_registry.json
5.1 KB
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
[
{
"id": "ceo",
"name": "CEO / 首席执行官",
"type": "cxo",
"description": "公司战略规划、团队管理、OKR 制定、任务委派",
"benchmarkScore": 97,
"trainedAt": "2026-03-20",
"usedCount": 10
},
{
"id": "qa-judge",
"name": "QA 仲裁者",
"type": "specialist",
"description": "质量验收、OKR 进度审查、标准化仲裁",
"benchmarkScore": 95,
"trainedAt": "2026-03-25",
"usedCount": 82
},
{
"id": "workgraph-planning",
"name": "WorkGraph 规划 Agent",
"type": "meta-planning",
"description": "全局唯一一份 work graph 规划本体（R-15.2）：每个 org node 每次接到任务时跑一次，读角色定义 + 任务 + artifacts/ + 上一轮 remediation_note，产出 work graph（org node 级 workflow）→ 过五项写入前校验 → 落库交回图 id。设计 docs/done/system-design/27-p3-p4-work-graph-console.md §四.D.0 / §五 T11",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 29
},
{
"id": "validator-triager",
"name": "失败归因 Agent",
"type": "validator",
"description": "org-validator work graph 的第一个节点：判失败根因，产出 fail_reason(impl_bug/input_underspecified/skill_defect/infra_flake) + remediation_note，透传回 org 层选边。skill_defect 挂起等人工，不触发 L4。设计 docs/done/system-design/27-p3-p4-work-graph-console.md §四.D.2.2",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 6
},
{
"id": "org-validator",
"name": "org 层全局验收 Agent",
"type": "validator",
"description": "org graph 末节点：按 artifacts/ 里的验收契约判「整个任务达成了吗」(task-goal 级)，收敛时透传 success/fail + fail_reason + remediation_note 三值。设计 docs/done/system-design/27-p3-p4-work-graph-console.md §四.D.3",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 11
},
{
"id": "planner-validator",
"name": "规划产物验收 Agent",
"type": "validator",
"description": "work 层验收本体：验 BRD/PRD/SDD，并额外审「这套验收标准是否可度量 / 覆盖原始需求 / 存在平凡通过漏洞」。设计 docs/done/system-design/27-p3-p4-work-graph-console.md §四.D.3 末段",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 10
},
{
"id": "coder-validator",
"name": "代码产出验收 Agent",
"type": "validator",
"description": "work 层验收本体：按 contract + acceptance_criteria 逐条验代码，测试必须真跑过，专抓空断言/跳过充数/只有 happy path 等假绿。设计 docs/done/system-design/27-p3-p4-work-graph-console.md §四.D.3",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 15
},
{
"id": "brd-planning",
"name": "BRD 规划 Agent",
"type": "planning",
"description": "work graph 内的规划节点：产出业务需求文档（BRD）。命名见 27-naming-conventions.md §二 规划类 {产出物}-planning。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 1
},
{
"id": "prd-planning",
"name": "PRD 规划 Agent",
"type": "planning",
"description": "work graph 内的规划节点：产出产品需求文档（PRD）。命名见 27-naming-conventions.md §二。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 7
},
{
"id": "sdd-planning",
"name": "SDD 规划 Agent",
"type": "planning",
"description": "work graph 内的规划节点：产出系统设计文档（SDD）。命名见 27-naming-conventions.md §二。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 5
},
{
"id": "goal-benchmark",
"name": "验收基准产出 Agent",
"type": "planning",
"description": "验收产出三件套之一：为当前 goal 产出可度量的验收基准，写入 root run 的 artifacts/。由 validator 侧消费执行（R-9.3 设计者定标准、执行者不定标准）。见 27-e2e-task-execution-flow.md §十一。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 1
},
{
"id": "benchmark-validator",
"name": "基准验收 Agent",
"type": "validator",
"description": "org-validator work graph 的节点：执行 goal-benchmark 产出的验收基准并给出判决。见 27-p3-p4-work-graph-console.md §四.D.3。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 3
},
{
"id": "testdata-validator",
"name": "测试数据验收 Agent",
"type": "validator",
"description": "org-validator work graph 的节点：执行 goal-test-data 产出的测试数据并给出判决。见 27-p3-p4-work-graph-console.md §四.D.3。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 1
},
{
"id": "task-planning",
"name": "需求拆分规划师",
"type": "meta-planning",
"description": "把一个人类需求（goal）切成 N 条可独立交付、可独立验收的 task（DAG），产出 draft 清单交人确认。见 27-p3c-goal-task-model.md §四.9。",
"benchmarkScore": null,
"trainedAt": null,
"usedCount": 0
}
]