_dir("daily")
-    filename = f"{date.strftime('%Y-%m-%d')}_digest.md"
-    filepath = os.path.join(output_dir, filename)
+    date_str = date.strftime("%Y-%m-%d")
+    daily_dir = os.path.join(get_output_dir("daily"), date_str)
+    os.makedirs(daily_dir, exist_ok=True)
+    now_str = datetime.now(TZ_BEIJING).strftime("%H%M%S")
+    filename = date_str + "_digest_" + now_str + ".md"
+    filepath = os.path.join(daily_dir, filename)

    with open(filepath, "w", encoding="utf-8") as f:
        f.write(content)

    logger.info(
-        "Daily digest written to %s | %d memos | %d categories",
-        filepath, len(memos), len(categories)
+        "Daily digest written to %s | %d memos | %d chars analysis",
+        filepath, len(memos), len(raw_response)
    )
+
+    # Step 6: Auto-push to Gitea
+    git_push(filepath)
+
    return filepath, len(memos)


 def main():
-    """CLI entry point."""
    logging.basicConfig(
        level=logging.INFO,
        format="%(asctime)s [%(levelname)s] %(name)s: %(message)s"
@ -161,7 +269,7 @@ def main():
    )

    filepath, count = run(memos_client, llm_client)
-    print(f"Done: {filepath} ({count} memos)")
+    print("Done: " + filepath + " (" + str(count) + " memos)")


 if __name__ == "__main__":