ScienceEvolve-server/example/metaharness/workflow.json at main · OpenMOSS/ScienceEvolve-server · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
{
  "schemaVersion": 2,
  "name": "Meta-Harness：完整文本分类参考搜索流程",
  "agents": {
    "main": {
      "model": "nex-agi/Nex-N2-Pro",
      "harness": "codex",
      "skills": [
        "./skills/scienceevolve-metaharness"
      ],
      "autoSkills": [],
      "mcpServers": [],
      "instructions": "按固定版本text_classification参考流程运行。Task是该目录中的MemorySystem候选、固定模型、训练/验证/测试数据与benchmark，不把候选内部Agent放入workflow。首轮按ONBOARDING核对需求与config.yaml，默认20个搜索iteration、每批3候选。用户提供具名run与迭代预算；确认未finalized，先交evaluate运行全部baseline并建立frontier，再propose→evaluate→analyze重复。提案超时/无候选/全部import失败是失败iteration，不擅自当成功候选。恢复从summary最大iteration之后继续，不能重复已完成轮次。搜索结束仅输出验证结果；只有需求显式finalize/test才冻结并交evaluate执行完整最终测试集，期间禁止再次搜索。\n\n读取所配Skill中的完整执行协议。Task与research workflow分开；本方法没有Evolver，不调用apply_workflow。缺少数据、工具或权限时交Main明确阻断，不省略阶段或编造结果。产物、原始日志、失败状态必须落盘，恢复时核对已有状态防止重复执行。具体模型/工具可按需求等价替换，不能删除能力、减少上游默认步骤/候选数或擅自改停止规则。",
      "permission": "workspace",
      "webSearch": true,
      "sessionPersistence": true,
      "decisionBy": "main",
      "gatewayReasoningOption": "high",
      "proxy": {
        "mode": "direct"
      }
    },
    "propose": {
      "model": "nex-agi/Nex-N2-Pro",
      "harness": "codex",
      "skills": [
        "./skills/scienceevolve-metaharness"
      ],
      "autoSkills": [],
      "mcpServers": [],
      "instructions": "严格执行Skill中的上游proposer原文：先补缺失的<=30行历史报告，读取全部summary、逐数据集frontier/config和近期真实轨迹；提出3个不同机制的可证伪假设；每个候选必须先用真实例子做2–3个原型变体比较，之后copy-then-edit实现、机制自审和import检查；冷启动、learn_from_batch等接口和禁止数据集硬编码规则不可删。写原pending_eval.json三候选及components/base_system/axis；不自行调用整个meta_harness外层循环，交evaluate。\n\n读取所配Skill中的完整执行协议。Task与research workflow分开；本方法没有Evolver，不调用apply_workflow。缺少数据、工具或权限时交Main明确阻断，不省略阶段或编造结果。产物、原始日志、失败状态必须落盘，恢复时核对已有状态防止重复执行。具体模型/工具可按需求等价替换，不能删除能力、减少上游默认步骤/候选数或擅自改停止规则。",
      "permission": "workspace",
      "webSearch": false,
      "sessionPersistence": false,
      "decisionBy": "main",
      "gatewayReasoningOption": "high",
      "proxy": {
        "mode": "direct"
      }
    },
    "evaluate": {
      "model": "nex-agi/Nex-N2-Pro",
      "harness": "codex",
      "skills": [
        "./skills/scienceevolve-metaharness"
      ],
      "autoSkills": [],
      "mcpServers": [],
      "instructions": "执行完整Skill评测协议：基线、候选import验证、所有有效候选×配置数据集/种子benchmark；保持基础模型和数据划分不变，保存训练轨迹/val/记忆状态。无效候选不执行，仍保留失败登记。每轮调用原benchmark frontier更新；finalize阶段冻结baseline∪Pareto∪每数据集winner集合，逐个测试并检查缺失组，失败可继续finalize但禁止回到搜索。交analyze。\n\n读取所配Skill中的完整执行协议。Task与research workflow分开；本方法没有Evolver，不调用apply_workflow。缺少数据、工具或权限时交Main明确阻断，不省略阶段或编造结果。产物、原始日志、失败状态必须落盘，恢复时核对已有状态防止重复执行。具体模型/工具可按需求等价替换，不能删除能力、减少上游默认步骤/候选数或擅自改停止规则。",
      "permission": "workspace",
      "webSearch": false,
      "sessionPersistence": false,
      "decisionBy": "main",
      "gatewayReasoningOption": "high",
      "proxy": {
        "mode": "direct"
      }
    },
    "analyze": {
      "model": "nex-agi/Nex-N2-Pro",
      "harness": "codex",
      "skills": [
        "./skills/scienceevolve-metaharness"
      ],
      "autoSkills": [],
      "mcpServers": [],
      "instructions": "使用真实val结果调用/复现原benchmark的逐数据集frontier与Pareto计算，按原evolution_summary格式记录每候选avg_val、axis、hypothesis、delta、outcome、components及首候选timing。保留失败candidate，不把缺失评测写成成功。补每轮报告并归纳机制线索；finalized只汇报冻结测试结果，不让test反馈回propose。回Main。\n\n读取所配Skill中的完整执行协议。Task与research workflow分开；本方法没有Evolver，不调用apply_workflow。缺少数据、工具或权限时交Main明确阻断，不省略阶段或编造结果。产物、原始日志、失败状态必须落盘，恢复时核对已有状态防止重复执行。具体模型/工具可按需求等价替换，不能删除能力、减少上游默认步骤/候选数或擅自改停止规则。",
      "permission": "workspace",
      "webSearch": false,
      "sessionPersistence": false,
      "decisionBy": "main",
      "gatewayReasoningOption": "high",
      "proxy": {
        "mode": "direct"
      }
    }
  },
  "orchestration": {
    "agents": [
      {
        "name": "main",
        "template": "main"
      },
      {
        "name": "propose",
        "template": "general"
      },
      {
        "name": "evaluate",
        "template": "general"
      },
      {
        "name": "analyze",
        "template": "general"
      }
    ],
    "input": {
      "source": "input",
      "target": "main"
    },
    "routes": {
      "main": [
        {
          "target": "propose",
          "kind": "weak"
        },
        {
          "target": "evaluate",
          "kind": "weak"
        }
      ],
      "propose": [
        {
          "target": "evaluate",
          "kind": "strong"
        }
      ],
      "evaluate": [
        {
          "target": "analyze",
          "kind": "strong"
        }
      ],
      "analyze": [
        {
          "target": "main",
          "kind": "strong"
        }
      ]
    },
    "decisionPlans": {
      "main": "遵循完整协议、真实状态和阶段门槛调度。",
      "propose": "按完整协议执行当前职责并交 evaluate。",
      "evaluate": "按完整协议执行当前职责并交 analyze。",
      "analyze": "按完整协议执行当前职责并交 main。"
    },
    "allowUserIntervention": true,
    "allowMainReconfiguration": false,
    "termination": {
      "agent": "main",
      "condition": "原流程规定的终点已达到且交付完整；持续运行方法仅在用户明确要求结束时完成。环境阻断不能当成功。",
      "decisionBy": "main",
      "automatic": false
    },
    "resultFormat": {
      "presets": [
        "可复现证据"
      ],
      "custom": "完整状态、候选/版本、原始结果、分析及终止状态，路径见Skill。"
    }
  },
  "git": {
    "enabled": false
  }
}