add-golden

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Add to Golden Dataset

添加至黄金数据集

Multi-agent curation workflow with quality score explanations, bias detection, and version tracking.

具备质量分数说明、偏差检测和版本跟踪功能的多Agent整理工作流。

Quick Start

快速开始

bash

/add-golden https://example.com/article
/add-golden https://arxiv.org/abs/2312.xxxxx

bash

/add-golden https://example.com/article
/add-golden https://arxiv.org/abs/2312.xxxxx

Task Management (CC 2.1.16)

任务管理（CC 2.1.16）

python

undefined

python

undefined

Create main curation task

创建主整理任务

TaskCreate( subject="Add to golden dataset: {url}", description="Multi-agent curation with quality explanation", activeForm="Curating document" )

Create subtasks for 9-phase process

为9阶段流程创建子任务

phases = ["Fetch content", "Run quality analysis", "Explain scores", "Check bias", "Check diversity", "Validate", "Get approval", "Write to dataset", "Update version"] for phase in phases: TaskCreate(subject=phase, activeForm=f"{phase}ing")

---

---

Workflow Overview

工作流概述

Phase	Activities	Output
1. Input Collection	Get URL, detect content type	Document metadata
2. Fetch and Extract	Parse document structure	Structured content
3. Quality Analysis	4 parallel agents evaluate	Raw scores
4. Quality Explanation	Explain WHY each score	Score rationale
5. Bias Detection	Check for bias in content	Bias report
6. Diversity Check	Assess dataset balance	Diversity metrics
7. Validation	Schema, duplicates, gates	Validation status
8. Silver-to-Gold	Promote or mark as silver	Classification
9. Version Tracking	Track changes, rollback	Version entry

阶段	活动	输出
1. 输入收集	获取URL，检测内容类型	文档元数据
2. 获取与提取	解析文档结构	结构化内容
3. 质量分析	4个并行Agent评估	原始分数
4. 质量分数说明	解释每个分数的原因	分数依据
5. 偏差检测	检查内容中的偏差	偏差报告
6. 多样性检查	评估数据集平衡性	多样性指标
7. 验证	Schema检查、重复项检查、准入校验	验证状态
8. 从白银到黄金	升级为黄金或标记为白银	分类结果
9. 版本跟踪	跟踪变更、支持回滚	版本记录

Phase 1-2: Input and Extraction

阶段1-2：输入与提取

Detect content type: article, tutorial, documentation, research_paper.

Extract: title, sections, code blocks, key terms, metadata (author, date).

检测内容类型：文章、教程、文档、研究论文。

提取内容：标题、章节、代码块、关键术语、元数据（作者、日期）。

Phase 3: Parallel Quality Analysis (4 Agents)

阶段3：并行质量分析（4个Agent）

Launch ALL agents in ONE message with

run_in_background=True

Agent	Focus	Output
code-quality-reviewer	Accuracy, coherence, depth, relevance	Quality scores
workflow-architect	Keyword directness, paraphrase, reasoning	Difficulty level
data-pipeline-engineer	Primary/secondary domains, skill level	Tags
test-generator	Direct, paraphrased, multi-hop queries	Test queries

See Quality Scoring for detailed criteria.

通过

run_in_background=True

在一条消息中启动所有Agent。

Agent	关注重点	输出
code-quality-reviewer	准确性、连贯性、深度、相关性	质量分数
workflow-architect	关键词匹配度、转述质量、推理能力	难度等级
data-pipeline-engineer	主次领域、技能水平	标签
test-generator	直接查询、转述查询、多跳查询	测试查询

详细标准请参考Quality Scoring。

Phase 4: Quality Explanation

阶段4：质量分数说明

Each dimension gets WHY explanation:

markdown

undefined

每个维度都会说明分数原因：

markdown

undefined

Accuracy: [N.NN]/1.0

准确性: [N.NN]/1.0

Why this score:

[Specific reason with evidence] What would improve it:
[Specific improvement]

---

分数原因:

[带证据的具体理由] 改进方向:
[具体改进建议]

---

Phase 5: Bias Detection

阶段5：偏差检测

See Bias Detection Guide for patterns.

Check for:

Technology bias (favors specific tools)
Recency bias (ignores LTS versions)
Complexity bias (assumed knowledge)
Vendor bias (promotes products)
Geographic/cultural bias

Bias Score	Action
0-2	Proceed normally
3-5	Add disclaimer
6-8	Require user review
9-10	Recommend against

请参考Bias Detection Guide中的模式。

检查以下类型的偏差：

技术偏差（偏好特定工具）
时效性偏差（忽略长期支持版本）
复杂度偏差（假设用户具备特定知识）
厂商偏差（推广特定产品）
地域/文化偏差

偏差分数	操作
0-2	正常推进
3-5	添加免责声明
6-8	需用户审核
9-10	建议拒绝

Phase 6: Diversity Dashboard

阶段6：多样性仪表盘

Track dataset balance across:

Domain distribution (AI/ML, Backend, Frontend, DevOps, Security)
Difficulty distribution (trivial, easy, medium, hard, adversarial)

Impact assessment: Does new document improve or worsen diversity?

跟踪数据集在以下维度的平衡性：

领域分布（AI/ML、后端、前端、DevOps、安全）
难度分布（简单、易、中等、难、对抗性）

影响评估: 新文档会改善还是恶化数据集的多样性？

Phase 7: Validation

阶段7：验证

URL validation (no placeholders)
Schema validation (required fields)
Duplicate check (>80% similarity)
Quality gates (min sections, content length)

URL验证（无占位符）
Schema验证（必填字段检查）
重复项检查（相似度>80%）
准入校验（最小章节数、内容长度）

Phase 8: Silver-to-Gold Workflow

阶段8：从白银到黄金工作流

See Silver-Gold Promotion for criteria.

Status	Criteria	Action
GOLD	Score >= 0.75, no bias	Add to main dataset
SILVER	Score 0.55-0.74	Add to silver, track
REJECT	Score < 0.55	Do not add

Promotion criteria: 7+ days in silver, quality >= 0.75, no negative feedback.

升级标准请参考Silver-Gold Promotion。

状态	标准	操作
黄金	分数≥0.75，无偏差	添加至主数据集
白银	分数0.55-0.74	添加至白银数据集并跟踪
拒绝	分数<0.55	不添加

升级标准: 在白银数据集存放7天以上，质量≥0.75，无负面反馈。

Phase 9: Version Tracking

阶段9：版本跟踪

json

{
  "version": "1.2.3",
  "change_type": "ADD|UPDATE|REMOVE|PROMOTE",
  "document_id": "doc-123",
  "quality_score": 0.82,
  "rollback_available": true
}

Update Type	Version Bump
Add/Update document	Patch (0.0.X)
Remove document	Minor (0.X.0)
Schema change	Major (X.0.0)