ai-worklog 1.0.2 → 1.0.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-worklog",
-  "version": "1.0.2",
+  "version": "1.0.4",
   "description": "AI 对话工作日志自动收集工具 —— 从 Claude Code / Codex 对话记录生成每日工作日志并推送到 GitLab",
   "bin": {
     "ai-worklog": "./bin/index.js"

package/scripts/collect_work_log.py

@@ -102,15 +102,17 @@ def utc_to_local_date(utc_ts: str) -> Optional[str]:
 
 # ─── Claude Code 数据收集 ─────────────────────────────────────────────────────
 
-def collect_claude_sessions(target_date: str) -> dict[str, list[str]]:
+def collect_claude_sessions(target_date: str) -> dict[str, list[list[tuple[str, str]]]]:
     """
     遍历 ~/.claude/projects/ 下所有 jsonl，
-    按 message timestamp 过滤当天（UTC→本地），提取 user 消息，
+    按 message timestamp 过滤当天（UTC→本地），
+    每个 jsonl 文件作为一个对话，提取 (用户消息, Claude回复) 轮次对，
     按项目名（cwd basename）分组。
 
-    返回: {project_name: [message1, message2, ...]}
+    返回: {project_name: [session1, session2, ...]}
+           session = [(user_msg, assistant_msg), ...]
     """
-    results: dict[str, list[str]] = {}
+    results: dict[str, list[list[tuple[str, str]]]] = {}
 
     if not CLAUDE_PROJECTS_DIR.exists():
         return results
@@ -124,10 +126,46 @@ def collect_claude_sessions(target_date: str) -> dict[str, list[str]]:
     return results
 
 
+def _extract_user_text(content) -> str:
+    """从 message.content 中提取用户文字（过滤 tool_result 等结构）"""
+    texts = []
+    if isinstance(content, str):
+        texts = [content]
+    elif isinstance(content, list):
+        for c in content:
+            if isinstance(c, dict) and c.get("type") == "text":
+                texts.append(c.get("text", ""))
+            elif isinstance(c, str):
+                texts.append(c)
+    return " ".join(t for t in texts if t).strip()
+
+
+def _extract_assistant_text(content) -> str:
+    """从 assistant message.content 中提取最终文字回复，跳过 thinking 和 tool_use"""
+    if not isinstance(content, list):
+        return ""
+    parts = [
+        c.get("text", "").strip()
+        for c in content
+        if isinstance(c, dict) and c.get("type") == "text" and c.get("text", "").strip()
+    ]
+    return " ".join(parts)
+
+
 def _parse_claude_jsonl(
-    jsonl_file: Path, target_date: str, results: dict[str, list[str]]
+    jsonl_file: Path, target_date: str, results: dict[str, list[list[tuple[str, str]]]]
 ) -> None:
-    """解析单个 Claude Code jsonl 文件"""
+    """
+    解析单个 Claude Code jsonl 文件。
+    每个文件视为一个对话 session，提取当天的 (用户消息, Claude回复) 轮次对。
+    Claude 回复只取 text 块，跳过 thinking 和 tool_use。
+    """
+    session_turns: list[tuple[str, str]] = []
+    pending_user: str = ""
+    pending_user_date: str = ""
+    pending_assistant: str = ""
+    cwd: str = ""
+
     try:
         with open(jsonl_file, encoding="utf-8", errors="ignore") as f:
             for line in f:
@@ -139,34 +177,43 @@ def _parse_claude_jsonl(
                 except json.JSONDecodeError:
                     continue
 
-                if entry.get("type") != "user":
-                    continue
+                if entry.get("cwd"):
+                    cwd = entry["cwd"]
 
-                # 按时间戳过滤日期
-                ts = entry.get("timestamp", "")
-                if ts and utc_to_local_date(ts) != target_date:
-                    continue
+                entry_type = entry.get("type", "")
 
-                # 提取项目名
-                cwd = entry.get("cwd", "")
-                project_name = os.path.basename(cwd) if cwd else "unknown"
-
-                # 提取消息内容
-                message = entry.get("message", {})
-                content = message.get("content", "")
-                texts = []
-                if isinstance(content, str):
-                    texts = [content]
-                elif isinstance(content, list):
-                    for c in content:
-                        if isinstance(c, dict) and c.get("type") == "text":
-                            texts.append(c.get("text", ""))
-                        elif isinstance(c, str):
-                            texts.append(c)
-
-                for text in texts:
-                    if text and not is_system_message(text):
-                        results.setdefault(project_name, []).append(text.strip())
+                if entry_type == "user":
+                    user_text = _extract_user_text(entry.get("message", {}).get("content", ""))
+
+                    if not user_text or is_system_message(user_text):
+                        continue  # tool_result 或系统消息，跳过
+
+                    # 保存上一个 turn（日期匹配才入列）
+                    if pending_user and pending_user_date == target_date:
+                        session_turns.append((pending_user, pending_assistant))
+
+                    # 开始新 turn
+                    ts = entry.get("timestamp", "")
+                    pending_user = user_text
+                    pending_user_date = utc_to_local_date(ts) if ts else ""
+                    pending_assistant = ""
+
+                elif entry_type == "assistant":
+                    if not pending_user:
+                        continue
+                    assistant_text = _extract_assistant_text(
+                        entry.get("message", {}).get("content", [])
+                    )
+                    if assistant_text:
+                        pending_assistant = assistant_text  # 用最新文字覆盖（多步工具调用后取最终回复）
+
+        # 最后一个 pending turn
+        if pending_user and pending_user_date == target_date:
+            session_turns.append((pending_user, pending_assistant))
+
+        if session_turns:
+            project_name = os.path.basename(cwd) if cwd else "unknown"
+            results.setdefault(project_name, []).append(session_turns)
 
     except Exception as e:
         print(f"警告: 解析 {jsonl_file} 失败: {e}", file=sys.stderr)
@@ -259,41 +306,68 @@ def _call_api_claude_cli(prompt: str) -> str:
 
 def generate_summary(
     target_date: str,
-    claude_data: dict[str, list[str]],
+    claude_data: dict[str, list[list[tuple[str, str]]]],
     codex_data: dict[str, list[str]],
 ) -> str:
     """调用 Claude API 生成结构化工作日志"""
 
     # 合并数据（同项目名合并）
     all_projects: dict[str, dict] = {}
-    for proj, msgs in claude_data.items():
-        all_projects.setdefault(proj, {"claude": [], "codex": []})["claude"].extend(msgs)
+    for proj, sessions in claude_data.items():
+        all_projects.setdefault(proj, {"sessions": [], "codex": []})["sessions"].extend(sessions)
     for proj, msgs in codex_data.items():
-        all_projects.setdefault(proj, {"claude": [], "codex": []})["codex"].extend(msgs)
+        all_projects.setdefault(proj, {"sessions": [], "codex": []})["codex"].extend(msgs)
 
-    total_claude = sum(len(v["claude"]) for v in all_projects.values())
-    total_codex = sum(len(v["codex"]) for v in all_projects.values())
+    total_claude = sum(
+        sum(len(turns) for turns in data["sessions"])
+        for data in all_projects.values()
+    )
+    total_codex = sum(len(data["codex"]) for data in all_projects.values())
     total_sessions = total_claude + total_codex
 
     if total_sessions == 0:
         return _generate_empty_log(target_date)
 
-    # 构建给 AI 的原始数据
+    def trim(text: str, max_chars: int) -> str:
+        return text if len(text) <= max_chars else text[:max_chars] + "…"
+
+    def dedup_and_trim(msgs: list[str], max_chars: int = 200) -> list[str]:
+        seen: set[str] = set()
+        result = []
+        for m in msgs:
+            key = m[:80]
+            if key in seen:
+                continue
+            seen.add(key)
+            result.append(trim(m, max_chars))
+        return result
+
     project_sections = []
     for proj_name, data in all_projects.items():
-        msgs_parts = []
-        if data["claude"]:
-            msgs_parts.append(f"[Claude Code 对话 {len(data['claude'])} 条]")
-            for i, m in enumerate(data["claude"], 1):
-                # 截断过长消息
-                msg = m if len(m) <= 500 else m[:500] + "..."
-                msgs_parts.append(f"  {i}. {msg}")
+        parts = []
+
+        # Claude 对话：按 session 分，每个 session 里是轮次对
+        for si, turns in enumerate(data["sessions"], 1):
+            parts.append(f"  [对话 {si}]（{len(turns)} 轮）")
+            for user_msg, asst_msg in turns:
+                parts.append(f"    [用户] {trim(user_msg, 200)}")
+                if asst_msg:
+                    parts.append(f"    [Claude] {trim(asst_msg, 150)}")
+
+        # Codex 对话：没有回复，直接列用户消息
         if data["codex"]:
-            msgs_parts.append(f"[Codex 对话 {len(data['codex'])} 条]")
-            for i, m in enumerate(data["codex"], 1):
-                msg = m if len(m) <= 500 else m[:500] + "..."
-                msgs_parts.append(f"  {i}. {msg}")
-        project_sections.append(f"项目: {proj_name}\n" + "\n".join(msgs_parts))
+            trimmed = dedup_and_trim(data["codex"])
+            parts.append(f"  [Codex 对话]（{len(data['codex'])} 条）")
+            for m in trimmed:
+                parts.append(f"    [用户] {m}")
+
+        total_turns = sum(len(t) for t in data["sessions"])
+        header = (
+            f"项目: {proj_name}"
+            f"（Claude {len(data['sessions'])} 个对话 {total_turns} 轮，"
+            f"Codex {len(data['codex'])} 条）"
+        )
+        project_sections.append(header + "\n" + "\n".join(parts))
 
     raw_data = "\n\n".join(project_sections)
 
@@ -462,7 +536,9 @@ def git_commit_and_push(log_file: Path, target_date: str, push: bool = True) ->
     """全自动 git init → GitLab 项目创建 → commit → push"""
 
     def run(cmd: list[str], check_err: bool = True) -> subprocess.CompletedProcess:
-        r = subprocess.run(cmd, cwd=REPO_DIR, capture_output=True, text=True)
+        env = os.environ.copy()
+        env["LC_ALL"] = "C"  # 强制英文输出，保证字符串匹配不受本地化影响
+        r = subprocess.run(cmd, cwd=REPO_DIR, capture_output=True, text=True, env=env)
         if check_err and r.returncode != 0:
             print(f"命令失败: {' '.join(cmd)}\n{r.stderr.strip()}", file=sys.stderr)
         return r
@@ -496,7 +572,7 @@ def git_commit_and_push(log_file: Path, target_date: str, push: bool = True) ->
         run(["git", "add", "-A"])
         run(["git", "commit", "-m", "init: 初始化工作日志仓库"])
 
-    # ── 4. 提交日志 ────────────────────────────────────────────────────────────
+    # ── 4. 提交日志 ───────────────────────────────────────────────────────────
     rel_path = log_file.relative_to(REPO_DIR)
     print(f"Git: 添加 {rel_path}")
     if run(["git", "add", str(rel_path)]).returncode != 0:
@@ -574,8 +650,11 @@ def main():
     # 1. 收集数据
     print("收集 Claude Code 对话记录...")
     claude_data = collect_claude_sessions(target_date)
-    total_claude = sum(len(v) for v in claude_data.values())
-    print(f"  找到 {len(claude_data)} 个项目，{total_claude} 条消息")
+    total_claude_turns = sum(
+        sum(len(turns) for turns in sessions) for sessions in claude_data.values()
+    )
+    total_claude_sessions = sum(len(sessions) for sessions in claude_data.values())
+    print(f"  找到 {len(claude_data)} 个项目，{total_claude_sessions} 个对话，{total_claude_turns} 轮交互")
 
     print("收集 Codex 对话记录...")
     codex_data = collect_codex_sessions(target_date)
@@ -584,10 +663,19 @@ def main():
 
     if args.dry_run:
         print("\n[dry-run] 数据预览：")
-        for proj, msgs in {**claude_data, **codex_data}.items():
-            print(f"  [{proj}] {len(msgs)} 条")
+        for proj, sessions in claude_data.items():
+            total_turns = sum(len(t) for t in sessions)
+            print(f"  [Claude/{proj}] {len(sessions)} 个对话，{total_turns} 轮")
+            for si, turns in enumerate(sessions[:2], 1):
+                print(f"    对话{si}: {len(turns)} 轮")
+                for user_msg, asst_msg in turns[:2]:
+                    print(f"      [用户] {user_msg[:60]}")
+                    if asst_msg:
+                        print(f"      [Claude] {asst_msg[:60]}")
+        for proj, msgs in codex_data.items():
+            print(f"  [Codex/{proj}] {len(msgs)} 条")
             for m in msgs[:2]:
-                print(f"    - {m[:80]}")
+                print(f"    - {m[:60]}")
         return
 
     # 2. 生成摘要（通过 claude -p 复用 Claude Code CLI 认证）