From 820ec69354088dc6cb0539bb3ca2db89bd6dec69 Mon Sep 17 00:00:00 2001
From: JOJO <1498581755@qq.com>
Date: Wed, 10 Jun 2026 15:14:01 +0800
Subject: [PATCH] =?UTF-8?q?feat(memory):=20=E6=96=B0=E5=A2=9E=E9=A1=B9?=
=?UTF-8?q?=E7=9B=AE=E8=AE=B0=E5=BF=86=E7=B3=BB=E7=BB=9F=EF=BC=8C=E6=95=B4?=
=?UTF-8?q?=E5=90=88=E6=80=BB=E4=BD=93+=E9=A1=B9=E7=9B=AE=E8=AE=B0?=
=?UTF-8?q?=E5=BF=86=E6=B3=A8=E5=85=A5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
- 新建 .agents/memory/ 项目记忆目录，支持 YAML frontmatter + Markdown
- 新增 update_project_memory / recall_project_memory 工具（复用 write_file/read_file）
- 新建 memory_system.txt prompt，总体长期记忆 + 项目记忆统一注入
- 记忆从 workspace_system.txt 中移出，独立为 memory_prompt 注入
- 重排 system prompt 注入顺序：主→权限→环境→对话→个性化→工作区→AGENTS.md→skills→记忆→禁用
- 压缩后选择性 invalidate 4 个 frozen keys（skills/workspace/personalization/memory）
- 前端：notebook/notebook-pen 图标、动画、状态文案、展开内容渲染
- 优化三个记忆工具的 description 和 memory_system prompt，引导模型主动调用
---
core/main_terminal_parts/context.py | 172 +++++++++++++-----
core/main_terminal_parts/tools_definition.py | 44 ++++-
core/main_terminal_parts/tools_execution.py | 51 ++++++
core/main_terminal_parts/tools_read.py | 15 ++
prompts/memory_system.txt | 14 ++
prompts/workspace_system.txt | 2 -
server/deep_compression.py | 13 +-
static/icons/notebook-pen.svg | 1 +
static/icons/notebook.svg | 1 +
.../components/chat/actions/toolRenderers.ts | 52 ++++++
static/src/stores/monitor.ts | 2 +
static/src/utils/chatDisplay.ts | 6 +
static/src/utils/icons.ts | 4 +
utils/context_manager.py | 1 -
14 files changed, 332 insertions(+), 46 deletions(-)
create mode 100644 prompts/memory_system.txt
create mode 100644 static/icons/notebook-pen.svg
create mode 100644 static/icons/notebook.svg
diff --git a/core/main_terminal_parts/context.py b/core/main_terminal_parts/context.py
index 8cd0e17..e07f5cf 100644
--- a/core/main_terminal_parts/context.py
+++ b/core/main_terminal_parts/context.py
@@ -225,6 +225,79 @@ class MainTerminalContextMixin:
# 构建上下文
return self.context_manager.build_main_context(memory)
+ def _scan_project_memories(self):
+ """扫描 .agents/memory/*.md 并解析 frontmatter 中的 name 和 description"""
+ try:
+ memory_dir = Path(self.project_path) / ".agents" / "memory"
+ except Exception:
+ return []
+ if not memory_dir.exists() or not memory_dir.is_dir():
+ return []
+ results = []
+ for md_file in sorted(memory_dir.glob("*.md")):
+ try:
+ text = md_file.read_text(encoding="utf-8")
+ name = None
+ description = None
+ if text.startswith("---"):
+ end = text.find("---", 3)
+ if end > 0:
+ frontmatter = text[3:end]
+ for line in frontmatter.strip().split("\n"):
+ line_stripped = line.strip()
+ if line_stripped.startswith("name:"):
+ name = line_stripped.split(":", 1)[1].strip()
+ elif line_stripped.startswith("description:"):
+ description = line_stripped.split(":", 1)[1].strip()
+ results.append({
+ "file": md_file.name,
+ "name": name or md_file.stem,
+ "description": description or "",
+ })
+ except Exception:
+ pass
+ return results
+
+ def _build_memory_system_content(self) -> str:
+ """构建记忆系统的 prompt 内容"""
+ template = self.load_prompt("memory_system").strip()
+ if not template:
+ return ""
+
+ try:
+ global_memory = self.memory_manager.read_main_memory()
+ global_memory_text = global_memory.strip() if global_memory else ""
+ except Exception:
+ global_memory_text = ""
+
+ project_memories = self._scan_project_memories()
+ if project_memories:
+ lines = []
+ for m in project_memories:
+ desc = m.get("description", "")
+ if desc:
+ lines.append(f".agents/memory/{m['file']}：{desc}")
+ else:
+ lines.append(f".agents/memory/{m['file']}")
+ project_memory_list = "\n".join(lines)
+ else:
+ project_memory_list = ""
+
+ result = template
+ if global_memory_text:
+ result = result.replace("{global_memory}", global_memory_text)
+ result = result.replace("[global_memory_empty]", "").replace("[/global_memory_empty]", "")
+ else:
+ result = result.replace("{global_memory}", "")
+
+ if project_memory_list:
+ result = result.replace("{project_memory_list}", project_memory_list)
+ result = result.replace("[project_memory_empty]", "").replace("[/project_memory_empty]", "")
+ else:
+ result = result.replace("{project_memory_list}", "")
+
+ return result.strip()
+
def _build_recent_conversations_message(self, limit: int = 10) -> Optional[str]:
"""构建最近对话提示（仅当前工作区）。"""
try:
@@ -318,7 +391,6 @@ class MainTerminalContextMixin:
container_memory=container_memory,
project_storage=project_storage,
file_tree=context["project_info"]["file_tree"],
- memory=context["memory"],
current_time=datetime.now().strftime("%Y-%m-%d %H"),
model_description=prompt_replacements.get("model_description", "")
)
@@ -334,31 +406,26 @@ class MainTerminalContextMixin:
personalization_config = getattr(self.context_manager, "custom_personalization_config", None) or load_personalization_config(self.data_dir)
shallow_replace_enabled = bool(personalization_config.get("auto_shallow_compress_enabled", False)) if isinstance(personalization_config, dict) else False
- skills_catalog = get_skills_catalog(private_dir=infer_private_skills_dir(self.data_dir))
- enabled_skills = merge_enabled_skills(
- personalization_config.get("enabled_skills") if isinstance(personalization_config, dict) else None,
- skills_catalog,
- personalization_config.get("skills_catalog_snapshot") if isinstance(personalization_config, dict) else None,
- )
- def _build_skills_system_prompt() -> str:
- skills_template = self.load_prompt("skills_system").strip()
- skills_list = build_skills_list(skills_catalog, enabled_skills)
- return build_skills_prompt(skills_template, skills_list) or ""
- skills_prompt = self._get_or_init_frozen_prompt(
- "frozen_skills_prompt",
- _build_skills_system_prompt,
- )
- if skills_prompt:
- messages.append({"role": "system", "content": skills_prompt})
+ # 顺序：主prompt → 权限模式 → 执行环境 → 最近对话 → 个性化配置 → 工作区信息 → AGENTS.md → skills → 记忆 → 自定义 → 禁用提示
- workspace_system = self._get_or_init_frozen_prompt(
- "frozen_workspace_prompt",
- lambda: self.context_manager._build_workspace_system_message(context) or "",
+ # 权限模式
+ permission_mode_message = self._get_or_init_frozen_mode_prompt(
+ "frozen_permission_prompt",
+ self._build_permission_mode_message,
)
- if workspace_system:
- messages.append({"role": "system", "content": workspace_system})
+ if permission_mode_message:
+ messages.append({"role": "system", "content": permission_mode_message})
+ # 执行环境
+ execution_mode_message = self._get_or_init_frozen_mode_prompt(
+ "frozen_execution_prompt",
+ self._build_execution_mode_message,
+ )
+ if execution_mode_message:
+ messages.append({"role": "system", "content": execution_mode_message})
+
+ # 最近对话
def _build_recent_conversations_prompt() -> str:
recent_conversations_enabled = (
bool(personalization_config.get("recent_conversations_prompt_enabled", False))
@@ -389,21 +456,7 @@ class MainTerminalContextMixin:
if recent_conversations_prompt:
messages.append({"role": "system", "content": recent_conversations_prompt})
- # 注入权限模式说明（根据当前模式动态生成）
- permission_mode_message = self._get_or_init_frozen_mode_prompt(
- "frozen_permission_prompt",
- self._build_permission_mode_message,
- )
- if permission_mode_message:
- messages.append({"role": "system", "content": permission_mode_message})
- execution_mode_message = self._get_or_init_frozen_mode_prompt(
- "frozen_execution_prompt",
- self._build_execution_mode_message,
- )
- if execution_mode_message:
- messages.append({"role": "system", "content": execution_mode_message})
-
- # 支持按对话覆盖的个性化配置
+ # 个性化配置
def _build_personalization_system_prompt() -> str:
personalization_block = build_personalization_prompt(personalization_config, include_header=False)
if not personalization_block:
@@ -422,7 +475,15 @@ class MainTerminalContextMixin:
if personalization_text:
messages.append({"role": "system", "content": personalization_text})
- # AGENTS.md 自动注入
+ # 工作区信息
+ workspace_system = self._get_or_init_frozen_prompt(
+ "frozen_workspace_prompt",
+ lambda: self.context_manager._build_workspace_system_message(context) or "",
+ )
+ if workspace_system:
+ messages.append({"role": "system", "content": workspace_system})
+
+ # AGENTS.md
def _build_agents_md_prompt() -> str:
agents_md_inject_enabled = bool(personalization_config.get("agents_md_auto_inject", False)) if isinstance(personalization_config, dict) else False
if not agents_md_inject_enabled:
@@ -442,8 +503,37 @@ class MainTerminalContextMixin:
if agents_md_text:
messages.append({"role": "system", "content": agents_md_text})
- # 支持按对话覆盖的自定义 system prompt（API 用途）。
- # 放在最后一个 system 消息位置，确保优先级最高，便于业务场景强约束。
+ # skills 列表
+ skills_catalog = get_skills_catalog(private_dir=infer_private_skills_dir(self.data_dir))
+ enabled_skills = merge_enabled_skills(
+ personalization_config.get("enabled_skills") if isinstance(personalization_config, dict) else None,
+ skills_catalog,
+ personalization_config.get("skills_catalog_snapshot") if isinstance(personalization_config, dict) else None,
+ )
+ def _build_skills_system_prompt() -> str:
+ skills_template = self.load_prompt("skills_system").strip()
+ skills_list = build_skills_list(skills_catalog, enabled_skills)
+ return build_skills_prompt(skills_template, skills_list) or ""
+
+ skills_prompt = self._get_or_init_frozen_prompt(
+ "frozen_skills_prompt",
+ _build_skills_system_prompt,
+ )
+ if skills_prompt:
+ messages.append({"role": "system", "content": skills_prompt})
+
+ # 记忆系统（总体长期记忆 + 项目记忆）
+ def _build_memory_system_prompt() -> str:
+ return self._build_memory_system_content()
+
+ memory_prompt = self._get_or_init_frozen_prompt(
+ "frozen_memory_prompt",
+ _build_memory_system_prompt,
+ )
+ if memory_prompt:
+ messages.append({"role": "system", "content": memory_prompt})
+
+ # API 自定义 system prompt
custom_system_prompt = self._get_or_init_frozen_prompt(
"frozen_custom_system_prompt",
lambda: (
@@ -455,7 +545,7 @@ class MainTerminalContextMixin:
if custom_system_prompt:
messages.append({"role": "system", "content": custom_system_prompt})
- # 禁用工具提示也作为“开头 system prompt”注入，便于后续统一合并。
+ # 禁用工具提示
disabled_notice = self._get_or_init_frozen_prompt(
"frozen_disabled_tools_prompt",
lambda: self._format_disabled_tool_notice() or "",
diff --git a/core/main_terminal_parts/tools_definition.py b/core/main_terminal_parts/tools_definition.py
index 3941e2b..ac95bb6 100644
--- a/core/main_terminal_parts/tools_definition.py
+++ b/core/main_terminal_parts/tools_definition.py
@@ -835,7 +835,7 @@ class MainTerminalToolsDefinitionMixin:
"type": "function",
"function": {
"name": "update_memory",
- "description": "按条目更新记忆列表（自动编号）。append 追加新条目；replace 用序号替换；delete 用序号删除。应积极记录长期偏好、项目事实、已验证经验、用户明确要求记住的内容；不要记录密钥、隐私或未确认猜测。",
+ "description": "按条目管理总体长期记忆（自动编号，跨项目通用）。append/replace/delete。当用户提到个人信息、偏好、跨项目习惯，或用户主动要求「记住xxx」时，应积极主动地调用本工具。不记录密钥、隐私、未确认猜测。与特定项目绑定的技术约定用 update_project_memory。",
"parameters": {
"type": "object",
"properties": self._inject_intent({
@@ -847,6 +847,48 @@ class MainTerminalToolsDefinitionMixin:
}
}
},
+ {
+ "type": "function",
+ "function": {
+ "name": "recall_project_memory",
+ "description": "读取指定的项目记忆文件，返回完整内容（含 frontmatter）。项目记忆存储在 .agents/memory/ 目录下。",
+ "parameters": {
+ "type": "object",
+ "properties": self._inject_intent({
+ "name": {
+ "type": "string",
+ "description": "记忆名称，对应 .agents/memory/{name}.md 文件名"
+ }
+ }),
+ "required": ["name"]
+ }
+ }
+ },
+ {
+ "type": "function",
+ "function": {
+ "name": "update_project_memory",
+ "description": "创建或覆盖项目记忆文件。当你发现有关当前项目的重要约定/决策/坑、用户表现出对项目的偏好，或用户主动要求「在当前项目里，下次要xxx/记住xxx」时，应积极主动地调用本工具。记忆名称用英文下划线，描述格式为'当xxxx时，应该索引本记忆'。不记录可从代码直接推断的或一次性信息。",
+ "parameters": {
+ "type": "object",
+ "properties": self._inject_intent({
+ "name": {
+ "type": "string",
+ "description": "记忆名称，同时作为文件名 .agents/memory/{name}.md。建议英文+下划线，如 docker_compose"
+ },
+ "description": {
+ "type": "string",
+ "description": "记忆简述，格式为'当xxxx时，应该索引本记忆'，用于在 prompt 中作为索引展示（≤100字）"
+ },
+ "content": {
+ "type": "string",
+ "description": "记忆正文（Markdown），记录具体内容"
+ }
+ }),
+ "required": ["name", "description", "content"]
+ }
+ }
+ },
{
"type": "function",
"function": {
diff --git a/core/main_terminal_parts/tools_execution.py b/core/main_terminal_parts/tools_execution.py
index d805272..219fdb8 100644
--- a/core/main_terminal_parts/tools_execution.py
+++ b/core/main_terminal_parts/tools_execution.py
@@ -185,6 +185,8 @@ class MainTerminalToolsExecutionMixin:
"vlm_analyze",
"ocr_image",
"update_memory",
+ "recall_project_memory",
+ "update_project_memory",
"conversation_search",
"conversation_review",
"todo_create",
@@ -535,6 +537,35 @@ class MainTerminalToolsExecutionMixin:
result.pop("target_dir", None)
return result
+ def _handle_update_project_memory(self, name: str, description: str, content: str) -> Dict[str, Any]:
+ """处理 update_project_memory：写入 .agents/memory/{name}.md"""
+ safe_name = str(name).strip()
+ if not safe_name or "/" in safe_name or "\\" in safe_name:
+ return {"success": False, "error": f"记忆名称不合法: {name}"}
+
+ # 确保 .agents/memory/ 目录存在
+ memory_dir = Path(self.project_path) / ".agents" / "memory"
+ try:
+ memory_dir.mkdir(parents=True, exist_ok=True)
+ except Exception as exc:
+ return {"success": False, "error": f"创建记忆目录失败: {exc}"}
+
+ # 拼接完整文件内容（YAML frontmatter + markdown）
+ full_content = f"---\nname: {safe_name}\ndescription: {description}\n---\n\n{content}\n"
+ file_path = memory_dir / f"{safe_name}.md"
+
+ try:
+ file_path.write_text(full_content, encoding="utf-8")
+ except Exception as exc:
+ return {"success": False, "error": f"写入记忆文件失败: {exc}"}
+
+ return {
+ "success": True,
+ "memory_name": safe_name,
+ "path": str(file_path),
+ "summary": f"已{'更新' if file_path.exists() else '创建'}项目记忆: {safe_name}",
+ }
+
def _mark_skill_read_from_result(self, tool_name: str, arguments: Dict[str, Any], result: Dict[str, Any]) -> None:
if tool_name not in {"read_file", "read_skill"}:
return
@@ -1504,6 +1535,26 @@ class MainTerminalToolsExecutionMixin:
index=index
)
+ elif tool_name == "recall_project_memory":
+ name = str(arguments.get("name", "")).strip()
+ if not name:
+ result = {"success": False, "error": "recall_project_memory 需要 name 参数"}
+ else:
+ result = self._handle_recall_project_memory(name)
+
+ elif tool_name == "update_project_memory":
+ name = str(arguments.get("name", "")).strip()
+ description = str(arguments.get("description", "")).strip()
+ content_text = str(arguments.get("content", "")).strip()
+ if not name:
+ result = {"success": False, "error": "update_project_memory 需要 name 参数"}
+ elif not description:
+ result = {"success": False, "error": "update_project_memory 需要 description 参数"}
+ elif not content_text:
+ result = {"success": False, "error": "update_project_memory 需要 content 参数"}
+ else:
+ result = self._handle_update_project_memory(name, description, content_text)
+
elif tool_name == "conversation_search":
query = str(arguments.get("query") or "").strip()
raw_keywords = arguments.get("keywords")
diff --git a/core/main_terminal_parts/tools_read.py b/core/main_terminal_parts/tools_read.py
index 57333a4..5068848 100644
--- a/core/main_terminal_parts/tools_read.py
+++ b/core/main_terminal_parts/tools_read.py
@@ -158,6 +158,21 @@ class MainTerminalToolsReadMixin:
result["skill_name"] = skill_name
return result
+ def _handle_recall_project_memory(self, name: str) -> Dict:
+ """处理 recall_project_memory：读取 .agents/memory/{name}.md"""
+ safe_name = str(name).strip()
+ if not safe_name or "/" in safe_name or "\\" in safe_name:
+ return {"success": False, "error": f"记忆名称不合法: {name}"}
+ file_path = f".agents/memory/{safe_name}.md"
+ read_args = {
+ "path": file_path,
+ "type": "read",
+ }
+ result = self._handle_read_tool(read_args)
+ if result.get("success"):
+ result["memory_name"] = safe_name
+ return result
+
@staticmethod
def _clamp_int(value, default, min_value=None, max_value=None):
"""将输入转换为整数并限制范围。"""
diff --git a/prompts/memory_system.txt b/prompts/memory_system.txt
new file mode 100644
index 0000000..d0a7351
--- /dev/null
+++ b/prompts/memory_system.txt
@@ -0,0 +1,14 @@
+项目记忆系统已启用。总体长期记忆来自用户的全局记录，项目记忆以文件形式存放在 .agents/memory/ 下。
+
+- 开始涉及项目约定、历史决策或已知问题的任务前，先检查项目记忆索引，如有相关记忆则用 recall_project_memory 读取
+- 当你发现有关当前项目的重要约定/决策/调试坑、用户表现出对项目的偏好，或用户说「在这个项目里下次/记住」时，用 update_project_memory 主动记录
+- 当用户提到个人信息、偏好，或说「记住xxx」时，用 update_memory 记录为总体长期记忆
+- 记忆内容为提示而非事实，执行前应与实际代码验证
+
+### 总体长期记忆
+{global_memory}
+[global_memory_empty]暂无总体长期记忆[/global_memory_empty]
+
+### 项目记忆（.agents/memory/）
+{project_memory_list}
+[project_memory_empty]暂无项目记忆[/project_memory_empty]
diff --git a/prompts/workspace_system.txt b/prompts/workspace_system.txt
index 86662fa..0e5d0b1 100644
--- a/prompts/workspace_system.txt
+++ b/prompts/workspace_system.txt
@@ -9,5 +9,3 @@
```
{file_tree}
```
-
-- **长期记忆**：{memory}
diff --git a/server/deep_compression.py b/server/deep_compression.py
index b263f93..0779eb0 100644
--- a/server/deep_compression.py
+++ b/server/deep_compression.py
@@ -457,6 +457,13 @@ async def run_deep_compression(
)
# 更新对话 metadata：压缩记录 + 清理压缩状态标记（同一对话，无切换）。
+ # 同时清除需要重建的 frozen prompt 缓存，使压缩后下一次请求自动重新加载动态内容。
+ REBUILD_FROZEN_KEYS = (
+ "frozen_skills_prompt",
+ "frozen_workspace_prompt",
+ "frozen_personalization_prompt",
+ "frozen_memory_prompt",
+ )
meta_updates = {
"compression_count": target_count,
"deep_compression_records": all_records,
@@ -469,10 +476,14 @@ async def run_deep_compression(
"compression_resume_payload": None,
"is_ultra_long_conversation": False,
}
+ for frozen_key in REBUILD_FROZEN_KEYS:
+ meta_updates[frozen_key] = None
cm.conversation_manager.update_conversation_metadata(conversation_id, meta_updates)
- # 同步内存中的 metadata，保证后续在同一对话内的判断读到最新值。
+ # 同步内存中的 metadata，清除 frozen 缓存
try:
if getattr(cm, "current_conversation_id", None) == conversation_id and isinstance(cm.conversation_metadata, dict):
+ for frozen_key in REBUILD_FROZEN_KEYS:
+ cm.conversation_metadata.pop(frozen_key, None)
cm.conversation_metadata.update(meta_updates)
except Exception:
pass
diff --git a/static/icons/notebook-pen.svg b/static/icons/notebook-pen.svg
new file mode 100644
index 0000000..6892d7c
--- /dev/null
+++ b/static/icons/notebook-pen.svg
@@ -0,0 +1 @@
+
\ No newline at end of file
diff --git a/static/icons/notebook.svg b/static/icons/notebook.svg
new file mode 100644
index 0000000..b50cf38
--- /dev/null
+++ b/static/icons/notebook.svg
@@ -0,0 +1 @@
+
\ No newline at end of file
diff --git a/static/src/components/chat/actions/toolRenderers.ts b/static/src/components/chat/actions/toolRenderers.ts
index 92d4664..685d426 100644
--- a/static/src/components/chat/actions/toolRenderers.ts
+++ b/static/src/components/chat/actions/toolRenderers.ts
@@ -100,6 +100,10 @@ export function renderEnhancedToolResult(
// 记忆类
else if (name === 'update_memory') {
return renderUpdateMemory(result, args);
+ } else if (name === 'recall_project_memory') {
+ return renderRecallProjectMemory(result, args);
+ } else if (name === 'update_project_memory') {
+ return renderUpdateProjectMemory(result, args);
} else if (name === 'conversation_search') {
return renderConversationSearch(result, args);
} else if (name === 'conversation_review') {
@@ -704,6 +708,54 @@ function renderUpdateMemory(result: any, args: any): string {
return html;
}
+function renderRecallProjectMemory(result: any, args: any): string {
+ const name = args.name || result.memory_name || '';
+ const status = formatToolStatusLabel(result, '✓ 已读取');
+ const content = result.content || result.text || '';
+
+ let html = '';
+
+ if (result.success && content) {
+ html += '';
+ html += `
${escapeHtml(content)}`;
+ html += '
${escapeHtml(String(result.error))}
`;
+ }
+
+ return html;
+}
+
+function renderUpdateProjectMemory(result: any, args: any): string {
+ const name = args.name || result.memory_name || '';
+ const description = args.description || '';
+ const content = args.content || '';
+ const status = formatToolStatusLabel(result, '✓ 已更新');
+
+ let html = '';
+
+ if (result.success && content) {
+ html += '';
+ html += '
记忆内容：
';
+ html += `
${escapeHtml(content)}`;
+ html += '