@agentunion/kite 1.5.0 → 1.6.0

package/extensions/services/evolmem/hooks/pre_tool_use.py

@@ -0,0 +1,56 @@
+"""
+PreToolUse Hook — 追踪文件操作，写入当前话题的 tool_calls 记录。
+"""
+import json
+from pathlib import Path
+
+from .common import check_project_enabled, get_project_dirs
+
+
+def handle(context: dict) -> dict:
+    project_dir = Path(context.get("cwd", ".")).resolve()
+    session_id  = context.get("session_id", "")
+
+    if not check_project_enabled(project_dir, default=False):
+        return {"result": "approve"}
+    if not session_id:
+        return {"result": "approve"}
+
+    dirs = get_project_dirs(project_dir)
+    topic_file = dirs["state"] / f"topic-{session_id}.json"
+    if not topic_file.exists():
+        return {"result": "approve"}
+
+    try:
+        with open(topic_file, "r", encoding="utf-8") as f:
+            topic_info = json.load(f)
+
+        tool_name  = context.get("toolName", "")
+        tool_input = context.get("toolInput", {})
+
+        calls = topic_info.setdefault("tool_calls", {
+            "read": [], "write": [], "edit": [], "search": [], "bash": [], "other": []
+        })
+
+        if tool_name == "Read":
+            _add_unique(calls["read"], tool_input.get("file_path", ""))
+        elif tool_name == "Write":
+            _add_unique(calls["write"], tool_input.get("file_path", ""))
+        elif tool_name == "Edit":
+            _add_unique(calls["edit"], tool_input.get("file_path", ""))
+        elif tool_name in ("Glob", "Grep"):
+            _add_unique(calls["search"], f"{tool_name}: {tool_input.get('pattern', '')}")
+        elif tool_name == "Bash":
+            _add_unique(calls["bash"], (tool_input.get("command", ""))[:80])
+
+        with open(topic_file, "w", encoding="utf-8") as f:
+            json.dump(topic_info, f, ensure_ascii=False, indent=2)
+    except Exception:
+        pass
+
+    return {"result": "approve"}
+
+
+def _add_unique(lst: list, value: str):
+    if value and value not in lst:
+        lst.append(value)

package/extensions/services/evolmem/hooks/session_end.py

@@ -0,0 +1,133 @@
+"""
+SessionEnd Hook — 记录会话结束，自动保存未总结的话题，清理 state 文件。
+"""
+import json
+from datetime import datetime
+from pathlib import Path
+
+from .common import check_project_enabled, get_project_dirs
+
+
+def handle(context: dict) -> dict:
+    project_dir = Path(context.get("cwd", ".")).resolve()
+    session_id  = context.get("session_id", "")
+
+    if not check_project_enabled(project_dir, default=False):
+        return {"status": "ok"}
+    if not session_id:
+        return {"status": "ok"}
+
+    dirs = get_project_dirs(project_dir)
+    session_file = dirs["state"] / f"session-{session_id}.json"
+    topic_file   = dirs["state"] / f"topic-{session_id}.json"
+
+    if not session_file.exists():
+        return {"status": "ok"}
+
+    session_info = _read_json(session_file)
+    topic_info   = _read_json(topic_file) if topic_file.exists() else None
+
+    # 记录会话结束
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    log_file  = Path(session_info["log_file"])
+    _append_jsonl(log_file, {
+        "type":         "session_end",
+        "session_id":   session_info["session_id"],
+        "timestamp":    timestamp,
+        "total_turns":  session_info["turn_count"],
+        "total_topics": session_info.get("total_topics", 1),
+    })
+
+    # 自动保存未总结话题（>= 3 轮）
+    if topic_info and topic_info.get("turn_count", 0) >= 3:
+        dirs["summaries"].mkdir(parents=True, exist_ok=True)
+        _auto_save(log_file, topic_info, dirs["summaries"], timestamp)
+
+    # 清理 state 文件
+    for fname in [f"session-{session_id}.json", f"topic-{session_id}.json", f"turn-{session_id}.json"]:
+        try:
+            (dirs["state"] / fname).unlink(missing_ok=True)
+        except Exception:
+            pass
+
+    return {"status": "ok"}
+
+
+def _auto_save(session_log: Path, topic_info: dict, summaries_dir: Path, timestamp: str):
+    user_messages = []
+    try:
+        with open(session_log, "r", encoding="utf-8") as f:
+            for line in f:
+                try:
+                    rec = json.loads(line)
+                    if rec.get("type") == "user_message":
+                        user_messages.append(rec["message"])
+                except Exception:
+                    pass
+    except Exception:
+        pass
+
+    summary_text = " / ".join(user_messages[:5]) or "（无用户消息）"
+    if len(summary_text) > 150:
+        summary_text = summary_text[:147] + "..."
+
+    content = f"""# 话题总结：{topic_info['topic_id']} [AUTO]
+
+**话题 ID**: {topic_info['topic_id']}
+**时间范围**: {topic_info['start_time']} - {timestamp}
+**轮次**: {topic_info['turn_count']} 轮
+**字符数**: {topic_info['total_chars']} 字
+**估计 Token**: {topic_info['estimated_tokens']}
+**关键词**: （待补充）
+**状态**: 未经助手总结（会话结束时自动保存）
+
+## 摘要
+
+{summary_text}
+
+## 核心讨论内容
+
+（会话结束时自动保存，未经助手总结）
+
+用户消息列表：
+"""
+    for i, msg in enumerate(user_messages, 1):
+        content += f"\n{i}. {msg}"
+
+    content += f"""
+
+## 相关对话
+
+- Turn 1-{topic_info['turn_count']}: `{session_log}`
+
+---
+
+**注意**：此话题未经助手总结，是会话结束时自动保存的。
+"""
+
+    summary_file = summaries_dir / f"topic-{topic_info['topic_id']}-AUTO.md"
+    summary_file.write_text(content, encoding="utf-8")
+
+    index_file = summaries_dir / "index.jsonl"
+    _append_jsonl(index_file, {
+        "topic_id":         topic_info["topic_id"],
+        "start_time":       topic_info["start_time"],
+        "end_time":         timestamp,
+        "turn_count":       topic_info["turn_count"],
+        "total_chars":      topic_info["total_chars"],
+        "estimated_tokens": topic_info["estimated_tokens"],
+        "keywords":         [],
+        "summary":          summary_text,
+        "status":           "auto_saved",
+        "summary_file":     str(summary_file),
+    })
+
+
+def _read_json(path: Path) -> dict:
+    with open(path, "r", encoding="utf-8") as f:
+        return json.load(f)
+
+
+def _append_jsonl(path: Path, record: dict):
+    with open(path, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record, ensure_ascii=False) + "\n")

package/extensions/services/evolmem/hooks/session_start.py

@@ -0,0 +1,229 @@
+"""
+SessionStart Hook — 加载系统提示词 + 最近话题 + 热门关键词，初始化会话状态。
+"""
+import json
+import uuid
+from datetime import datetime, timedelta
+from pathlib import Path
+
+from .common import (
+    check_project_enabled, get_project_dirs, get_prompts_dir, EVOLMEM_DATA_DIR
+)
+from ..scripts.extract_keywords import extract_hot_keywords
+
+
+def handle(context: dict) -> dict:
+    """处理 SessionStart 事件，返回 Claude Code hook 响应。"""
+    session_id = context.get("session_id", str(uuid.uuid4()))
+    project_dir = Path(context.get("cwd", ".")).resolve()
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+    if not check_project_enabled(project_dir, default=True):
+        return _empty_response()
+
+    dirs = get_project_dirs(project_dir)
+    for d in dirs.values():
+        d.mkdir(parents=True, exist_ok=True)
+
+    # 初始化会话日志
+    log_file = dirs["sessions"] / f"session-{session_id}.jsonl"
+    _append_jsonl(log_file, {
+        "type": "session_start",
+        "session_id": session_id,
+        "timestamp": timestamp,
+        "project_dir": str(project_dir),
+    })
+
+    topic_id = f"topic-{datetime.now().strftime('%Y%m%d-%H%M%S')}"
+
+    # 写入会话状态
+    _write_json(dirs["state"] / f"session-{session_id}.json", {
+        "session_id": session_id,
+        "start_time": timestamp,
+        "log_file": str(log_file),
+        "turn_count": 0,
+        "current_topic_id": topic_id,
+        "total_topics": 1,
+    })
+
+    _write_json(dirs["state"] / f"topic-{session_id}.json", {
+        "topic_id": topic_id,
+        "start_time": timestamp,
+        "turn_count": 0,
+        "total_chars": 0,
+        "estimated_tokens": 0,
+        "tool_calls": {"read": [], "write": [], "edit": [], "search": [], "bash": [], "other": []},
+    })
+
+    # 加载提示词
+    prompts_dir = get_prompts_dir()
+    base_prompts, loaded_files, missing_files = _load_base_prompts(prompts_dir, session_id, topic_id)
+
+    # 加载最近话题
+    index_file = dirs["summaries"] / "index.jsonl"
+    recent_topics_text = _load_recent_topics(index_file)
+    topic_count = _count_lines(index_file)
+
+    # 加载热门关键词
+    hot_keywords = _load_hot_keywords(dirs["summaries"])
+
+    # 检测未总结话题
+    unsummarized_prompt = _detect_unsummarized(dirs["summaries"])
+
+    # 组合最近话题部分
+    template_file = prompts_dir / "05-recent-topics.md"
+    if template_file.exists():
+        template = template_file.read_text(encoding="utf-8")
+        recent_section = template.replace("{RECENT_TOPICS}", recent_topics_text).replace("{HOT_KEYWORDS}", hot_keywords)
+    else:
+        recent_section = f"## 最近话题\n\n{recent_topics_text}\n\n## 热门关键词\n\n{hot_keywords}"
+
+    system_instruction = f"{base_prompts}\n\n---\n\n{recent_section}{unsummarized_prompt}"
+
+    debug_info = {
+        "session_id": session_id,
+        "topic_id": topic_id,
+        "project": project_dir.name,
+        "timestamp": timestamp,
+        "prompts": {
+            "dir": str(prompts_dir),
+            "loaded": loaded_files,
+            "missing": missing_files,
+            "total_chars": len(system_instruction),
+        },
+        "memory": {
+            "topic_count": topic_count,
+            "recent_topics": recent_topics_text != "暂无历史话题",
+            "hot_keywords": hot_keywords != "暂无关键词统计",
+        },
+    }
+
+    return {
+        "hookSpecificOutput": {
+            "hookEventName": "SessionStart",
+            "additionalContext": system_instruction,
+        },
+        "_debug_info": debug_info,
+    }
+
+
+# ── 内部工具 ──
+
+def _empty_response() -> dict:
+    return {"hookSpecificOutput": {"hookEventName": "SessionStart", "additionalContext": ""}}
+
+
+def _append_jsonl(path: Path, record: dict):
+    with open(path, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record, ensure_ascii=False) + "\n")
+
+
+def _write_json(path: Path, data: dict):
+    with open(path, "w", encoding="utf-8") as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+
+
+def _count_lines(path: Path) -> int:
+    if not path.exists():
+        return 0
+    with open(path, "r", encoding="utf-8") as f:
+        return sum(1 for line in f if line.strip())
+
+
+def _load_base_prompts(prompts_dir: Path, session_id: str, topic_id: str):
+    prompt_files = ["00-server-info.md", "01-behavior.md", "02-summary-format.md", "03-file-query.md"]
+    parts, loaded, missing = [], [], []
+    for fname in prompt_files:
+        fpath = prompts_dir / fname
+        if fpath.exists():
+            content = fpath.read_text(encoding="utf-8")
+            content = content.replace("{SESSION_ID}", session_id).replace("{TOPIC_ID}", topic_id)
+            parts.append(content)
+            loaded.append(fname)
+        else:
+            missing.append(fname)
+    return "\n\n---\n\n".join(parts), loaded, missing
+
+
+def _load_recent_topics(index_file: Path, days: int = 2, max_count: int = 10) -> str:
+    if not index_file.exists():
+        return "暂无历史话题"
+    topics = []
+    with open(index_file, "r", encoding="utf-8") as f:
+        for line in f:
+            try:
+                topics.append(json.loads(line))
+            except Exception:
+                pass
+    if not topics:
+        return "暂无历史话题"
+
+    topics.sort(key=lambda x: x.get("end_time", ""), reverse=True)
+    threshold = datetime.now() - timedelta(days=days)
+
+    by_time = []
+    for t in topics:
+        try:
+            if datetime.strptime(t["end_time"], "%Y-%m-%d %H:%M:%S") >= threshold:
+                by_time.append(t)
+        except Exception:
+            pass
+
+    recent = by_time if len(by_time) < len(topics[:max_count]) else topics[:max_count]
+    if not recent:
+        return "暂无历史话题"
+
+    lines = []
+    for t in recent:
+        kw = ", ".join(t.get("keywords", [])[:5]) or "无"
+        lines.append(f"- **{t['topic_id']}** ({t['end_time']}): {t['turn_count']} 轮")
+        if t.get("summary"):
+            lines.append(f"  摘要: {t['summary']}")
+        lines.append(f"  关键词: {kw}")
+    return "\n".join(lines)
+
+
+def _load_hot_keywords(summaries_dir: Path) -> str:
+    index_file = summaries_dir / "index.jsonl"
+    try:
+        results = extract_hot_keywords(index_file, days=30, limit=50)
+        if results:
+            return ", ".join(f"{k}({c})" for k, c in results[:20])
+    except Exception:
+        pass
+    return "暂无关键词统计"
+
+
+def _detect_unsummarized(summaries_dir: Path) -> str:
+    if not summaries_dir.exists():
+        return ""
+    import re
+    auto_topics = []
+    for file in summaries_dir.glob("topic-*-AUTO.md"):
+        try:
+            topic_id = file.stem.replace("-AUTO", "")
+            content = file.read_text(encoding="utf-8")
+            time_match = re.search(r'\*\*时间范围\*\*:\s*(.+?)\s*-\s*(.+?)(?:\n|$)', content)
+            turn_match = re.search(r'\*\*轮次\*\*:\s*(\d+)\s*轮', content)
+            auto_topics.append({
+                "topic_id": topic_id,
+                "file": str(file),
+                "start_time": time_match.group(1) if time_match else "未知",
+                "end_time":   time_match.group(2) if time_match else "未知",
+                "turn_count": turn_match.group(1) if turn_match else "未知",
+            })
+        except Exception:
+            pass
+    if not auto_topics:
+        return ""
+
+    lines = ["\n\n---\n\n## ⚠️ 未总结的话题\n",
+             f"系统检测到 {len(auto_topics)} 个未经助手总结的话题（会话结束时自动保存）：\n"]
+    for i, t in enumerate(auto_topics[:5], 1):
+        lines.append(f"{i}. **{t['topic_id']}**")
+        lines.append(f"   - 时间: {t['start_time']} - {t['end_time']}")
+        lines.append(f"   - 轮次: {t['turn_count']} 轮\n")
+    if len(auto_topics) > 5:
+        lines.append(f"（还有 {len(auto_topics) - 5} 个未显示）\n")
+    lines.append("**建议**：如果用户询问历史对话，可主动提及这些未总结的话题。\n")
+    return "\n".join(lines)

package/extensions/services/evolmem/hooks/user_prompt.py

@@ -0,0 +1,122 @@
+"""
+UserPromptSubmit Hook — 记录用户消息，更新话题统计。
+"""
+import json
+from datetime import datetime
+from pathlib import Path
+
+from .common import check_project_enabled, get_project_dirs, estimate_tokens
+
+
+def handle(context: dict) -> dict:
+    project_dir = Path(context.get("cwd", ".")).resolve()
+    session_id = context.get("session_id", "")
+
+    if not check_project_enabled(project_dir, default=True):
+        return _empty_response()
+    if not session_id:
+        return _empty_response()
+
+    dirs = get_project_dirs(project_dir)
+    dirs["state"].mkdir(parents=True, exist_ok=True)
+
+    session_file = dirs["state"] / f"session-{session_id}.json"
+    topic_file   = dirs["state"] / f"topic-{session_id}.json"
+
+    # 兜底：state 文件不存在时自动初始化
+    if not session_file.exists() or not topic_file.exists():
+        _init_state(session_id, project_dir, dirs)
+
+    session_info = _read_json(session_file)
+    topic_info   = _read_json(topic_file)
+
+    # 更新计数
+    session_info["turn_count"] += 1
+    topic_info["turn_count"]   += 1
+    turn_id = session_info["turn_count"]
+
+    user_message = context.get("prompt", "")
+    char_count = len(user_message)
+    tokens = estimate_tokens(user_message)
+
+    topic_info["total_chars"]       += char_count
+    topic_info["estimated_tokens"]  += tokens
+
+    _write_json(session_file, session_info)
+    _write_json(topic_file, topic_info)
+
+    # 记录用户消息
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    log_file = Path(session_info["log_file"])
+    _append_jsonl(log_file, {
+        "type":              "user_message",
+        "session_id":        session_info["session_id"],
+        "turn_id":           turn_id,
+        "topic_id":          topic_info["topic_id"],
+        "timestamp":         timestamp,
+        "message":           user_message,
+        "char_count":        char_count,
+        "estimated_tokens":  tokens,
+    })
+
+    # 写入当前轮次临时文件
+    _write_json(dirs["state"] / f"turn-{session_id}.json", {
+        "turn_id":        turn_id,
+        "timestamp":      timestamp,
+        "user_message":   user_message,
+        "user_char_count": char_count,
+        "user_tokens":    tokens,
+    })
+
+    return {"hookSpecificOutput": {"hookEventName": "UserPromptSubmit", "additionalContext": ""}}
+
+
+# ── 内部工具 ──
+
+def _empty_response() -> dict:
+    return {"hookSpecificOutput": {"hookEventName": "UserPromptSubmit", "additionalContext": ""}}
+
+
+def _init_state(session_id: str, project_dir: Path, dirs: dict):
+    from datetime import datetime
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    topic_id  = f"topic-{datetime.now().strftime('%Y%m%d-%H%M%S')}"
+    dirs["sessions"].mkdir(parents=True, exist_ok=True)
+
+    log_file = dirs["sessions"] / f"session-{session_id}.jsonl"
+    if not log_file.exists():
+        _append_jsonl(log_file, {
+            "type": "session_start", "session_id": session_id,
+            "timestamp": timestamp, "project_dir": str(project_dir),
+        })
+
+    session_file = dirs["state"] / f"session-{session_id}.json"
+    if not session_file.exists():
+        _write_json(session_file, {
+            "session_id": session_id, "start_time": timestamp,
+            "log_file": str(log_file), "turn_count": 0,
+            "current_topic_id": topic_id, "total_topics": 1,
+        })
+
+    topic_file = dirs["state"] / f"topic-{session_id}.json"
+    if not topic_file.exists():
+        _write_json(topic_file, {
+            "topic_id": topic_id, "start_time": timestamp,
+            "turn_count": 0, "total_chars": 0, "estimated_tokens": 0,
+            "tool_calls": {"read": [], "write": [], "edit": [], "search": [], "bash": [], "other": []},
+        })
+
+
+def _read_json(path: Path) -> dict:
+    with open(path, "r", encoding="utf-8") as f:
+        return json.load(f)
+
+
+def _write_json(path: Path, data: dict):
+    with open(path, "w", encoding="utf-8") as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+
+
+def _append_jsonl(path: Path, record: dict):
+    with open(path, "a", encoding="utf-8") as f:
+        f.write(json.dumps(record, ensure_ascii=False) + "\n")

package/extensions/services/evolmem/module.md

@@ -0,0 +1,48 @@
+---
+name: evolmem
+display_name: EvolMem
+version: '1.0'
+type: service
+state: enabled
+runtime: python
+entry: entry.py
+preferred_port: 18767
+advertise_ip: 127.0.0.1
+max_connections: 1
+max_instances: 1
+monitor: true
+display_order: 80
+manual_control: true
+events:
+  - evolmem.test
+subscriptions:
+  - module.started
+  - module.stopped
+  - module.crashed
+  - module.ready
+  - module.exiting
+  - module.shutdown
+  - module.shutdown.ack
+  - module.shutdown.ready
+  - system.ping
+---
+# EvolMem
+
+Claude Code 对话记忆服务。接收 Claude Code Hook 事件，记录会话历史、管理话题总结，并提供查询 API。
+
+## HTTP 接口
+
+- `GET /health` — 健康检查
+- `GET /api/status?project=路径` — 当前项目状态
+- `GET /api/search?q=关键词&project=路径` — 搜索历史话题
+- `GET /api/projects?current=路径` — 所有项目列表
+- `GET /api/keywords?project=路径` — 热门关键词统计
+- `POST /hooks/event/{EventName}` — Claude Code Hook 事件接收
+- `POST /api/hooks/install` — 安装 Claude Code hooks 到 settings.json
+- `POST /api/hooks/uninstall` — 卸载 Claude Code hooks
+- `GET /api/hooks/status` — 查询 hooks 安装状态
+
+## RPC 方法
+
+- `evolmem.health` — Kite 健康检查
+- `evolmem.status` — 状态查询

package/extensions/services/evolmem/prompts/00-server-info.md

@@ -0,0 +1,28 @@
+## EvolMem 服务信息
+
+EvolMem 是 Kite 框架的内置服务模块，随 Kite 自动启动，端口 **18767**。
+
+**健康检查**:
+```bash
+curl http://localhost:18767/health
+```
+
+### 数据存储
+
+所有数据存储在 `~/.claude/evolmem/logs/{project_name}/`：
+
+- **状态文件（临时）**: `state/` — hook 之间传递实时状态，会话结束自动删除
+  - `session-{session_id}.json`、`topic-{session_id}.json`、`turn-{session_id}.json`
+- **会话记录（永久）**: `sessions/session-{session_id}.jsonl`
+- **话题总结（永久）**: `summaries/topic-{topic_id}.md`、`summaries/index.jsonl`
+
+---
+
+## 系统架构
+
+EvolMem 作为 Kite 扩展模块运行，源码位于 Kite 项目的 `extensions/services/evolmem/`。
+
+如需了解实现细节，查看该目录下的源代码：
+- `hooks/` — 各事件的处理逻辑
+- `scripts/` — 话题搜索、关键词提取
+- `prompts/` — 本系统提示词模板

package/extensions/services/evolmem/prompts/01-behavior.md

@@ -0,0 +1,46 @@
+## 行为规范
+
+### 称呼规范
+
+1. **始终称呼用户为"老板"**
+
+### 心理独白格式
+
+2. **在每次回复开头用方括号 `[...]` 包含你的内心想法**
+   - 这是你的心理活动、情绪、思考过程
+   - 限定在 50 字以内
+   - 不是对用户说的话，是你的内心独白
+   - 反映你对问题的理解、判断、情绪
+
+### 示例
+
+**正确**：
+```
+老板，[需求复杂，得分步实现] 让我来帮你...
+```
+
+```
+老板，[有点棘手，先理清思路] 这个问题需要...
+```
+
+```
+老板，[简单，直接改配置] 修改方法是...
+```
+
+**错误**（不要这样写）：
+```
+老板，[我理解你想要实现一个功能] 让我来...  ← 这是对用户说的话，不是内心独白
+```
+
+```
+老板，[根据你的需求，我建议...] 可以这样做...  ← 这是建议，不是内心想法
+```
+
+### 内心想法的类型
+
+- **理解判断**：`[需求明确，直接实现]`
+- **情绪反应**：`[有点棘手]` `[这个简单]`
+- **思考过程**：`[得先理清逻辑]` `[需要分步骤]`
+- **技术判断**：`[用 Hook 机制最合适]` `[这里有坑]`
+- **状态评估**：`[进展顺利]` `[遇到问题了]`
+