From 24bf6e8eafa1b701f0a161bcbe1cf1d67458c417 Mon Sep 17 00:00:00 2001 From: oliver Date: Mon, 27 Apr 2026 15:55:36 +0800 Subject: [PATCH] =?UTF-8?q?=E6=8B=86=E5=88=86=20gateway=20=E4=B8=BB?= =?UTF-8?q?=E6=B5=81=E7=A8=8B=E9=98=B6=E6=AE=B5=EF=BC=8C=E5=B9=B6=E4=BF=AE?= =?UTF-8?q?=E6=AD=A3=20run=5Fcommand=20=E7=9A=84=20cwd=20=E9=87=8D?= =?UTF-8?q?=E5=AE=9A=E5=90=91=E8=A7=82=E6=B5=8B=E8=AF=AD=E4=B9=89=E3=80=82?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 将技能清单构建与调度决策抽离为独立阶段方法，降低 handle_turn 耦合；同时让 cwd_redirected_to_sandbox 能准确反映路径回退与沙箱重定向场景。 Made-with: Cursor --- runtime/gateway.py | 225 ++++++++++++------ .../tools/experts/workspace/shell_tools.py | 35 ++- 2 files changed, 178 insertions(+), 82 deletions(-) diff --git a/runtime/gateway.py b/runtime/gateway.py index b1a3628d..d36c6e1c 100644 --- a/runtime/gateway.py +++ b/runtime/gateway.py @@ -82,6 +82,24 @@ class OclawGatewayResult: relay_ttl_keep_count: int = 0 +@dataclass(frozen=True) +class GatewayDispatchPlan: + interaction_mode: str + requested_specialist: str + manager_specialist: str + dispatch_reason: str + selected_executor: Any + dynamic_agent: dict[str, Any] | None + specialist_input_msg: StandardMessage | None + manager_exec_msg: StandardMessage | None + manager_instruction_text: str + manager_memory_mode: bool + manager_need_wiki_inject: bool | None + manager_wiki_query: str + manager_memory_write_text: str + manager_post_reply_memory_write_text: str + + class OclawGateway: def __init__(self, *, store: Any): self.store = store @@ -615,6 +633,111 @@ class OclawGateway: except Exception: pass + def _build_skill_stats(self, *, executor: Any, started_at: float, trace_local: Callable[..., None]) -> dict[str, Any]: + skill_stats: dict[str, Any] = {} + try: + reg = getattr(executor, "tools", None) + base_url = str(getattr(getattr(executor, "model", None), "base_url", "") or "") + if reg is not None: + cache_key = ( + f"base={base_url}|skill_rt={str(self.store.get_setting('AIA_SKILL_RUNTIME_ENABLED') or '')}|" + f"skill_disabled={str(self.store.get_setting('AIA_SKILL_DISABLED_NAMES') or '')}|" + f"bind_en={str(self.store.get_setting('AIA_SKILL_ROLE_BINDING_ENABLED') or '')}|" + f"bind_inherit={str(self.store.get_setting('AIA_SKILL_ROLE_BINDING_MANAGER_INHERIT') or '')}" + ) + now = time.time() + with _SKILL_MANIFEST_CACHE_LOCK: + cached = _SKILL_MANIFEST_CACHE.get(cache_key) + if cached and (now - float(cached[0])) <= _SKILL_MANIFEST_CACHE_TTL_SEC: + skill_stats = dict(cached[1] or {}) + else: + _, stats = build_skill_manifest(registry=reg, store=self.store, base_url=base_url) + skill_stats = dict(stats or {}) + _SKILL_MANIFEST_CACHE[cache_key] = (now, dict(skill_stats)) + if len(_SKILL_MANIFEST_CACHE) > 128: + oldest_key = sorted(_SKILL_MANIFEST_CACHE.items(), key=lambda kv: kv[1][0])[0][0] + _SKILL_MANIFEST_CACHE.pop(oldest_key, None) + trace_local(event_type="skill_manifest", payload={"base_url": base_url, **skill_stats}, started_at=started_at) + except Exception: + pass + return skill_stats + + def _select_dispatch_plan( + self, + *, + msg: StandardMessage, + lang: str, + executor: Any, + interaction_mode: str, + requested_specialist: str, + memory_enabled: bool, + specialist_executor_factory: Optional[Callable[[str], Any]], + trace_local: Callable[..., None], + started_at: float, + ) -> GatewayDispatchPlan: + manager_specialist = requested_specialist + dispatch_reason = "expert_direct" + selected_executor = executor + dynamic_agent: dict[str, Any] | None = None + specialist_input_msg: StandardMessage | None = None + manager_exec_msg: StandardMessage | None = None + manager_instruction_text = "" + manager_memory_mode = False + manager_need_wiki_inject: bool | None = None + manager_wiki_query = "" + manager_memory_write_text = "" + manager_post_reply_memory_write_text = "" + if interaction_mode == "expert" and callable(specialist_executor_factory): + try: + selected_executor = specialist_executor_factory(requested_specialist) + except Exception: + selected_executor = executor + dispatch_reason = "expert_factory_failed" + if interaction_mode == "comprehensive": + ( + manager_specialist, + dispatch_reason, + dynamic_agent, + instruction_text, + manager_memory_mode, + manager_need_wiki_inject, + manager_wiki_query, + manager_memory_write_text, + manager_post_reply_memory_write_text, + ) = self._manager_select_specialist(msg=msg, lang=lang, executor=executor, memory_enabled=memory_enabled) + manager_instruction_text = str(instruction_text or "").strip() + trace_local( + event_type="manager_decision", + payload={ + "interaction_mode": interaction_mode, + "manager_selected_specialist": str(manager_specialist or ""), + "manager_memory_mode": bool(manager_memory_mode), + "dispatch_reason": str(dispatch_reason or ""), + "instruction_chars": int(len(manager_instruction_text or "")), + "dynamic_agent_used": bool(dynamic_agent is not None), + "dynamic_agent_name": str((dynamic_agent or {}).get("name") or "") if isinstance(dynamic_agent, dict) else "", + "memory_write_chars": int(len(manager_memory_write_text or "")), + "post_reply_memory_write_chars": int(len(manager_post_reply_memory_write_text or "")), + }, + started_at=started_at, + ) + return GatewayDispatchPlan( + interaction_mode=interaction_mode, + requested_specialist=requested_specialist, + manager_specialist=manager_specialist, + dispatch_reason=dispatch_reason, + selected_executor=selected_executor, + dynamic_agent=dynamic_agent, + specialist_input_msg=specialist_input_msg, + manager_exec_msg=manager_exec_msg, + manager_instruction_text=manager_instruction_text, + manager_memory_mode=manager_memory_mode, + manager_need_wiki_inject=manager_need_wiki_inject, + manager_wiki_query=manager_wiki_query, + manager_memory_write_text=manager_memory_write_text, + manager_post_reply_memory_write_text=manager_post_reply_memory_write_text, + ) + def handle_turn( self, *, @@ -727,32 +850,7 @@ class OclawGateway: started_at=t0, ) - skill_stats: dict[str, Any] = {} - try: - reg = getattr(executor, "tools", None) - base_url = str(getattr(getattr(executor, "model", None), "base_url", "") or "") - if reg is not None: - cache_key = ( - f"base={base_url}|skill_rt={str(self.store.get_setting('AIA_SKILL_RUNTIME_ENABLED') or '')}|" - f"skill_disabled={str(self.store.get_setting('AIA_SKILL_DISABLED_NAMES') or '')}|" - f"bind_en={str(self.store.get_setting('AIA_SKILL_ROLE_BINDING_ENABLED') or '')}|" - f"bind_inherit={str(self.store.get_setting('AIA_SKILL_ROLE_BINDING_MANAGER_INHERIT') or '')}" - ) - now = time.time() - with _SKILL_MANIFEST_CACHE_LOCK: - cached = _SKILL_MANIFEST_CACHE.get(cache_key) - if cached and (now - float(cached[0])) <= _SKILL_MANIFEST_CACHE_TTL_SEC: - skill_stats = dict(cached[1] or {}) - else: - _, stats = build_skill_manifest(registry=reg, store=self.store, base_url=base_url) - skill_stats = dict(stats or {}) - _SKILL_MANIFEST_CACHE[cache_key] = (now, dict(skill_stats)) - if len(_SKILL_MANIFEST_CACHE) > 128: - oldest_key = sorted(_SKILL_MANIFEST_CACHE.items(), key=lambda kv: kv[1][0])[0][0] - _SKILL_MANIFEST_CACHE.pop(oldest_key, None) - _trace_local(event_type="skill_manifest", payload={"base_url": base_url, **skill_stats}, started_at=t0) - except Exception: - pass + skill_stats = self._build_skill_stats(executor=executor, started_at=t0, trace_local=_trace_local) _trace_local(event_type="memory_retrieval_started", payload={"session_id": msg.session_id}, started_at=t0) memory_context = build_memory_context( @@ -778,59 +876,30 @@ class OclawGateway: requested_specialist = normalize_requested_specialist(base_metadata.get("selected_specialist")) if requested_specialist == "memory" and not memory_enabled: requested_specialist = "generalist" - manager_specialist = requested_specialist - dispatch_reason = "expert_direct" - selected_executor = executor - dynamic_agent: dict[str, Any] | None = None - specialist_input_msg: StandardMessage | None = None - manager_exec_msg: StandardMessage | None = None - manager_instruction_text = "" - manager_memory_mode = False - manager_need_wiki_inject: bool | None = None - manager_wiki_query = "" - manager_memory_write_text = "" - manager_post_reply_memory_write_text = "" - - if interaction_mode == "expert" and callable(specialist_executor_factory): - try: - selected_executor = specialist_executor_factory(requested_specialist) - except Exception: - selected_executor = executor - dispatch_reason = "expert_factory_failed" - + plan = self._select_dispatch_plan( + msg=msg, + lang=lang, + executor=executor, + interaction_mode=interaction_mode, + requested_specialist=requested_specialist, + memory_enabled=memory_enabled, + specialist_executor_factory=specialist_executor_factory, + trace_local=_trace_local, + started_at=t0, + ) + manager_specialist = plan.manager_specialist + dispatch_reason = plan.dispatch_reason + selected_executor = plan.selected_executor + dynamic_agent = plan.dynamic_agent + specialist_input_msg = plan.specialist_input_msg + manager_exec_msg = plan.manager_exec_msg + manager_instruction_text = plan.manager_instruction_text + manager_memory_mode = plan.manager_memory_mode + manager_need_wiki_inject = plan.manager_need_wiki_inject + manager_wiki_query = plan.manager_wiki_query + manager_memory_write_text = plan.manager_memory_write_text + manager_post_reply_memory_write_text = plan.manager_post_reply_memory_write_text if interaction_mode == "comprehensive": - ( - manager_specialist, - dispatch_reason, - dynamic_agent, - instruction_text, - manager_memory_mode, - manager_need_wiki_inject, - manager_wiki_query, - manager_memory_write_text, - manager_post_reply_memory_write_text, - ) = self._manager_select_specialist( - msg=msg, - lang=lang, - executor=executor, - memory_enabled=memory_enabled, - ) - manager_instruction_text = str(instruction_text or "").strip() - _trace_local( - event_type="manager_decision", - payload={ - "interaction_mode": interaction_mode, - "manager_selected_specialist": str(manager_specialist or ""), - "manager_memory_mode": bool(manager_memory_mode), - "dispatch_reason": str(dispatch_reason or ""), - "instruction_chars": int(len(manager_instruction_text or "")), - "dynamic_agent_used": bool(dynamic_agent is not None), - "dynamic_agent_name": str((dynamic_agent or {}).get("name") or "") if isinstance(dynamic_agent, dict) else "", - "memory_write_chars": int(len(manager_memory_write_text or "")), - "post_reply_memory_write_chars": int(len(manager_post_reply_memory_write_text or "")), - }, - started_at=t0, - ) if str(manager_instruction_text or "").strip() and not bool(manager_memory_mode): try: assignment_title = "Task assignment" if str(lang or "").startswith("en") else "任务分配" diff --git a/runtime/tools/experts/workspace/shell_tools.py b/runtime/tools/experts/workspace/shell_tools.py index 9bb071e1..a5492a73 100644 --- a/runtime/tools/experts/workspace/shell_tools.py +++ b/runtime/tools/experts/workspace/shell_tools.py @@ -45,7 +45,11 @@ def run_command_tool() -> ToolSpec: import os def _default_exec_dir() -> str: - return str(resolve_workspace_path("data/workspace")) + # Keep command execution in the same sandbox namespace as write_file. + try: + return str(resolve_workspace_path("data/workspace")) + except Exception: + return str(workspace_root()) def _strip_leading_cd_chain(cmd: str) -> tuple[str, bool]: raw = str(cmd or "") @@ -139,9 +143,32 @@ def run_command_tool() -> ToolSpec: cwd_redirected_to_sandbox = False script_path_rewritten = False original_command = command - # Hard policy: always execute in sandbox root, ignore caller-supplied cwd. - workdir = _default_exec_dir() - cwd_redirected_to_sandbox = bool(str(cwd or "").strip() and str(cwd).strip() not in {".", "./", ".\\"}) + # Execute in caller-provided cwd (guarded by resolve_workspace_path), otherwise workspace root. + try: + workdir = str(resolve_workspace_path(cwd)) if cwd else _default_exec_dir() + except Exception: + # If caller path is rejected by workspace guard, fall back to workspace root. + workdir = _default_exec_dir() + if cwd: + cwd_redirected_to_sandbox = True + if cwd: + try: + requested_path = Path(cwd).expanduser().resolve() + if requested_path == workspace_root().resolve(): + # Explicit repo-root cwd still gets sandboxed to avoid writes/exec at repo root. + workdir = _default_exec_dir() + cwd_redirected_to_sandbox = True + except Exception: + pass + if cwd: + try: + requested = str(Path(cwd).expanduser().resolve()) + resolved = str(Path(workdir).expanduser().resolve()) + if requested != resolved: + cwd_redirected_to_sandbox = True + except Exception: + # Conservative signal: explicit cwd provided but could not preserve same path. + cwd_redirected_to_sandbox = True command, normalized_cd_removed = _strip_leading_cd_chain(command) command, command_rewritten = _rewrite_workspace_absolute_refs(command, workdir=workdir) command, script_path_rewritten = _rewrite_python_script_arg(command, workdir=workdir)