claim-investigation
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseClaim Investigation: Systematic Fact-Checking Skill
声明调查：系统性事实核查Skill
You help systematically investigate claims from social media and other sources, separating verifiable facts from narrative interpretation and identifying what can and cannot be confirmed.
你帮助系统性调查来自社交媒体及其他来源的声明，区分可验证事实与叙事解读，并明确识别可确认与不可确认的内容。
Core Principle
核心原则
Complex claims typically combine verifiable facts with unverifiable interpretations. Effective investigation decomposes claims into atomic components, verifies each independently, and clearly distinguishes between confirmed facts and narrative framing.
复杂声明通常结合可验证事实与不可验证解读。有效的调查会将声明拆解为原子组件，逐一独立验证，并清晰区分已确认事实与叙事框架。
Phase 1: Claim Decomposition
第一阶段：声明拆解
1.1 Extract Atomic Claims
1.1 提取原子声明
Break the statement into individual verifiable claims. Each should be:
- A single factual assertion
- Independently verifiable
- Free of narrative interpretation
Example Decomposition:
Original: "The House Leader refusing to seat the newly-elected AZ-07 special election winner because she'd vote to release the Epstein files"
Atomic claims:
- There is a House Leader (entity exists)
- There was an AZ-07 special election (event occurred)
- Someone won that election (result exists)
- The winner has not been seated (current state)
- A refusal action occurred (specific action claim)
- Causal relationship with Epstein files (causation claim)
将陈述拆解为独立的可验证声明。每个原子声明应满足：
- 单一事实断言
- 可独立验证
- 无叙事解读
拆解示例：
原文："众议院领袖拒绝为新当选的AZ-07特别选举获胜者安排席位，因为她会投票要求公布爱泼斯坦文件"
原子声明：
- 存在众议院领袖（实体存在）
- 举办过AZ-07特别选举（事件发生）
- 有人赢得该选举（结果存在）
- 获胜者未被安排席位（当前状态）
- 发生了拒绝行为（具体行为声明）
- 与爱泼斯坦文件存在因果关系（因果声明）
1.2 Classify Each Component
1.2 对每个组件分类
| Type | Description | Verifiability |
|---|---|---|
| ENTITY | Person, organization, place | Usually verifiable |
| EVENT | Something that allegedly happened | Often verifiable |
| STATE | Current condition or status | Usually verifiable |
| PROCESS | Official procedure or mechanism | Verifiable |
| CAUSATION | Claimed reason or motivation | Rarely verifiable |
| NARRATIVE | Interpretive framing | Not directly verifiable |
| 类型 | 描述 | 可验证性 |
|---|---|---|
| ENTITY | 个人、组织、地点 | 通常可验证 |
| EVENT | 据称发生的事情 | 通常可验证 |
| STATE | 当前状况或状态 | 通常可验证 |
| PROCESS | 官方流程或机制 | 可验证 |
| CAUSATION | 声称的原因或动机 | 极少可验证 |
| NARRATIVE | 解读性框架 | 无法直接验证 |
1.3 Identify Missing Information
1.3 识别缺失信息
Note what's conspicuously absent:
- Unnamed entities ("the winner" instead of a name)
- Unspecified dates
- Missing procedural context
- Absent opposing perspectives
记录明显缺失的内容：
- 未命名实体（如用"获胜者"替代姓名）
- 未指定日期
- 缺失流程背景
- 缺少对立观点
Phase 2: Entity Resolution
第二阶段：实体解析
2.1 Resolve Vague References
2.1 解析模糊指代
Convert vague references to specific, searchable terms:
- "House Leader" → Current House Speaker/Majority Leader name
- "newly-elected winner" → Candidate names from election results
- "Epstein files" → Specific documents/investigations
将模糊指代转换为具体、可搜索的术语：
- "众议院领袖" → 当前众议院议长/多数党领袖姓名
- "新当选的获胜者" → 选举结果中的候选人姓名
- "爱泼斯坦文件" → 特定文件/调查
2.2 Establish Timeline
2.2 建立时间线
For each event:
- When did it allegedly occur?
- What is normal timeline for this type of event?
- Are there procedural deadlines involved?
针对每个事件：
- 据称发生的时间？
- 此类事件的正常时间线是什么？
- 是否涉及流程截止日期？
2.3 Identify Key Actors
2.3 识别关键参与者
- Primary actors (those taking alleged actions)
- Secondary actors (those affected)
- Official bodies with relevant authority
- Potential sources of verification
- 主要参与者（采取据称行为的主体）
- 次要参与者（受影响的主体）
- 拥有相关权限的官方机构
- 潜在的验证来源
Phase 3: Systematic Verification
第三阶段：系统性验证
3.1 Verify Foundational Facts First
3.1 首先验证基础事实
Start with most basic, verifiable claims:
- Did the event occur?
- Do the entities exist?
- Are basic facts correct?
Search Strategy:
- Official sources first (.gov, electoral bodies)
- Cross-reference multiple news sources
- Look for primary documents
从最基础、可验证的声明开始：
- 事件是否发生？
- 实体是否存在？
- 基础事实是否正确？
搜索策略：
- 优先使用官方来源（.gov、选举机构）
- 交叉参考多个新闻来源
- 查找原始文件
3.2 Investigate Procedural Context
3.2 调查流程背景
For any claimed action/inaction:
- What is normal procedure?
- What are requirements?
- What is typical timeline?
- What are legitimate reasons for delays?
针对任何声称的行为/不作为：
- 正常流程是什么？
- 有哪些要求？
- 典型时间线是什么？
- 延迟的合理原因有哪些？
3.3 Examine Causation Claims
3.3 审查因果声明
For any "because" or causal claim:
Direct Evidence:
- Quoted statements from alleged actor
- Official statements or press releases
- Video/audio of relevant statements
Indirect Evidence:
- Other explanations for observed facts
- Standard reasons for similar situations
- Procedural explanations
Context:
- Previous positions by involved parties
- Historical precedents
- Timeline compatibility
针对任何包含"因为"或因果关系的声明：
直接证据：
- 据称行为主体的公开声明
- 官方声明或新闻稿
- 相关声明的视频/音频
间接证据：
- 观察到的事实的其他解释
- 类似情况的标准原因
- 流程层面的解释
背景信息：
- 参与方之前的立场
- 历史先例
- 时间线兼容性
Phase 4: Source Evaluation
第四阶段：来源评估
4.1 Source Priority Order
4.1 来源优先级
- Official government records/databases
- Direct statements from involved parties
- Court documents or legal filings
- Contemporary news reports (multiple outlets)
- Analysis or opinion pieces (noted as such)
- 官方政府记录/数据库
- 参与方的直接声明
- 法庭文件或法律诉状
- 同期新闻报道（多渠道）
- 分析或评论文章（需标注）
4.2 Credibility Markers
4.2 可信度标记
For each source, note:
- Type (official, news, advocacy, social media)
- Date relative to events
- Whether claims are attributed
- Presence of supporting documentation
- Corrections or updates issued
针对每个来源，记录：
- 类型（官方、新闻、倡导组织、社交媒体）
- 与事件的时间相关性
- 声明是否有引用来源
- 是否有支持性文档
- 发布的更正或更新
4.3 Bias Indicators
4.3 偏见指标
Document without dismissing:
- Source's typical political alignment
- Stakeholder relationships
- Pattern of coverage
- Language choices (neutral vs charged)
记录但不否定：
- 来源的典型政治倾向
- 利益相关者关系
- 报道模式
- 语言选择（中立 vs 带情绪）
Phase 5: Narrative Pattern Recognition
第五阶段：叙事模式识别
5.1 Identify Narrative Constructions
5.1 识别叙事构建
Patterns indicating narrative rather than fact:
- Causal chains without evidence ("X because Y because Z")
- Mind-reading claims ("thinks that," "wants to")
- Selective fact inclusion
- Temporal conflation (mixing time periods)
- False dichotomies
表明是叙事而非事实的模式：
- 无证据支撑的因果链（"X因为Y因为Z"）
- 读心式声明（"认为"、"想要"）
- 选择性事实纳入
- 时间混淆（mixing time periods）
- 虚假二分法
5.2 Find Counter-Narratives
5.2 寻找反叙事
For each narrative:
- What facts support it?
- What facts complicate it?
- What alternative narratives explain same facts?
- What facts are excluded?
针对每个叙事：
- 哪些事实支持它？
- 哪些事实使它复杂化？
- 哪些替代叙事可以解释相同事实？
- 哪些事实被排除在外？
5.3 Missing Context
5.3 缺失的背景信息
What would change interpretation:
- Standard procedures being followed
- Similar historical cases
- Full quotes vs partial quotes
- Events immediately before/after
哪些内容会改变解读：
- 正在遵循的标准流程
- 类似的历史案例
- 完整引用 vs 部分引用
- 事件前后的直接相关事件
Phase 6: Synthesis and Reporting
第六阶段：综合与报告
6.1 Report Structure
6.1 报告结构
VERIFIED FACTS:
- [Fact] (Source: [citation])
DISPUTED/UNCLEAR:
- [Claim]:
- Supporting: [source]
- Contradicting: [source]
- Unable to verify: [what's missing]
CONTEXT NEEDED:
- [Procedural context]
- [Historical precedent]
- [Timeline considerations]
NARRATIVE ELEMENTS:
- [Claim]
- Facts that support: [list]
- Facts that complicate: [list]
- Alternative explanations: [list]已核实事实：
- [事实]（来源：[引用]）
有争议/不明确：
- [声明]：
- 支持：[来源]
- 矛盾：[来源]
- 无法验证：[缺失内容]
所需背景：
- [流程背景]
- [历史先例]
- [时间线考量]
叙事元素：
- [声明]
- 支持的事实：[列表]
- 使情况复杂的事实：[列表]
- 替代解释：[列表]6.2 Confidence Levels
6.2 置信度等级
| Level | Meaning |
|---|---|
| Certain | Multiple primary sources confirm |
| Probable | Multiple credible sources align, no contradictions |
| Possible | Some evidence supports, gaps remain |
| Unclear | Contradictory evidence or insufficient info |
| False | Contradicted by authoritative sources |
| 等级 | 含义 |
|---|---|
| 确定 | 多个原始来源确认 |
| 可能 | 多个可信来源一致，无矛盾 |
| 有可能 | 部分证据支持，仍存在缺口 |
| 不明确 | 证据矛盾或信息不足 |
| 错误 | 被权威来源反驳 |
Phase 7: Meta-Analysis
第七阶段：元分析
7.1 Information Gaps
7.1 信息缺口
Document what couldn't be determined:
- Information that should exist but wasn't found
- Questions that remain unanswered
- Time constraints on verification
记录无法确定的内容：
- 应该存在但未找到的信息
- 仍未解答的问题
- 验证的时间限制
7.2 Manipulation Indicators
7.2 操纵指标
Patterns suggesting intentional misrepresentation:
- Key facts consistently omitted
- Misquoted or out-of-context statements
- Conflation of different events/people
- Old events presented as new
表明存在故意误导的模式：
- 关键事实被持续省略
- 引用错误或断章取义的声明
- 混淆不同事件/人物
- 将旧事件呈现为新事件
7.3 Further Investigation
7.3 进一步调查
If initial investigation reveals deeper issues:
- What additional tools/access would help?
- What questions should be asked of officials?
- What documents should be requested?
如果初步调查发现更深层次问题：
- 哪些额外工具/访问权限会有帮助？
- 应该向官员提出哪些问题？
- 应该请求哪些文件？
Search Query Construction
搜索查询构建
- Start broad, then narrow
- Use multiple phrasings for same concept
- Include date ranges when relevant
- Search for both supporting and contradicting evidence
- Use exact phrases for quotes, broad terms for concepts
- 先广泛搜索，再逐步缩小范围
- 对同一概念使用多种表述
- 相关时包含日期范围
- 同时搜索支持和反驳的证据
- 对引用内容使用精确短语，对概念使用宽泛术语
Output Principles
输出原则
- Lead with verified facts
- Clearly separate facts from analysis
- Include all relevant context
- Present multiple valid interpretations where applicable
- Never assert causation without evidence
- Acknowledge investigation limitations
- 以已核实事实开头
- 明确区分事实与分析
- 包含所有相关背景
- 适用时呈现多种有效解读
- 无证据时绝不断言因果关系
- 承认调查的局限性
Output Persistence
输出持久化
Output Discovery
输出查找
- Check for in the project
context/output-config.md - If found, look for this skill's entry
- If not found, ask user: "Where should I save investigation reports?"
- Suggest: or
research/investigations/explorations/research/
- 检查项目中的
context/output-config.md - 如果找到，查找本Skill的条目
- 如果未找到，询问用户："调查报告应保存至何处？"
- 建议路径：或
research/investigations/explorations/research/
Primary Output
主要输出
- Decomposed claims - Atomic components with classifications
- Verification results - Confidence levels per component
- Context documentation - Procedural and historical context
- Synthesis report - Using standard report structure
- 拆解后的声明 - 带有分类的原子组件
- 验证结果 - 每个组件的置信度等级
- 背景文档 - 流程和历史背景
- 综合报告 - 使用标准报告结构
File Naming
文件命名
Pattern:
{topic}-investigation-{date}.md模式：
{topic}-investigation-{date}.mdVerification (Oracle)
验证（Oracle）
What This Skill Can Verify
本Skill可验证的内容
- Decomposition complete - All atomic claims identified? (High confidence)
- Entity resolution - Vague references resolved? (High confidence)
- Source evaluation - Credibility markers documented? (High confidence)
- 拆解完整性 - 是否识别了所有原子声明？（高置信度）
- 实体解析 - 模糊指代是否已解析？（高置信度）
- 来源评估 - 是否记录了可信度标记？（高置信度）
What Requires Human Judgment
需要人工判断的内容
- Source reliability - Contextual trust assessment
- Narrative interpretation - Which framing is most accurate?
- Manipulation detection - Intent behind information gaps
- 来源可靠性 - 上下文信任评估
- 叙事解读 - 哪种框架最准确？
- 操纵检测 - 信息缺口背后的意图
Oracle Limitations
Oracle局限性
- Cannot assess motivations behind claims
- Cannot predict how information will evolve
- 无法评估声明背后的动机
- 无法预测信息的演变
Feedback Loop
反馈循环
Session Persistence
会话持久化
- Output location: See
context/output-config.md - What to save: Decomposition, verification, context, synthesis
- Naming pattern:
{topic}-investigation-{date}.md
- 输出位置：参见
context/output-config.md - 需保存内容：拆解、验证、背景、综合报告
- 命名模式：
{topic}-investigation-{date}.md
Cross-Session Learning
跨会话学习
- Check for prior investigations on related topics
- Build on previous source evaluations
- Failed verifications inform methodology
- 检查相关主题的先前调查
- 基于之前的来源评估进行构建
- 验证失败案例为方法提供参考
Design Constraints
设计约束
This Skill Assumes
本Skill假设
- A specific claim to investigate (not general research)
- Verifiable components exist within the claim
- Sources are accessible for verification
- 存在特定的待调查声明（非一般性研究）
- 声明中存在可验证的组件
- 可访问用于验证的来源
This Skill Does Not Handle
本Skill不处理的内容
- General research - Route to: research
- AI output verification - Route to: fact-check
- Media pattern analysis - Route to: media-meta-analysis
- 一般性研究 - 转至：research
- AI输出验证 - 转至：fact-check
- 媒体模式分析 - 转至：media-meta-analysis
Degradation Signals
性能退化信号
- Single-source verification (confirmation rush)
- Accepting causation without evidence
- Dismissing entire claims for single errors
- 单来源验证（急于确认）
- 无证据时接受因果关系
- 因单个错误而否定整个声明
Reasoning Requirements
推理要求
Standard Reasoning
标准推理
- Single claim decomposition
- Basic entity resolution
- Simple source evaluation
- 单一声明拆解
- 基础实体解析
- 简单来源评估
Extended Reasoning (ultrathink)
扩展推理（ultrathink）
- Multi-claim investigation - [Why: claims interact and context builds]
- Narrative analysis - [Why: detecting manipulation patterns]
- Deep source tracing - [Why: finding original sources through citation chains]
Trigger phrases: "full investigation", "trace all sources", "analyze the narrative"
- 多声明调查 - [原因：声明相互影响，背景逐步构建]
- 叙事分析 - [原因：检测操纵模式]
- 深度来源追踪 - [原因：通过引用链找到原始来源]
触发短语："全面调查"、"追踪所有来源"、"分析叙事"
Execution Strategy
执行策略
Sequential (Default)
顺序执行（默认）
- Decomposition before verification
- Foundational facts before causation claims
- Individual components before synthesis
- 先拆解再验证
- 先基础事实再因果声明
- 先单个组件再综合
Parallelizable
可并行执行
- Verifying independent atomic claims
- Researching multiple sources simultaneously
- 验证独立的原子声明
- 同时研究多个来源
Subagent Candidates
子Agent候选
| Task | Agent Type | When to Spawn |
|---|---|---|
| Source research | general-purpose | When tracing claim origins |
| Timeline construction | general-purpose | When mapping event sequences |
| 任务 | Agent类型 | 触发时机 |
|---|---|---|
| 来源研究 | 通用型 | 追踪声明起源时 |
| 时间线构建 | 通用型 | 绘制事件序列时 |
Context Management
上下文管理
Approximate Token Footprint
大致Token占用
- Skill base: ~3.5k tokens (phases + templates)
- With examples: ~4.5k tokens
- With full output structure: ~5k tokens
- Skill基础部分：约3.5k tokens（阶段+模板）
- 含示例：约4.5k tokens
- 含完整输出结构：约5k tokens
Context Optimization
上下文优化
- Focus on current investigation phase
- Report structure is reference, not in-context
- Examples optional
- 聚焦当前调查阶段
- 报告结构仅作参考，不纳入上下文
- 示例为可选内容
When Context Gets Tight
上下文紧张时的处理
- Prioritize: Current phase, active claims
- Defer: Full template structure, all phases
- Drop: Meta-analysis section, search examples
- 优先保留：当前阶段、活跃声明
- 延迟处理：完整模板结构、所有阶段
- 移除：元分析部分、搜索示例
Anti-Patterns
反模式
1. Confirmation Rush
1. 急于确认
Pattern: Finding one source that matches the claim and declaring it verified.
Why it fails: Single-source verification misses errors, biases, and coordinated misinformation where multiple outlets repeat the same false claim without independent verification.
Fix: Require at least 2-3 independent sources. Trace claims back to primary sources. Check if "multiple sources" are actually just repeating the same original source.
模式：找到一个匹配声明的来源后即宣布已验证。
失败原因：单来源验证会遗漏错误、偏见，以及多渠道重复同一虚假声明而无独立验证的协同误导。
修复方案：要求至少2-3个独立来源。将声明追溯至原始来源。检查"多来源"是否实际上只是重复同一原始来源。
2. Causation Collapse
2. 因果坍塌
Pattern: Accepting "X happened because Y" claims when only "X happened" and "Y exists" are verified.
Why it fails: Correlation proves co-occurrence, not causation. Human pattern-matching fills in causal links that may not exist. Political narratives especially exploit this gap.
Fix: Demand direct evidence for causation (stated intent, documented decisions). When causation can't be verified, report it as "alleged motivation" or "claimed reason."
模式：仅验证了"X发生"和"Y存在"，就接受"X发生是因为Y"的声明。
失败原因：相关性仅证明同时发生，而非因果关系。人类的模式匹配会填补可能不存在的因果联系。政治叙事尤其会利用这一漏洞。
修复方案：要求因果关系的直接证据（明确的意图陈述、有记录的决策）。当因果关系无法验证时，将其报告为"声称的动机"或"声称的原因"。
3. Premature Debunking
3. 过早驳斥
Pattern: Finding one fact wrong and dismissing the entire claim without investigating other components.
Why it fails: Complex claims often mix true and false elements. Dismissing everything because one part is wrong misses real issues embedded in the narrative.
Fix: Decompose fully, verify each component independently. Report accuracy per-component: "Claims A and C are verified; claim B is false; claim D is unverifiable."
模式：发现一个事实错误后，未调查其他组件就否定整个声明。
失败原因：复杂声明通常混合真实与虚假元素。因一部分错误而否定全部内容会遗漏叙事中隐含的真实问题。
修复方案：完整拆解声明，逐一独立验证每个组件。按组件报告准确性："声明A和C已验证；声明B为虚假；声明D无法验证。"
4. Authority Fallacy
4. 权威谬误
Pattern: Accepting official sources uncritically because they're "authoritative."
Why it fails: Official sources can be wrong, incomplete, outdated, or deliberately misleading. Authority reduces probability of error but doesn't eliminate it.
Fix: Cross-reference official sources with other evidence. Note when official sources have incentives to misrepresent. Distinguish between "official position" and "verified fact."
模式：因为来源"权威"就不加批判地接受官方来源。
失败原因：官方来源可能存在错误、不完整、过时或故意误导。权威性降低了错误概率，但无法消除错误。
修复方案：将官方来源与其他证据交叉参考。记录官方来源是否有误导的动机。区分"官方立场"与"已验证事实"。
5. Narrative Anchoring
5. 叙事锚定
Pattern: Starting with a hypothesis about what's "really happening" and investigating to prove it.
Why it fails: Confirmation bias shapes what evidence you seek and how you interpret it. You'll find "evidence" for any narrative if you look hard enough.
Fix: Start with the specific claims made. Investigate each on its own terms. Actively seek disconfirming evidence. Document alternative explanations that fit the same facts.
模式：从"实际发生了什么"的假设开始，然后调查以证明该假设。
失败原因：确认偏差会影响你寻找的证据和解读方式。只要刻意寻找，你总能为任何叙事找到"证据"。
修复方案：从提出的具体声明开始。逐一独立调查每个声明。主动寻找反驳证据。记录符合相同事实的替代解释。
Integration
集成
Inbound (feeds into this skill)
入站（供本Skill使用）
| Skill | What it provides |
|---|---|
| research | Initial source discovery and query expansion |
| media-meta-analysis | Understanding of source biases and media patterns |
| Skill | 提供内容 |
|---|---|
| research | 初始来源发现和查询扩展 |
| media-meta-analysis | 来源偏见和媒体模式的理解 |
Outbound (this skill enables)
出站（本Skill为其提供支持）
| Skill | What this provides |
|---|---|
| fact-check | Verified facts for post-generation checking |
| sensitivity-check | Context for evaluating representation claims |
| Skill | 提供内容 |
|---|---|
| fact-check | 用于生成后核查的已验证事实 |
| sensitivity-check | 评估代表性声明的背景信息 |
Complementary
互补关系
| Skill | Relationship |
|---|---|
| research | Use research for broad information gathering, claim-investigation for specific claim verification |
| fact-check | Use claim-investigation for external claims, fact-check for AI-generated content verification |
| Skill | 关系 |
|---|---|
| research | 用research进行广泛信息收集，用claim-investigation进行特定声明验证 |
| fact-check | 用claim-investigation验证外部声明，用fact-check验证AI生成内容 |