| name | fallacy-detector |
| description | Use when someone presents a chain of reasoning or a story explaining why something happened, and the logic may have a structural flaw or the narrative may be too tidy. Covers logical fallacies (false dichotomy, post hoc, straw man) and narrative... |
| scenarios | ["Check my reasoning for logical holes","This explanation feels too neat — is it a narrative fallacy?","They said 'if we do X then Y will happen' — is that valid?","이 논리에 오류가 있는지 봐줘","이 설명이 너무 깔끔한 것 같아, 확인해줘","이 주장의 구조가 맞는지 분석해줘"] |
| compatibility | {"recommended":["think-tool"],"optional":["sequential-thinking"],"remote_mcp_note":"think-tool이 있으면 논리적 오류와 서사 오류를 체계적으로 스캔하고 핵심 문제를 정밀하게 짚어낼 수 있습니다. Claude 설정 → MCP Servers에서 remote SSE 엔드포인트를 추가하세요."} |
Fallacy Detector
When to Use / When Not to Use
Use when:
- An argument has a conclusion that needs testing before you act on it
- Someone is explaining why something happened and the story sounds too clean
- A debate or persuasion attempt needs structural analysis
Not for:
- Diagnosing why someone reasons poorly (use bias-auditor)
- Calibrating confidence levels (use epistemic-reasoner)
Process
Step 1 — State the argument or story. Confirm the core claim being analyzed.
Step 2 — Scan for logical fallacies. Flag only what's actually present.
| Category | Fallacies |
|---|
| Structural | False dichotomy, slippery slope, circular reasoning, straw man |
| Authority/Social | Ad hominem, appeal to authority, bandwagon |
| Causation | Post hoc ergo propter hoc, cum hoc, hasty generalization |
Step 3 — Scan for narrative fallacy. Separate scan — check for:
- Hindsight coherence: outcome sounds inevitable in retrospect
- Single-cause attribution: complex event explained by one clean cause
- Character-driven causation: traits driving outcome, ignoring systemic/situational forces
- Omission of counterfactual: similar cases with different outcomes ignored
- Emotional arc: events map too cleanly onto a satisfying story shape
Step 4 — Deliver the analysis. If no fallacies are present, say so.
Output Template
논리적 오류 / Logical Fallacies Detected:
[Fallacy name]: [Where it appears in the argument, specifically]
Why this matters: [What it invalidates or weakens in the conclusion]
서사 오류 / Narrative Fallacy Assessment:
[Present or absent. If present: which element of the story appears constructed rather than accurate]
핵심 문제 / Core Issue:
[The single most damaging error — the one that most undermines the argument's validity]
What Claude Does / What You Do
| Claude | You |
|---|
| Scans systematically across fallacy categories | Provide the argument or explanatory story |
| Names only fallacies that are actually present | Confirm whether the identified pattern applies |
| Explains why each fallacy undermines the conclusion | Decide whether to revise, discard, or defend the argument |
| Runs a separate narrative fallacy scan on explanatory stories | Update your reasoning based on the structural findings |
Key constraint: Naming a fallacy without explaining which step is unjustified and why is taxonomy, not analysis.
Related Skills
bias-auditor — for cognitive biases rather than argument structureepistemic-reasoner — for calibrating confidence to evidenceassumption-extractor — for surfacing hidden premises beneath arguments
