ai-worklog 1.0.4 → 1.0.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-worklog",
-  "version": "1.0.4",
+  "version": "1.0.6",
   "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

@@ -19,6 +19,80 @@ from pathlib import Path
 from typing import Optional
 
 
+# ─── TUI 输出 ────────────────────────────────────────────────────────────────
+
+_COLOR = sys.stdout.isatty()
+
+def _c(text: str, code: str) -> str:
+    return f"\033[{code}m{text}\033[0m" if _COLOR else text
+
+def bold(t):   return _c(t, "1")
+def dim(t):    return _c(t, "2")
+def green(t):  return _c(t, "32")
+def cyan(t):   return _c(t, "36")
+def yellow(t): return _c(t, "33")
+def gray(t):   return _c(t, "90")
+def red(t):    return _c(t, "31")
+def blue(t):   return _c(t, "34")
+
+
+def print_header(date_str: str, incremental: bool = False, prev_time: str = "") -> None:
+    line = f"  ai-worklog · {date_str}  "
+    w = max(len(line), 44)
+    print(f"\n╭{'─' * w}╮")
+    print(f"│{bold(line):<{w + (10 if _COLOR else 0)}}│")
+    print(f"╰{'─' * w}╯")
+    if incremental:
+        print(f"\n  {yellow('↻')} 增量更新模式  {dim(f'上次生成于 {prev_time}')}")
+
+
+def print_step(msg: str) -> None:
+    print(f"\n{bold(cyan('●'))} {bold(msg)}")
+
+
+def print_ok(msg: str) -> None:
+    print(f"  {green('✓')} {msg}")
+
+
+def print_info(msg: str) -> None:
+    print(f"  {dim('│')} {dim(msg)}")
+
+
+def print_warn(msg: str) -> None:
+    print(f"  {yellow('⚠')} {msg}", file=sys.stderr)
+
+
+def print_fail(msg: str) -> None:
+    print(f"  {red('✗')} {msg}", file=sys.stderr)
+
+
+def print_project_tree(data: dict, is_sessions: bool = True) -> None:
+    """打印项目列表（树形）"""
+    if not data:
+        print(f"  {gray('└─')} {gray('(无记录)')}")
+        return
+    items = list(data.items())
+    max_len = max(len(p) for p in data)
+    for i, (proj, val) in enumerate(items):
+        connector = gray("└─") if i == len(items) - 1 else gray("├─")
+        if is_sessions:
+            n_sessions = len(val)
+            n_turns = sum(len(t) for t in val)
+            stats = gray(f"{n_sessions} 对话  {n_turns} 轮")
+        else:
+            stats = gray(f"{len(val)} 条")
+        print(f"  {connector} {bold(proj.ljust(max_len))}  {stats}")
+
+
+def print_footer(log_file: Path, total_projects: int, total_turns: int) -> None:
+    w = 46
+    print(f"\n{'─' * w}")
+    rel = str(log_file).replace(str(Path.home()), "~")
+    print(f"  {green('✓')} {bold('完成')}  {total_projects} 个项目 · {total_turns} 轮对话")
+    print(f"  {dim(rel)}")
+    print(f"{'─' * w}\n")
+
+
 # ─── 配置 ───────────────────────────────────────────────────────────────────
 
 WORKLOG_CONFIG_FILE = Path.home() / ".config" / "worklog.json"
@@ -102,28 +176,42 @@ 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[list[tuple[str, str]]]]:
+def collect_claude_sessions(
+    target_date: str,
+    prev_file_counts: Optional[dict[str, int]] = None,
+) -> tuple[dict[str, list[list[tuple[str, str]]]], dict[str, int]]:
     """
-    遍历 ~/.claude/projects/ 下所有 jsonl，
-    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。
+    遍历 ~/.claude/projects/ 下所有 jsonl，提取当天的对话轮次对。
+
+    prev_file_counts: 上次生成时各文件已处理的轮次数。
+                      传入时只返回新增轮次（增量模式）。
 
-    返回: {project_name: [session1, session2, ...]}
-           session = [(user_msg, assistant_msg), ...]
+    返回:
+      sessions:    {project_name: [session, ...]}，session = [(user, assistant), ...]
+      file_counts: {jsonl_file_path: 本次处理的总轮次数}（用于保存 checkpoint）
     """
-    results: dict[str, list[list[tuple[str, str]]]] = {}
+    sessions: dict[str, list[list[tuple[str, str]]]] = {}
+    file_counts: dict[str, int] = {}
 
     if not CLAUDE_PROJECTS_DIR.exists():
-        return results
+        return sessions, file_counts
 
     for project_dir in CLAUDE_PROJECTS_DIR.iterdir():
         if not project_dir.is_dir():
             continue
         for jsonl_file in project_dir.glob("*.jsonl"):
-            _parse_claude_jsonl(jsonl_file, target_date, results)
+            turns, cwd = _parse_claude_jsonl(jsonl_file, target_date)
+            file_key = str(jsonl_file)
+            file_counts[file_key] = len(turns)
 
-    return results
+            prev = (prev_file_counts or {}).get(file_key, 0)
+            new_turns = turns[prev:]  # 增量模式下只取新增部分；首次 prev=0 取全部
+
+            if new_turns:
+                project_name = os.path.basename(cwd) if cwd else "unknown"
+                sessions.setdefault(project_name, []).append(new_turns)
+
+    return sessions, file_counts
 
 
 def _extract_user_text(content) -> str:
@@ -153,14 +241,14 @@ def _extract_assistant_text(content) -> str:
 
 
 def _parse_claude_jsonl(
-    jsonl_file: Path, target_date: str, results: dict[str, list[list[tuple[str, str]]]]
-) -> None:
+    jsonl_file: Path, target_date: str
+) -> tuple[list[tuple[str, str]], str]:
     """
     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。
+    返回 (当天所有轮次对, cwd)。
     Claude 回复只取 text 块，跳过 thinking 和 tool_use。
     """
-    session_turns: list[tuple[str, str]] = []
+    turns: list[tuple[str, str]] = []
     pending_user: str = ""
     pending_user_date: str = ""
     pending_assistant: str = ""
@@ -190,7 +278,7 @@ def _parse_claude_jsonl(
 
                     # 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))
+                        turns.append((pending_user, pending_assistant))
 
                     # 开始新 turn
                     ts = entry.get("timestamp", "")
@@ -205,19 +293,17 @@ def _parse_claude_jsonl(
                         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）
+                        pending_assistant = assistant_text  # 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)
+            turns.append((pending_user, pending_assistant))
 
     except Exception as e:
         print(f"警告: 解析 {jsonl_file} 失败: {e}", file=sys.stderr)
 
+    return turns, cwd
+
 
 # ─── Codex 数据收集 ───────────────────────────────────────────────────────────
 
@@ -455,10 +541,93 @@ def save_log(target_date: str, content: str) -> Path:
     log_dir.mkdir(parents=True, exist_ok=True)
     log_file = log_dir / f"{target_date}.md"
     log_file.write_text(content, encoding="utf-8")
-    print(f"日志已保存: {log_file}")
     return log_file
 
 
+# ─── Checkpoint（增量更新支持）────────────────────────────────────────────────
+
+def _checkpoint_path(target_date: str) -> Path:
+    year = target_date[:4]
+    return LOGS_DIR / year / f"{target_date}.meta.json"
+
+
+def load_checkpoint(target_date: str) -> Optional[dict]:
+    """读取上次生成的 checkpoint，不存在或解析失败返回 None"""
+    path = _checkpoint_path(target_date)
+    if not path.exists():
+        return None
+    try:
+        return json.loads(path.read_text())
+    except Exception:
+        return None
+
+
+def save_checkpoint(target_date: str, file_counts: dict[str, int]) -> None:
+    """保存本次处理的 checkpoint（各文件已处理轮次数）"""
+    path = _checkpoint_path(target_date)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    data = {
+        "generated_at": datetime.now().isoformat(timespec="seconds"),
+        "file_counts": file_counts,
+    }
+    path.write_text(json.dumps(data, ensure_ascii=False, indent=2))
+
+
+def generate_incremental_update(
+    target_date: str,
+    existing_log: str,
+    new_claude_data: dict[str, list[list[tuple[str, str]]]],
+    new_codex_data: dict[str, list[str]],
+) -> str:
+    """基于已有日志 + 新增对话，调用 AI 生成更新后的完整日志"""
+
+    def trim(text: str, max_chars: int) -> str:
+        return text if len(text) <= max_chars else text[:max_chars] + "…"
+
+    project_sections = []
+    for proj_name, sessions in new_claude_data.items():
+        parts = []
+        for si, turns in enumerate(sessions, 1):
+            parts.append(f"  [新增对话 {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)}")
+        total_turns = sum(len(t) for t in sessions)
+        project_sections.append(
+            f"项目: {proj_name}（新增 {len(sessions)} 个对话 {total_turns} 轮）\n" + "\n".join(parts)
+        )
+    for proj_name, msgs in new_codex_data.items():
+        parts = [f"  [Codex] {trim(m, 200)}" for m in msgs]
+        project_sections.append(f"项目: {proj_name}（Codex 新增 {len(msgs)} 条）\n" + "\n".join(parts))
+
+    new_data_text = "\n\n".join(project_sections)
+    now_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+    prompt = f"""以下是今天已生成的工作日志：
+
+{existing_log}
+
+---
+以下是在上次生成之后新增的 AI 对话记录：
+
+{new_data_text}
+
+请在已有日志基础上进行更新，要求：
+1. 更新「今日概览」中的统计数字（累计到最新）
+2. 在对应项目的「主要工作」中追加新工作内容（保留原有内容，不要重复）
+3. 如有新项目，新增对应章节
+4. 更新「今日总结」以反映全天整体工作
+5. 将页脚的生成时间改为 {now_str}
+6. 只输出完整的更新后日志，不要加任何额外说明"""
+
+    try:
+        return _call_api_claude_cli(prompt)
+    except Exception as e:
+        print(f"API 请求失败: {e}", file=sys.stderr)
+        return existing_log  # 失败时保留原有日志
+
+
 GITLAB_HOST = "gitcode.lingjingai.cn"
 
 
@@ -540,67 +709,67 @@ def git_commit_and_push(log_file: Path, target_date: str, push: bool = True) ->
         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)
+            print_fail(f"{' '.join(cmd[:2])}: {r.stderr.strip()}")
         return r
 
+    print_step("同步 Git")
+
     # ── 1. 初始化 git 仓库 ──────────────────────────────────────────────────
     is_git = run(["git", "rev-parse", "--git-dir"], check_err=False).returncode == 0
     if not is_git:
-        print("Git: 初始化仓库...")
+        print_info("初始化仓库...")
         if run(["git", "init"]).returncode != 0:
             return
         run(["git", "checkout", "-b", "main"], check_err=False)
+        print_ok("git init")
 
     # ── 2. 检查并创建 GitLab 远程（无需 glab）────────────────────────────────
     has_remote = run(["git", "remote", "get-url", "origin"], check_err=False).returncode == 0
     if not has_remote:
         repo_name = REPO_DIR.name
-        print(f"GitLab: 创建公开项目 {repo_name} ...")
+        print_info(f"在 GitLab 创建公开项目 {repo_name} ...")
         try:
             cfg = load_gitlab_config()
             ssh_url = gitlab_create_project(cfg, repo_name)
             run(["git", "remote", "add", "origin", ssh_url])
-            print(f"GitLab: 项目已创建 → {ssh_url}")
+            print_ok(f"远程仓库 → {ssh_url}")
         except Exception as e:
-            print(f"GitLab 创建失败: {e}", file=sys.stderr)
-            print("请手动设置 remote: git remote add origin <url>", file=sys.stderr)
+            print_fail(f"GitLab 创建失败: {e}")
+            print_info("请手动: git remote add origin <url>")
             return
 
     # ── 3. 确保有初始提交 ─────────────────────────────────────────────────────
     if run(["git", "rev-parse", "HEAD"], check_err=False).returncode != 0:
-        print("Git: 创建初始提交...")
         run(["git", "add", "-A"])
         run(["git", "commit", "-m", "init: 初始化工作日志仓库"])
 
     # ── 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:
         return
 
     commit_msg = f"docs: 工作日志 {target_date}"
-    print(f"Git: 提交 '{commit_msg}'")
     r = run(["git", "commit", "-m", commit_msg], check_err=False)
     if r.returncode != 0:
         if "nothing to commit" in r.stdout + r.stderr:
-            print("Git: 无变更，跳过提交（已在初始提交中）")
+            print_info("无新变更，跳过提交")
         else:
-            print(f"Git commit 失败: {r.stderr.strip()}", file=sys.stderr)
+            print_fail(f"commit 失败: {r.stderr.strip()}")
             return
-        # nothing to commit 不代表不需要 push，继续走推送流程
+    else:
+        print_ok(f"commit: {dim(commit_msg)}")
 
     # ── 5. 推送 ────────────────────────────────────────────────────────────────
     if push:
-        print("Git: 推送到远程...")
         r = run(["git", "push", "--set-upstream", "origin", "main"], check_err=False)
         if r.returncode != 0:
             r = run(["git", "push"])
         if r.returncode == 0:
-            print("Git: 推送成功 ✓")
+            print_ok("push → origin/main")
         else:
-            print(f"Git push 失败: {r.stderr.strip()}", file=sys.stderr)
+            print_fail(f"push 失败: {r.stderr.strip()}")
     else:
-        print("Git: 已跳过 push（--no-push 模式）")
+        print_info("已跳过 push（--no-push）")
 
 
 # ─── 主入口 ───────────────────────────────────────────────────────────────────
@@ -645,51 +814,61 @@ def main():
     args = parser.parse_args()
 
     target_date = parse_date_arg(args.date)
-    print(f"=== 收集 {target_date} 的工作日志 ===")
 
-    # 1. 收集数据
-    print("收集 Claude Code 对话记录...")
-    claude_data = collect_claude_sessions(target_date)
-    total_claude_turns = sum(
-        sum(len(turns) for turns in sessions) for sessions in claude_data.values()
+    # 1. 检查 checkpoint
+    checkpoint = load_checkpoint(target_date)
+    is_incremental = checkpoint is not None
+    prev_file_counts = checkpoint["file_counts"] if is_incremental else {}
+    prev_time = checkpoint["generated_at"].split("T")[-1] if is_incremental else ""
+
+    print_header(target_date, is_incremental, prev_time)
+
+    # 2. 收集 Claude Code 对话
+    print_step("收集 Claude Code 对话记录")
+    claude_data, file_counts = collect_claude_sessions(
+        target_date, prev_file_counts if is_incremental else None
     )
-    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_project_tree(claude_data, is_sessions=True)
+    total_claude_turns = sum(sum(len(t) for t in s) for s in claude_data.values())
 
-    print("收集 Codex 对话记录...")
+    # 3. 收集 Codex 对话
+    print_step("收集 Codex 对话记录")
     codex_data = collect_codex_sessions(target_date)
+    print_project_tree(codex_data, is_sessions=False)
     total_codex = sum(len(v) for v in codex_data.values())
-    print(f"  找到 {len(codex_data)} 个项目，{total_codex} 条消息")
+
+    total_projects = len(set(list(claude_data.keys()) + list(codex_data.keys())))
+    total_turns = total_claude_turns + total_codex
 
     if args.dry_run:
-        print("\n[dry-run] 数据预览：")
-        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[:60]}")
+        print(f"\n{dim('[dry-run] 预览完毕，未调用 API')}\n")
         return
 
-    # 2. 生成摘要（通过 claude -p 复用 Claude Code CLI 认证）
-    print("调用 Claude API 生成摘要...")
-    content = generate_summary(target_date, claude_data, codex_data)
+    # 4. 生成日志
+    if is_incremental:
+        if total_claude_turns == 0 and total_codex == 0:
+            print(f"\n  {yellow('─')} 没有新增内容，无需更新\n")
+            return
+        print_step(f"调用 {bold(MODEL)} 更新日志（增量）")
+        log_file_path = LOGS_DIR / target_date[:4] / f"{target_date}.md"
+        existing_log = log_file_path.read_text(encoding="utf-8") if log_file_path.exists() else ""
+        content = generate_incremental_update(target_date, existing_log, claude_data, codex_data)
+    else:
+        print_step(f"调用 {bold(MODEL)} 生成摘要")
+        content = generate_summary(target_date, claude_data, codex_data)
 
-    # 4. 保存文件
+    # 5. 保存文件 + checkpoint
+    print_step("保存日志")
     log_file = save_log(target_date, content)
+    save_checkpoint(target_date, file_counts)
+    rel = str(log_file).replace(str(Path.home()), "~")
+    print_ok(rel)
 
-    # 5. Git 操作
+    # 6. Git 操作
     if not args.no_git:
         git_commit_and_push(log_file, target_date, push=not args.no_push)
 
-    print(f"\n完成！日志文件: {log_file}")
+    print_footer(log_file, total_projects, total_turns)
 
 
 if __name__ == "__main__":