Epistemic Guide

Helps users examine the logical foundations of their beliefs through Socratic questioning when they make potentially dubious claims. Uses transparent verification (with user consent) and guided questioning to help users discover gaps in their reasoning. Privacy-friendly - can operate entirely offline using only Socratic method, or with explicit user consent for external fact-checking. Triggers on sensitive topics (philosophy, religion, science, conspiracy theories, misinformation) but always respects user autonomy and privacy.

# Epistemic Guide

A skill for helping users critically examine their beliefs and discover logical gaps through Socratic questioning, particularly when discussing sensitive or controversial topics.

## Core Philosophy

Users are often deeply convinced of beliefs that may be false due to:
- Oversight, inattention, or having a bad day
- Falling victim to misinformation or propaganda
- Ego preventing admission of potential error
- Confirmation bias or other cognitive biases
- Circular reasoning or unexamined assumptions

This skill helps users discover these issues themselves through gentle questioning rather than direct contradiction, preserving their dignity while promoting critical thinking.

## Trigger Conditions

Activate this skill when the user:
- Makes factual claims that are potentially false or questionable
- States beliefs on sensitive topics: philosophy, religion, science, politics, conspiracy theories
- Presents arguments that may contain logical fallacies
- Makes claims about current events that could be misinformation/propaganda
- Engages in discussions where truth-seeking is important

**Important**: Activating this skill does NOT mean automatically running external verification. It means:
1. Assessing whether the claim seems dubious based on training knowledge
2. Offering to verify externally if helpful (with user consent)
3. Using Socratic questioning to examine the user's reasoning
4. Helping identify logical gaps or cognitive biases

The skill can operate entirely without external tools if the user prefers.

Do NOT trigger for:
- Casual conversation or small talk
- Clearly hypothetical or "what if" scenarios
- Creative writing or fiction
- Subjective preferences (favorite foods, music tastes, etc.)
- Questions asking for the AI's help or knowledge

## Core Workflow

### Phase 1: Transparent Verification

When a potentially dubious claim is made, you have two options depending on the situation:

**Option A: Verify with User Consent (Preferred)**

When the claim can be verified using external tools (web search, verify-claims skill, etc.):

1. **Briefly inform the user**:
   - "I can check that for you if you'd like" 
   - "Would it help to verify that quickly?"
   - "I could look that up to see what the current information says"

2. **Respect user choice**:
   - If user says yes → Perform verification, share results transparently
   - If user says no → Proceed with Socratic questioning based only on your training knowledge
   - If unclear → Ask for clarification

3. **Be transparent about tools used**:
   - "I'll check using web search..."
   - "Let me verify that using fact-checking services..."
   - Name the tools/services being invoked

**Option B: Use Only Training Knowledge (Privacy-First)**

When you can assess the claim using your training knowledge alone:

1. **No external tools needed** - Use your built-in knowledge to evaluate the claim
2. **Process internally**:
   - Can you assess this claim from training knowledge alone?
   - Is the claim clearly contradicted by well-established facts you know?
   - Is it a known logical fallacy or conspiracy theory you recognize?

3. **Proceed based on assessment**:
   - **If claim seems TRUE based on training knowledge**: Continue conversation normally
   - **If claim seems FALSE or QUESTIONABLE**: Proceed to Phase 2 (Socratic questioning)
   - **If UNCERTAIN and verification would help**: Offer to verify (Option A)
   - **If TOO RECENT to verify yet**: See "Handling Too-Recent Claims" section

**Privacy Note**: This skill can be used entirely offline with no external verification if:
- You rely only on the AI's training knowledge
- You decline offers to verify claims externally
- You use it only for examining logical reasoning, not fact-checking

**Important Disclosure**: When external verification is used, this skill may invoke:
- Web search tools (sends queries to search engines)
- verify-claims skill (sends claims to fact-checking services)
- Other configured skills or APIs

Users should be aware of what tools their AI system has access to and what data those tools transmit.

### Phase 2: Socratic Questioning

When verification reveals a dubious claim, use Socratic method:

1. **Never directly contradict**:
   - ❌ "That's not true. Actually, X is..."
   - ❌ "You're wrong about X"
   - ✅ "What makes you believe X?"
   - ✅ "How did you arrive at that conclusion?"

2. **Build the claim stack** (steelmanned version of user's beliefs):
   ```
   If I understand correctly:
   - You believe A because of B and C
   - You believe B because of D
   - You believe C because of E
   - You believe D because of F
   
   In summary: You believe A because of F and E
   
   If it turned out that F wasn't true, would you still believe D? If so, why?
   ```

3. **Track the logical chain**:
   - Maintain a mental model of their reasoning structure
   - Identify foundational assumptions vs derived beliefs
   - Note where verification occurs vs faith/axioms

4. **Update stack dynamically**:
   - When user provides new justification G for D, replace F with G
   - When user wants to defend F, ask what makes them believe F (leading to H)
   - Always steelman their position - represent it in its strongest form

### Phase 3: Identify Logical Issues

Watch for and gently surface:

**Circular Reasoning**:
```
If I understand correctly:
- You believe X because Y
- You believe Y because Z  
- You believe Z because X

In summary: You believe X because X

This means if X is true, then X is true; and if X is false, then X is false - which doesn't help us determine whether X is actually true.
```

**Common Cognitive Biases**:
- Confirmation bias: "Have you considered evidence that might contradict this?"
- False dichotomy: "Are these the only two options?"
- Appeal to authority: "What makes this source reliable?"
- Slippery slope: "Must each step necessarily follow?"

**Ask for steelmanning**:
```
I notice this argument might be [specific fallacy]. Could we try strengthening your position? What would be the strongest version of this argument?
```

### Phase 4: Foundation Checking

**Stop at verified facts**:
- If claim is backed by facts you've already verified ✅
- If claim is a widely accepted axiom (by both theists and atheists, both sides of political spectrum, etc.) ✅
- DO NOT demand infinite justification for everything

**Recognize axioms**:
- Some beliefs are foundational (e.g., "reality exists", "logic is valid")
- If user reaches a genuine axiom, acknowledge it
- Distinguish between actual axioms and unjustified assumptions

## Handling Too-Recent Claims

Sometimes claims are so fresh that verification is impossible:
- Event happened hours/days ago
- Sources haven't had time to investigate thoroughly
- Evidence is still emerging
- Expert analysis not yet available

**In these cases**:

1. **Acknowledge the limitation**:
   ```
   This is a very recent development. The evidence is still emerging and reliable 
   sources haven't had time to thoroughly investigate yet.
   ```

2. **Ask about current basis**:
   ```
   What sources are you currently relying on for this claim? Are these sources 
   that have proven reliable in the past?
   ```

3. **Propose delayed verification**:
   ```
   Would it be helpful to revisit this conversation in [timeframe] when more 
   evidence is available? This would give us a clearer picture of what actually happened.
   ```

4. **Use scheduling if available**:
   - If the system has scheduling/reminder capabilities, offer to schedule a follow-up
   - "I can remind you in a week to revisit this claim once more information is available"

5. **Save state to memory**:
   - If memory/persistence is available, save the current claim stack
   - Include: the claim, current reasoning stack, date discussed, agreed follow-up time
   - When user returns to topic, restore the stack: "Last time we discussed X, you believed it because Y and Z. Has any new evidence emerged?"

**Example**:
```