From a3256d15f4b2d9cc85aecc8ce58ecd42a558c721 Mon Sep 17 00:00:00 2001 From: hc Date: Sun, 12 Apr 2026 00:46:49 +0800 Subject: [PATCH] =?UTF-8?q?=E6=9D=83=E9=99=90=E5=92=8C=E6=A8=A1=E5=BC=8F?= =?UTF-8?q?=E7=BB=9F=E4=B8=80?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- AGENTS.md | 6 +++ SKILLS/cli-core.md | 1 + src/cc_slim/commands.py | 82 +++++++++++++++++++++++--------------- src/cc_slim/engine.py | 4 +- src/cc_slim/main.py | 10 ++--- src/cc_slim/permissions.py | 21 ++++++---- 6 files changed, 76 insertions(+), 48 deletions(-) diff --git a/AGENTS.md b/AGENTS.md index 3a36995..db6d855 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -31,12 +31,18 @@ REPL 内的会话管理命令优先使用 slash command。 REPL 内所有 slash command 由程序直接处理，不进入 agent loop。 +REPL 会显示当前模式提示符，例如 `build >` 或 `plan >`。 + 当前项目支持最小 memory：使用 `/remember` 保存长期信息，使用 `/memory` 查看。 写操作和 Bash 默认需要确认，可通过 `--auto-approve` 或 `/permissions auto-on` 跳过。 +工作区外访问和 `plan` 模式限制属于硬性边界，不通过审批放行。 + 支持 `/mode build` 和 `/mode plan` 两种模式，`plan` 为只读规划模式。 +`/clear` 是清空当前上下文的主别名，等价于 `/new`。 + 未明确说明时，使用以下默认值： - 工作目录：当前进程目录 diff --git a/SKILLS/cli-core.md b/SKILLS/cli-core.md index 4627c90..dc2b61b 100644 --- a/SKILLS/cli-core.md +++ b/SKILLS/cli-core.md @@ -24,6 +24,7 @@ - 修改已有文件时，优先使用 `Edit`，而不是 `Bash` 或 `Write` - 需要创建或覆盖文件时，优先使用 `Write` 工具，而不是 `Bash` - 会话与 memory 管理优先使用 slash command，而不是自然语言或 `Bash` 探查对应文件 +- 优先通过 `/help` 查看当前命令，而不是猜测命令格式 - 长期有价值的项目约束或偏好，优先使用 `/remember` 保存 - 高风险操作会触发确认，优先先读再改，减少无意义审批 - 复杂任务可先切换到 `/mode plan` 进行规划，再切换回 `/mode build` 执行 diff --git a/src/cc_slim/commands.py b/src/cc_slim/commands.py index 9e22807..ce40685 100644 --- a/src/cc_slim/commands.py +++ b/src/cc_slim/commands.py @@ -42,45 +42,54 @@ def handle_command( args = parsed["args"] if name == "/help": - console.print("/help - 显示命令帮助") - console.print("/clear - 清空当前上下文并创建新 session") - console.print("/history - 查看历史 session") - console.print("/resume - 恢复指定 session") - console.print("/new - 创建全新 session") - console.print("/remember - 保存长期 memory") - console.print("/memory - 查看当前项目 memory") - console.print("/mode - 查看当前模式") - console.print("/mode build - 切换到 build 模式") - console.print("/mode plan - 切换到 plan 模式") - console.print("/build - 切换到 build 模式") - console.print("/plan - 切换到 plan 模式") - console.print("/permissions - 查看当前权限状态") - console.print("/permissions auto-on - 开启自动批准") - console.print("/permissions auto-off - 关闭自动批准") + console.print("会话命令") + console.print(" /history 查看当前工作目录的历史会话") + console.print(" /resume 恢复指定会话") + console.print(" /new 创建新会话") + console.print(" /clear /new 的别名，用于清空当前上下文") + console.print("") + console.print("Memory 命令") + console.print(" /memory 查看当前项目 memory") + console.print(" /remember 追加保存长期信息") + console.print("") + console.print("模式命令") + console.print(" /mode 查看当前模式") + console.print(" /mode build 切换到 build 模式") + console.print(" /mode plan 切换到 plan 模式") + console.print(" /build /mode build 的简写") + console.print(" /plan /mode plan 的简写") + console.print("") + console.print("权限命令") + console.print(" /permissions 查看当前权限状态") + console.print(" /permissions auto-on 开启自动批准") + console.print(" /permissions auto-off 关闭自动批准") + console.print("") + console.print("退出方式") + console.print(" exit / quit 退出 REPL") return agent if name in {"/clear", "/new"}: session_meta = store.create_session(config.model) - console.print(f"已创建新 session: {session_meta.get('session_id')}") + console.print(f"已创建新会话: {session_meta.get('session_id')}") return build_agent(root, config, store, permissions, session_meta, []) if name in {"/build", "/plan"}: target_mode = "build" if name == "/build" else "plan" mode.set_mode(target_mode) permissions.set_mode(target_mode) - console.print(f"当前模式: {target_mode}") + console.print(f"模式已切换为: {target_mode}") return agent if name == "/mode": if not args: - console.print(f"当前模式: {mode.mode}") + console.print(f"模式状态\n 当前模式: {mode.mode}") return agent if args in {"build", "plan"}: mode.set_mode(args) permissions.set_mode(args) - console.print(f"当前模式: {args}") + console.print(f"模式已切换为: {args}") return agent - console.print("[red]error:[/red] 仅支持 /mode build 或 /mode plan") + console.print("命令错误: 仅支持 /mode build 或 /mode plan") return agent if name == "/history": @@ -89,45 +98,52 @@ def handle_command( if name == "/resume": if not args: - console.print("[red]error:[/red] 缺少 resume 目标") + console.print("命令错误: 缺少 resume 目标") return agent session_id = store.resolve_session_id(args) session_meta = store.load_meta(session_id) restored_history = store.load_messages(session_id) - console.print(f"已恢复 session: {session_meta.get('session_id')}") + console.print(f"已恢复会话: {session_meta.get('session_id')}") return build_agent(root, config, store, permissions, session_meta, restored_history) if name == "/memory": content = memory.read() - console.print(content or "当前项目还没有 memory。") + console.print(content or "当前项目还没有已保存的 memory。") return agent if name == "/permissions": if args == "auto-on": permissions.set_auto_approve(True) - console.print("已开启 auto_approve") + console.print("权限状态\n auto_approve: 开启") return agent if args == "auto-off": permissions.set_auto_approve(False) - console.print("已关闭 auto_approve") + console.print("权限状态\n auto_approve: 关闭") return agent status = permissions.status() - console.print(f"auto_approve: {status['auto_approve']}") - console.print(f"mode: {status['mode']}") - console.print(f"workspace: {status['workspace']}") - console.print(f"boundary: {status['boundary_policy']}") - console.print(f"自动允许: {', '.join(status['auto_allowed_tools'])}") - console.print(f"需要确认: {', '.join(status['confirm_required_tools'])}") + auto_allowed_items = list(status["auto_allowed_tools"] if isinstance(status["auto_allowed_tools"], list) else []) + confirm_required_items = list( + status["confirm_required_tools"] if isinstance(status["confirm_required_tools"], list) else [] + ) + auto_allowed = ", ".join(str(item) for item in auto_allowed_items) + confirm_required = ", ".join(str(item) for item in confirm_required_items) + console.print("权限状态") + console.print(f" auto_approve: {'开启' if status['auto_approve'] else '关闭'}") + console.print(f" mode: {status['mode']}") + console.print(f" workspace: {status['workspace']}") + console.print(f" boundary: {status['boundary_policy']}") + console.print(f" 自动允许: {auto_allowed}") + console.print(f" 需要确认: {confirm_required}") return agent if name == "/remember": if not args: - console.print("[red]error:[/red] 缺少需要保存的 memory 内容") + console.print("命令错误: 缺少需要保存的 memory 内容") return agent path = memory.append(args) console.print(f"已写入 memory: {path}") return agent - console.print(f"[red]error:[/red] 不支持的命令: {name}，输入 /help 查看帮助") + console.print(f"命令错误: 不支持的命令 {name}，输入 /help 查看帮助") return agent diff --git a/src/cc_slim/engine.py b/src/cc_slim/engine.py index 70cc58e..9d2c1f0 100644 --- a/src/cc_slim/engine.py +++ b/src/cc_slim/engine.py @@ -289,9 +289,9 @@ class Agent: def _check_tool_permission(self, name: str, payload: dict[str, Any]) -> str | None: if not self.permission_checker: return None - if not self.permission_checker.is_allowed(name, payload): + if self.permission_checker.is_hard_blocked(name, payload): return self.permission_checker.denial_reason(name, payload) - if self.permission_checker.requires_confirmation(name): + if self.permission_checker.requires_confirmation(name, payload): if not self.confirm_tool: return self.permission_checker.denial_reason(name, payload) if not self.confirm_tool(name, payload): diff --git a/src/cc_slim/main.py b/src/cc_slim/main.py index 40f42a4..47c6755 100644 --- a/src/cc_slim/main.py +++ b/src/cc_slim/main.py @@ -33,9 +33,9 @@ def render_stream(agent: Agent, user_input: str) -> None: elif event["type"] == "tool_result": output = str(event.get("output", "")) if ( - output.startswith("Tool blocked in plan mode:") - or output.startswith("Permission denied for tool:") - or output.startswith("Bash command appears to target paths outside the workspace") + output.startswith("当前处于 plan 模式，禁止执行工具:") + or output.startswith("当前未批准工具执行:") + or output.startswith("当前命令疑似指向工作区外路径，已拒绝执行 Bash") ): console.print(f"[red]{output}[/red]") else: @@ -148,10 +148,10 @@ def run( render_stream(agent, prompt) return - console.print("[bold cyan]cc-slim[/bold cyan] REPL，输入 exit 或 quit 退出。") + console.print("[bold cyan]cc-slim[/bold cyan] REPL，输入 /help 查看命令，输入 exit 或 quit 退出。") while True: try: - user_input = typer.prompt(">") + user_input = typer.prompt(f"[{mode.mode}] >") except (EOFError, KeyboardInterrupt): console.print() break diff --git a/src/cc_slim/permissions.py b/src/cc_slim/permissions.py index dbfba90..d6ca577 100644 --- a/src/cc_slim/permissions.py +++ b/src/cc_slim/permissions.py @@ -13,23 +13,28 @@ class PermissionChecker: self.auto_allowed_tools = ["Read", "Glob", "Grep"] self.confirm_required_tools = ["Write", "Edit", "Bash"] - def is_allowed(self, tool_name: str, payload: dict[str, object] | None = None) -> bool: + def is_hard_blocked(self, tool_name: str, payload: dict[str, object] | None = None) -> bool: if self.mode == "plan" and tool_name in self.confirm_required_tools: - return False + return True if tool_name == "Bash" and payload and self._bash_targets_outside_workspace(str(payload.get("command", ""))): - return False - return True + return True + return False + + def is_allowed(self, tool_name: str, payload: dict[str, object] | None = None) -> bool: + return not self.is_hard_blocked(tool_name, payload) def denial_reason(self, tool_name: str, payload: dict[str, object] | None = None) -> str: if self.mode == "plan" and tool_name in self.confirm_required_tools: - return f"Tool blocked in plan mode: {tool_name}" + return f"当前处于 plan 模式，禁止执行工具: {tool_name}" if tool_name == "Bash" and payload and self._bash_targets_outside_workspace(str(payload.get("command", ""))): - return "Bash command appears to target paths outside the workspace" - return f"Permission denied for tool: {tool_name}" + return "当前命令疑似指向工作区外路径，已拒绝执行 Bash" + return f"当前未批准工具执行: {tool_name}" - def requires_confirmation(self, tool_name: str) -> bool: + def requires_confirmation(self, tool_name: str, payload: dict[str, object] | None = None) -> bool: if self.auto_approve: return False + if self.is_hard_blocked(tool_name, payload): + return False return tool_name in self.confirm_required_tools def set_auto_approve(self, enabled: bool) -> None: