agentsclimarketplace

Credit risk offline

Skill sadjjk/public-agent-skills-plugins/skills/credit-risk-offline

风控离线建模辅助工具(Credit Risk Offline Modeling),管理样本/特征/模型三层全流程。 **当以下情况时使用此 Skill**: (1) 查看风控项目、新建风控项目、按日期或名称筛选项目 (2) 样本操作:添加样本、生成样本报告、样本确认 (3) 特征操作:添加特征、特征筛选、特征探查、特征确认 (4) 模型操作:模型训练、模型评估、模型确认 (5) 需要查看样本状态、特征状态、模型状态From its SKILL.md

Install
npx -y skills add sadjjk/public-agent-skills-plugins --skill credit-risk-offline

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 3 commands, including `cd {skill_dir}/scripts && python3 -m cli sample --project <project_root>` and 2 more.

SKILL.md

3.4 KB, 954 tokens by cl100k_base, as published. Nobody here has run it

风控离线建模助手

作用范围

用于风控离线建模项目的三层协作:sample / feature / model。

当前支持:

  • sample 层完整流程
  • feature 层完整流程
  • model 层完整流程

命令别名

所有 references 文件统一使用以下别名,不再写完整命令:

  • $SAMPLE_SCRIPT = cd {skill_dir}/scripts && python3 -m cli sample --project <project_root>
  • $FEATURE_SCRIPT = cd {skill_dir}/scripts && python3 -m cli feature --project <project_root>
  • $MODEL_SCRIPT = cd {skill_dir}/scripts && python3 -m cli model --project <project_root>

核心流程

  1. 先识别用户意图属于项目、样本、特征还是模型
  2. 先确认 project_root;只有命中本轮会话内的明确继承条件时,轻入口才允许跳过项目选择
  3. 若 project_root 未确认,立即读取 references/common/project-bootstrap.md
    • 对"新增样本""新增特征""模型训练"这类轻入口,默认先按 project-bootstrap 自动发现并列出候选项目
    • 只有在没有候选项目,或用户明确要求直接给路径时,才向用户索要绝对路径
  4. 项目确认后,根据意图类型加载对应 knowledge:
    • 流程意图(添加/筛选/确认等推进状态的动作):references/sample/sample.md、references/feature/feature.md、references/model/model.md
    • 查询意图(查列表/查状态/查详情,不推进状态):references/sample/sample-query.md、references/feature/feature-query.md
  5. 只有当前动作确实需要时,才补读对应 references/*

核心约束

  • 先确认 project_root,再进入具体层级流程;不能把 workspace 当成业务项目目录
  • 轻入口默认先选项目;不要因为历史绑定或想象中的"当前项目"直接继承
  • 只认本轮会话内建立的项目绑定;不要把 memory、系统注入摘要或旧会话记录当作当前项目证据
  • 项目确认后遵循最小推进:只进入当前入口的下一步必填信息或必要动作
  • 项目内 sample.yaml / feature.yaml / model.yaml 只通过脚本访问,不手工编辑
  • 只有项目来自已有项目选择或本轮合法继承时,才可展开样本/特征/模型多入口菜单

按需补读

  • 风控项目 (查看/确认/选择/初始化):references/common/project-bootstrap.md
  • 样本流程(添加/修复/报告/确认):references/sample/sample.md
  • 样本通用查询(查列表/查状态/查详情):references/sample/sample-query.md
  • 特征流程(添加/筛选/报告/确认):references/feature/feature.md
  • 特征通用查询(查列表/查状态/查详情):references/feature/feature-query.md
  • 模型流程:references/model/model.md

硬停止

出现以下念头时,立即停下并回到最小推进路径:

  • 还没确认项目,就直接进入 sample / feature / model 流程
  • 用户只给了轻入口动作,却跳过项目选择直接默认继承历史项目

What ships with it: 30 files

8421.5 KB alongside SKILL.md

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.