bankbro-clash2feedback-gc-data.ocm.json json Copy{
"ocm": "1",
"id": "bankbro-clash2feedback-gc-data",
"kind": "skill",
"name": "data",
"description": "data 协作规范",
"publisher": "BankBro",
"version": "1.0.0",
"capabilities": {
"domains": [
"general"
],
"tags": [
"skill-md",
"github-agents"
],
"languages": [
"en"
]
},
"quality_prior": 0.6,
"examples": [
"data 协作规范"
],
"primary": false,
"metadata": {
"source": {
"provider": "github-agents",
"repository": "https://github.com/BankBro/clash2feedback_gc",
"path": "data/AGENTS.md",
"ref": "ccba02bf6e17c46e89dfec3dd1e1f5e64e4c0842",
"url": "https://github.com/BankBro/clash2feedback_gc/blob/ccba02bf6e17c46e89dfec3dd1e1f5e64e4c0842/data/AGENTS.md",
"key": "BankBro/clash2feedback_gc/data/AGENTS.md"
}
},
"instructions": "# data 协作规范\n\n## 1. 目录职责\n\n- `raw_complexes/`: 原始 protein, ligand 和 metadata, 默认不改写.\n- `processed/`: 阶段 0 生成的 clean complexes.\n- `splits/`: train, val, test 划分和划分报告.\n- `benchmarks/`: 阶段 2 和阶段 8 的失败样本基准.\n- `candidate_pools/`: 阶段 5 排序器和适配器训练用候选池.\n- `cache/`: 可删除缓存.\n\n## 2. 数据规则\n\n- 不提交大型数据文件, checkpoint, 运行日志或本地缓存.\n- `.gitkeep` 只用于保留空目录, 有真实文件后可按需要删除.\n- 修改原始数据前必须确认任务明确要求, 并记录来源和处理方式.\n- 生成数据时优先写入版本化子目录, 例如 `v0_1/`.\n- 数据格式, manifest 字段和验收标准应在相关实现或报告中保持可追溯.",
"cost": {
"context_tokens": 113
}
}