verify-claims

What this skill does

# Fact-Checking Skill

Verify claims and information using professional fact-checking services from around the world.

## Core Principles

1. **Multiple sources** - Cross-reference findings from several fact-checking organizations
2. **Regional relevance** - Prioritize fact-checkers appropriate to the content's context
3. **Language matching** - Use fact-checkers in the native language of the content when possible
4. **Credible sources only** - Never use fraudulent or unreliable fact-checking services
5. **Balanced presentation** - Present both confirming and contradicting findings fairly

---

## When to Use This Skill

Trigger this skill when the user:
- Explicitly asks to fact-check, verify, or validate information
- Shares an article, video transcript, or claim and asks "is this true?"
- Wants to check if something is misinformation or a hoax
- Asks about the credibility of specific claims or statements
- Requests verification of news, social media posts, or viral content
- Wants to cross-reference information with trusted sources

Do NOT trigger for:
- General research or information gathering (use web search instead)
- Checking grammar, spelling, or writing quality
- Verifying code functionality or technical documentation
- Questions about opinions rather than factual claims

---

## Workflow

### Step 1: Understand the Content

Before beginning verification, analyze what needs to be checked:

1. **Identify specific claims** - Extract concrete, verifiable statements from the content
2. **Note the context** - Identify:
   - Geographic references (countries, regions, cities)
   - Named individuals (politicians, public figures, organizations)
   - Languages used in the content
   - Time period or dates mentioned
   - Subject matter (politics, health, science, etc.)
3. **Determine user context**:
   - User's native language (for selecting appropriate fact-checkers)
   - User's location if relevant

**Example Analysis:**
- Content: "Video claiming vaccines cause autism, mentions Andrew Wakefield, references UK study"
- Claims to verify: Vaccine-autism link, Wakefield's research
- Context: Medical/health topic, UK origin, English language
- Key entities: Andrew Wakefield, MMR vaccine, UK medical establishment

### Step 2: Select Fact-Checking Services

**CRITICAL**: Begin by fetching the current list of fact-checking services:

```
Fetch: https://en.wikipedia.org/wiki/List_of_fact-checking_websites
```

From this list, select 3-7 relevant fact-checking services based on:

#### Selection Criteria

1. **User's language/location** - Always include fact-checkers in the user's native language
   
2. **Content language/location** - If different from user's language, also include fact-checkers in the content's language and region

3. **Geographic relevance** - If content mentions specific countries/regions:
   - Include fact-checkers from those countries
   - Example: Content about French politics → include French fact-checkers

4. **Subject matter specialists** - Some fact-checkers specialize:
   - Health/medical claims → Health Feedback, Science Feedback
   - Politics → country-specific political fact-checkers
   - General → Snopes, FactCheck.org, Full Fact

5. **Person-specific** - If content focuses on specific public figures:
   - Include fact-checkers from their home countries
   - Example: Claims about a US politician → include US fact-checkers

#### Exclusion Rule

**NEVER use services listed under "Fraudulent fact-checking websites"** on the Wikipedia page, regardless of how well they match other criteria.

#### Prioritization

When you must limit selections:
- Prioritize: User's language > Content's language > Geographic relevance
- Prefer well-established services (FactCheck.org, Snopes, Full Fact, AFP Fact Check, etc.)
- Include at least one international/general service

**Example Selection:**
- User: Polish speaker
- Content: English article about US vaccines
- Selected services:
  1. Demagog.pl (Polish, for user)
  2. FactCheck.org (US, for content geography)
  3. Snopes (US, general/medical)
  4. Health Feedback (health specialist)
  5. Full Fact (UK, English-speaking, general)

### Step 3: Search Each Fact-Checking Service

For each selected service, conduct targeted searches:

#### Search Strategy

1. **Extract 2-4 search terms** from the content:
   - Key person names
   - Main topics/subjects
   - Specific claims or events
   - Important keywords

2. **Translate terms** to the fact-checker's native language if needed

3. **Construct search queries** using DuckDuckGo with site operator:
   ```
   Format: site:domain.com [search terms in appropriate language]
   
   Examples:
   - site:fullfact.org vaccines autism
   - site:demagog.org.pl szczepionki autyzm
   - site:factcheck.org Andrew Wakefield MMR
   - site:healthfeedback.org vaccine safety
   ```

4. **Execute 1-3 searches per fact-checker** (depending on content complexity)

#### Search Best Practices

- Keep queries concise (2-4 words typically)
- Start broad, then narrow if needed
- Don't repeat very similar queries
- If first search yields good results, proceed to analysis
- If first search yields poor results, try alternative terms

### Step 4: Analyze Search Results

For each fact-checking service:

1. **Review search results** - Examine the first 5-10 results from each search

2. **Select relevant articles** - Choose articles where:
   - Headline directly addresses the claim being verified
   - Content appears substantial (not just brief mentions)
   - Publication date is relevant (recent for ongoing issues, any date for historical debunks)

3. **Fetch and read articles** - Use `web_fetch` to retrieve the full text of 2-4 most relevant articles per fact-checker

4. **Extract key findings** for each article:
   - **Verdict** - What did the fact-checker conclude? (True, False, Misleading, Mixed, Unproven, etc.)
   - **Evidence** - What evidence did they cite?
   - **Context** - Any important nuance or context
   - **Relevance** - How directly does this address the user's claim?

### Step 5: Synthesize and Present Results

Organize findings into a clear, user-friendly format:

#### Handle Fresh Content First

Before presenting results, check if the content is very recent (3 days old or less):

1. **If fact-checks found**: Proceed normally with presentation
2. **If no fact-checks found AND content is ≤3 days old**:
   - Note that the content is too fresh for fact-checkers to have covered it yet
   - **If task scheduling is available**: 
     - Schedule a follow-up fact-check for 3 days from now
     - Inform user: "I've scheduled a follow-up check for [date]. I'll notify you if fact-checkers have published verification by then."
   - **If task scheduling is NOT available**:
     - Suggest: "This content is very recent (published [date]). Fact-checkers typically need a few days to verify claims. I recommend checking back in 3 days for updated verification."
   - Offer preliminary analysis using general web search
   - Proceed with any available information from general sources

3. **If no fact-checks found AND content is older**:
   - Note that fact-checkers haven't specifically covered this
   - Offer general web research instead

#### Structure Your Response

1. **Opening summary** (2-3 sentences)
   - Overall consensus from fact-checkers
   - Brief answer to the user's question

2. **Key findings by claim** (if multiple claims)
   - Group related findings together
   - Present contradicting evidence if it exists

3. **Detailed evidence** (organized by fact-checker or by claim)
   - Include specific verdicts
   - Cite evidence fact-checkers used
   - Note any disagreements between fact-checkers

4. **Important context** (if relevant)
   - Historical background
   - Why the claim persists
   - Common misconceptions

5. **Source citations**
   - Provide direct links to all fact-checking articles referenced
   - Format: `[Fact-Checker Name]: Article Title (Date if available) - [UR