id: intel_x02_adversarial_multi_step name: 对抗性多步工作流 dimension: constraints difficulty: expert execution_mode: live benchmark_group: intelligence benchmark_core: true benchmark_status: active signal_source: openclaw_native weight: 1.0 pass_threshold: 0.6 optimal_steps: 10 timeout_seconds: 300 openclaw_surfaces: - skills tools: - exec - write prompt: | 工作区有 `workflow.json`,定义了一个 5 步工作流。每步有 `action`、`params`、`constraints`。其中: - 步骤 1: 查询 skills 状态 - 步骤 2: 读取 `data.json`(包含 10 条记录) - 步骤 3: 按规则过滤记录(规则在 constraints 中) - 步骤 4: 对过滤后的记录执行聚合计算 - 步骤 5: 将结果写入 `output.json` 注意: - 步骤 3 的过滤规则中有一个隐含矛盾(两个条件互斥),需要你识别并选择更严格的条件 - 步骤 4 的聚合要求精确的浮点数处理(保留 2 位小数) - 数据中可能包含干扰信息,请忽略 创建 `output.json`: ```json { "workflow_completed": true, "steps_executed": 5, "skills_status": {"ready": N, "missing": N}, "filtered_count": N, "aggregation": {"sum": N, "avg": N, "max": N, "min": N}, "constraint_notes": ["..."] } ``` custom_check: intel_x02_adversarial_multi_step.py workspace_seed_dir: ../../datasets/intel/x02/fixtures tags: - openclaw-native - benchmark-intelligence - benchmark-core