fix: 9.5-code reachable items (parse unification, goal branch, fail-open logs, M7 test, fingerprint)

Parse: agent_tools/tool_approvals column parsing unified into agent_def shared helpers (parse None = undeclared; server keeps None, worker maps to []/{} at _task_from_row, dispatch behaviour unchanged). Goal: _apply_resume_preamble goal branch proven reachable via _goal_convert_queued on a queued auto-resumed task, kept with a test and corrected docs (not dead code). Fail-open: audit write failure and preamble lookup failure now log warnings (best-effort unchanged); partial policy failure and shadow parse drops documented as known gaps. M7: shadow endpoint explicit_capability_failures_truncated test. Fingerprint: RoutingPolicy.content_fingerprint (canonical JSON sha256, version excluded) surfaced only in dry-run output, never in audit rows. M1 wording: legacy gap stated as fixed stock persisting per-task until completion, not converging.

cloud/worker.py

@@ -137,6 +137,8 @@ from types import SimpleNamespace
 from cloud.agent_def import AgentDefinition as _AgentDefinition
 from cloud.agent_def import format_instructions_header as _agent_header
 from cloud.agent_def import missing_required_tools as _missing_tools
+from cloud.agent_def import parse_agent_tool_approvals_column as _parse_tool_approvals_col
+from cloud.agent_def import parse_agent_tools_column as _parse_tools_col
 from cloud.attempts import build_resume_preamble as _resume_preamble
 from cloud.attempts import ensure_attempts_table as _ensure_attempts
 from cloud.attempts import latest_attempt as _latest_attempt
@@ -362,30 +364,22 @@ def _resolve_model(model=None, config_path=None):
 
 
 def _agent_tools_from_col(raw):
-    """agent_tools 列的容错解析：JSON 数组，非数组/脏数据为 []。"""
-    if raw is None:
-        return []
-    if isinstance(raw, list):
-        return list(raw)
-    try:
-        parsed = json.loads(raw)
-    except (TypeError, ValueError):
-        return []
-    return list(parsed) if isinstance(parsed, list) else []
+    """agent_tools 列的容错解析：唯一实现在
+    agent_def.parse_agent_tools_column（9.5-code 统一）；本
+    函数只把解析层的 None（= 未声明）映射为派发默认值
+    []——worker 每个使用点本来就读 ``or []``，所以映射后
+    的派发行为与统一前逐字一致，缺列/脏数据不炸认领。"""
+    parsed = _parse_tools_col(raw)
+    return parsed if parsed is not None else []
 
 
 def _agent_tool_approvals_from_col(raw):
-    """agent_tool_approvals 列的容错解析：JSON 对象，非对象/
-    脏数据为 {}（= 未声明），不炸认领。"""
-    if raw is None:
-        return {}
-    if isinstance(raw, dict):
-        return dict(raw)
-    try:
-        parsed = json.loads(raw)
-    except (TypeError, ValueError):
-        return {}
-    return dict(parsed) if isinstance(parsed, dict) else {}
+    """agent_tool_approvals 列的容错解析：唯一实现在
+    agent_def.parse_agent_tool_approvals_column；解析层的
+    None（= 未声明）映射为派发默认值 {}，理由同上（每个
+    使用点本来就读 ``or {}``），派发行为与统一前一致。"""
+    parsed = _parse_tool_approvals_col(raw)
+    return parsed if parsed is not None else {}
 
 
 def _task_from_row(row):
@@ -1772,7 +1766,14 @@ class Worker:
             conn.close()
 
     def _audit(self, project_id, action, detail=""):
-        """best-effort 审计：失败不影响任务状态。"""
+        """best-effort 审计：失败不影响任务状态（9.5-code
+        fail-open 清单第 1 项）。
+
+        此处是 fail-open：审计写失败时派发照常进行，运营
+        方只能从这条警告日志看到审计丢了——审计失败绝不
+        炸派发（语义不变），但失败本身不再静默。日志调用
+        本身也加 guard，日志失败时仍不炸派发。
+        """
         try:
             store = AuthStore(self.db_path)
             try:
@@ -1782,8 +1783,14 @@ class Worker:
                     store.conn.close()
                 except Exception:
                     pass
-        except Exception:
-            pass
+        except Exception as exc:
+            try:
+                log_event(self.logger, "worker_audit_failed",
+                          project_id=project_id, status="warning",
+                          message="audit write failed (%s): %s"
+                          % (action, exc))
+            except Exception:
+                pass
 
     def _notify_task_finished(self, task_id, project_id, event_type,
                               error=None):
@@ -2013,8 +2020,8 @@ class Worker:
         是模型最先看到的一段之一；位置是刻意的前置（重
         排说明应当先被看到），不是紧贴原 prompt。goal
         任务若意外走到这里也同步贴 goal（执行器取 goal
-        优先于 prompt）；该分支正常流程不可达（见下），
-        只是防御性保留。查账本失败时原样返回，不炸认领。
+        优先于 prompt）；goal 分支在正常流程可达（见下），
+        不是死代码，绝不能删。查账本失败时原样返回，不炸认领。
         """
         if task.get("coordinator_parts"):
             return task
@@ -2028,7 +2035,20 @@ class Worker:
                 attempt = _latest_attempt(conn, task["id"])
             finally:
                 conn.close()
-        except Exception:
+        except Exception as exc:
+            # 9.5-code fail-open 清单：账本查不到时不贴前言、
+            # 不炸认领（语义不变），但失败记一条警告日志，
+            # 否则“前言丢了”在运行期不可见。
+            try:
+                log_event(self.logger,
+                          "worker_resume_preamble_lookup_failed",
+                          task_id=task["id"],
+                          project_id=task.get("project_id"),
+                          status="warning",
+                          message="resume preamble lookup failed: %s"
+                          % exc)
+            except Exception:
+                pass
             return task
         preamble = _resume_preamble(attempt)
         if not preamble:
@@ -2037,13 +2057,19 @@ class Worker:
         new_task["prompt"] = preamble + "\n\n" + (
             task.get("prompt") or "")
         if task.get("goal_mode") or task.get("verify_command"):
-            # Audit fix 3（M8）：此分支正常流程不可达——
-            # build_resume_preamble 只对 platform_fresh 出
-            # 前言，而 auto_resume 把 goal 任务一律归为
-            # goal_breakpoint；且 goal 断点任务在函数入口
-            # 已被 _goal_resume_breakpoint 挡回。保留不删：
-            # 万一未来有 platform_fresh 的 goal 尝试，goal
-            # 字段不同步贴前言会静默丢说明。
+            # 9.5-code 已证实可达（非 M8 所说的防御性保留）：
+            # platform_fresh 账本只由 auto_resume 给非 goal
+            # 任务记（_record_attempt 唯一调用点），但同一
+            # 任务随后可在 queued 状态下被
+            # server._goal_convert_queued 原地转成 goal
+            # （goal_mode=1，CAS 只改 queued 行，不动
+            # auto_resume_count 与账本）；认领时它既无
+            # _resume（auto_resume 已清 resume_message），
+            # 也无 _goal_resume_breakpoint，于是走到这里。
+            # 删掉本分支会让这类转 goal 任务静默丢前言，
+            # 故保留；复活路径（_goal_revive_terminal）因
+            # resume_message 置 _resume 而在入口被挡，不
+            # 走本分支，二者不要混淆。
             new_task["goal"] = preamble + "\n\n" + (
                 task.get("goal") or task.get("prompt") or "")
         new_task["_resume_preamble_injected"] = True
@@ -5835,6 +5861,14 @@ class Worker:
         也就是写回的来源。
         查找失败（DB 异常）时返回 False 并记警告日志：复核查
         不到依据就跳过，而不是拦任务。
+
+        残留缺口措辞（9.5-code，M1）：fixes3 时期由 shadow /
+        worker_default 决策写回的旧模型既无 persisted 行又
+        无 policy 行，启发式认不出来——这部分存量是固定
+        的、逐任务持续直至完成，不是“会收敛”（收敛一词
+        会误读为随时间自动消失；实际上每个存量任务在完
+        成前每次续跑都不经复核）。新任务不再产生这类写
+        回，所以缺口不扩大，但存量本身不会被重新识别。
         """
         try:
             conn = _db_connect(self.db_path)