{"id":"a85fa149-3331-444d-a5c4-c9f22a05f7e1","entityType":"agent","slug":"clawhub-deciqai-logical-fallacies","name":"Logical Fallacies","canonicalUrl":"https://www.xpersona.co/agent/clawhub-deciqai-logical-fallacies","canonicalPath":"/agent/clawhub-deciqai-logical-fallacies","generatedAt":"2026-10-11T16:01:10.104Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T13:20:10.514Z","emptyReason":null},"description":"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning... Skill: Logical Fallacies Owner: deciqai Summary: Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning... Tags: latest:1.0.5 Version history: v1.0.5 | 2026-07-16T18:05:10.879Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/logical-fallacies.json) v1.0.4 | 2026-07-10T10:26:53.800","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.1K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s17a4mqcnk515kvaca5ze55d0x88pfpx:logical-fallacies","sourceUrl":"https://clawhub.ai/deciqai/logical-fallacies","homepage":"https://clawhub.ai/deciqai/skills/logical-fallacies","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/deciqai/logical-fallacies","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/deciqai/skills/logical-fallacies","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning..."},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T13:20:10.514Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T13:20:10.514Z","emptyReason":null},"stars":null,"forks":null,"downloads":1058,"packageName":null,"latestVersion":"1.0.5","tractionLabel":"1.1K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T13:20:10.436Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T13:20:10.514Z","lastCrawledAt":"2026-10-11T13:20:10.436Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T13:20:10.436Z","lastVerifiedAt":null,"highlights":[{"version":"1.0.5","createdAt":"2026-07-16T18:05:10.879Z","changelog":"Description tail link + agents machine-readable metadata line (deciqai.com/s/logical-fallacies.json)","fileCount":6,"zipByteSize":15151},{"version":"1.0.4","createdAt":"2026-07-10T10:26:53.800Z","changelog":"Add 2024-2026 AI-era worked example + updated sources","fileCount":6,"zipByteSize":14892},{"version":"1.0.3","createdAt":"2026-07-08T11:08:12.814Z","changelog":"Footer now uses /c/<slug> short link (fixes UTM truncation when SKILL.md is read in a terminal)","fileCount":5,"zipByteSize":9675},{"version":"1.0.2","createdAt":"2026-07-08T00:52:59.849Z","changelog":"Refreshed content + GitHub star link in footer","fileCount":5,"zipByteSize":9559},{"version":"1.0.1","createdAt":"2026-07-07T22:28:13.788Z","changelog":"Add catalog categories and topics","fileCount":5,"zipByteSize":9645},{"version":"1.0.0","createdAt":"2026-06-29T11:17:42.214Z","changelog":"Initial publish","fileCount":5,"zipByteSize":9391}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17a4mqcnk515kvaca5ze55d0x88pfpx:logical-fallacies","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T16:01:10.100Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-deciqai-logical-fallacies/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-11T13:20:10.514Z","emptyReason":null},"readme":"Skill: Logical Fallacies\n\nOwner: deciqai\n\nSummary: Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning...\n\nTags: latest:1.0.5\n\nVersion history:\n\nv1.0.5 | 2026-07-16T18:05:10.879Z | user\n\nDescription tail link + agents machine-readable metadata line (deciqai.com/s/logical-fallacies.json)\n\nv1.0.4 | 2026-07-10T10:26:53.800Z | user\n\nAdd 2024-2026 AI-era worked example + updated sources\n\nv1.0.3 | 2026-07-08T11:08:12.814Z | user\n\nFooter now uses /c/<slug> short link (fixes UTM truncation when SKILL.md is read in a terminal)\n\nv1.0.2 | 2026-07-08T00:52:59.849Z | user\n\nRefreshed content + GitHub star link in footer\n\nv1.0.1 | 2026-07-07T22:28:13.788Z | user\n\nAdd catalog categories and topics\n\nv1.0.0 | 2026-06-29T11:17:42.214Z | user\n\nInitial publish\n\nArchive index:\n\nArchive v1.0.5: 6 files, 15151 bytes\n\nFiles: examples/ai-hype-discourse-fallacies-2024-2026.md (10010b), examples/tversky-kahnemans-linda-problem-1983.md (4966b), references/sources.md (3043b), skill-card.md (3327b), SKILL.md (9120b), _meta.json (136b)\n\nFile v1.0.5:SKILL.md\n\n---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low. More: deciqai.com/c/logical-fallacies\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: `critical-thinking` audits evidence quality and framing; `first-principles` attacks premises; `mece` catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n- You're weighing an AI hype or AI-adoption claim (\"a lab CEO said it's near,\" \"it passed the benchmark so it's intelligent,\" \"doom vs. utopia\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not advance until user responds]**\n\n## The Process\n\nRun the **Fallacy Audit** in four passes — structure, language, cognition, rhetoric — then judge.\n\n1. **State the argument cleanly.** Rewrite as premises → conclusion. If you cannot, first finding: it's an assertion dressed as an argument.\n2. **Structural pass.** Begging the question, affirming the consequent, denying the antecedent, false cause (*post hoc*), hasty generalization, accident, complex question, ignoratio elenchi.\n3. **Linguistic pass.** Equivocation (key term shifts meaning), amphiboly, composition/division (part↔whole), accent/figure of speech.\n4. **Cognitive pass.** Conjunction fallacy (P(A∧B) > P(A) from representativeness), base-rate neglect, availability, anchoring — see `anchoring`.\n5. **Rhetorical-trap pass.** Ad hominem, appeal to authority (exception: expert in own domain), appeal to popularity (weak prior only), appeal to emotion (evidence vs. substitute), false dichotomy, straw man, slippery slope, tu quoque.\n6. **Judge the argument, not the moves.** A fallacy means the inference fails — not that the conclusion is false.\n7. **Fallacy-fallacy check.** If you can't articulate *why this instance fails*, you have a dismissal, not a finding.\n8. **Output:** for each fallacy — (a) which and where, (b) what the inference fails to establish, (c) what would actually support the conclusion.\n\n### Output: the Fallacy Audit\n\n```\nArgument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>\n```\n\n*→ Method in Action: [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md)*\n*→ 2026 lens: [Four Fallacies in the AI Debate (2024–2026)](examples/ai-hype-discourse-fallacies-2024-2026.md)*\n\n## Pack: Common Fallacy Patterns by Domain\n\n- **Startup/business:** survivorship bias (hasty generalization on filtered data), appeal to authority (\"Sequoia said X\"), anecdote as evidence, false dichotomy (\"raise now or die\")\n- **Policy/public debate:** undocumented slippery slope, straw man, ad hominem, appeal to consequences\n- **Internal arguments (most expensive):** conjunction fallacy, confirmation bias, sunk cost, optimism bias — see `expected-value-and-kelly` and `probabilistic-thinking`\n- **Statistical/data:** base-rate neglect, Texas sharpshooter (target drawn after the fact), cherry-picking, regression to mean confused with causation\n\n## Applying It Well\n\nRun the filter on your *own* arguments first. The ones that survive are worth more than the ones you flag in opponents. Fallacy-detection at speed is suspicious — it usually means surface pattern-matching, not an actual audit. Slow is good.\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I named the fallacy, so I refuted the argument\" | A fallacious argument doesn't make the conclusion false — it fails to *establish* it. Refuting an argument ≠ refuting a claim. |\n| [D] \"Citing an expert is appeal to authority\" | Expert testimony in the expert's actual domain is legitimate evidence. The fallacy is treating a citation as *proof* or citing outside their domain. |\n| [D] \"Any analogy is a false analogy\" | Reasoning by analogy is often valid. The fallacy is asserting similarity on the relevant dimensions without showing it. |\n| [D] Spotting fallacies only in arguments you disagree with | If the filter has a personal valence, it is broken. Run it on your own most-loved arguments. |\n| [D] \"No formal fallacy → the argument is sound\" | Formal fallacies are a small fraction. Most modern errors are informal. Run all four passes. |\n| [D] \"Knowing the conjunction fallacy means I won't commit it\" | Tversky-Kahneman 1983 showed otherwise. Only the explicit check prevents the error. |\n| [D] \"I caught the fallacy quickly, so I'm good at this\" | Speed signals surface pattern-matching, not an actual audit. Slow is good. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Fallacy label with no articulation of which inference step fails and why\n- Audit found fallacies only in arguments the auditor already disagreed with\n- Only the structural pass ran; linguistic, cognitive, or rhetorical skipped\n- No distinction between \"the argument fails\" and \"the conclusion is false\"\n- Output is a list of labels with no \"what would actually support this claim\" section\n\n## Verification\n\n- [ ] Argument rewritten as premises → conclusion (or flagged as not an argument)\n- [ ] All four passes run: structural, linguistic, cognitive, rhetorical\n- [ ] Each fallacy specifies exact location and exact failure\n- [ ] Fallacy-fallacy check performed — no findings are dismissals dressed as fallacies\n- [ ] Verdict separates \"argument fails\" from \"conclusion is false\"\n- [ ] Output names what evidence would actually support the claim\n- [ ] At least one cognitive fallacy (modern empirical category) considered\n- [ ] Filter applied to arguments the auditor agrees with, not only opponents'\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/logical-fallacies** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\n*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/logical-fallacies.json*\n\nFile v1.0.5:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.5\",\n  \"publishedAt\": 1784225110879\n}\n\nFile v1.0.5:references/sources.md\n\n# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n- Stanford Institute for Human-Centered AI (HAI), *AI Index Report* (annual, 2024 edition and later). Documents the measurement problems behind AI capability claims — benchmark saturation and data contamination (test items leaking into training data), which is what makes \"it passed the benchmark, therefore it's intelligent\" an instance of affirming the consequent rather than a valid inference. https://aiindex.stanford.edu/report/\n- Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). \"On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?\" *Proceedings of FAccT '21*, 610–623. A durable reference for the equivocation-on-\"intelligence\" and hasty-generalization patterns in AI discourse — arguing that fluent output can be mistaken for understanding. https://doi.org/10.1145/3442188.3445922\n\nFile v1.0.5:examples/ai-hype-discourse-fallacies-2024-2026.md\n\n# Method in Action: Four Fallacies in the 2024–2026 AI Debate\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nThe Linda problem shows the fallacy filter under laboratory conditions. This example runs the same **Fallacy Audit** on messy public discourse — the argument about artificial intelligence as it circulated across roughly 2024–2025. This is the environment the skill was built for: claims that *sound* authoritative, move fast, and carry high stakes, where naming the fallacy is only useful if you can also say what would actually settle the question.\n\nWe audit four representative arguments, each a real pattern that recurred across essays, interviews, and social threads in this period. No individual quote is reconstructed verbatim below; each argument is stated as a clean paraphrase of a widely-circulated *pattern* of reasoning, then run through the process.\n\n---\n\n## Step 1 — State each argument cleanly (premises → conclusion)\n\n**Argument A (authority).**\n- P1: The CEO of a leading AI lab says transformative AI is a few years away.\n- P2: They run the lab and see the frontier models first.\n- C: Therefore transformative AI is a few years away.\n\n**Argument B (false dilemma).**\n- P1: AI leads either to catastrophe (\"doom\") or to radical abundance (\"utopia\").\n- P2: The doom scenario is implausible / the utopia scenario is implausible (whichever the speaker rejects).\n- C: Therefore the other outcome is what we should expect.\n\n**Argument C (hasty generalization).**\n- P1: A single demo went viral showing a model doing task X impressively.\n- C: Therefore models can now do X (and tasks like X) reliably.\n\n**Argument D (affirming the consequent).**\n- P1: If a system is intelligent, it will pass benchmark B.\n- P2: This system passed benchmark B.\n- C: Therefore this system is intelligent.\n\nAll four are genuine arguments (premises and a conclusion), so none collapses at Step 1 into \"an assertion dressed as an argument.\" Good — that means the work is in the passes.\n\n## Step 2 — Structural pass\n\n**Argument D is a textbook formal fallacy: affirming the consequent.** The form is \"If P then Q; Q; therefore P.\" That is invalid: Q can be true for reasons unrelated to P. Passing benchmark B is consistent with intelligence *and* with narrow pattern-matching, benchmark contamination (test items leaking into training data), or overfitting to the benchmark's format. The inference fails because P1 only licenses the reverse direction (intelligent → passes B), not (passes B → intelligent). Benchmark saturation across this period — models scoring very high on tests that older models failed — is exactly the observation that makes the invalid direction tempting.\n\n**Argument C is hasty generalization.** One vivid, curated, possibly cherry-picked instance is generalized to reliable capability across a class of tasks. A viral demo is a maximally-filtered sample: the impressive run is the one that got posted. The inference from \"did X once, on camera\" to \"does X reliably\" ignores variance, failure rate, and selection.\n\n## Step 3 — Linguistic pass\n\n**Argument D also hides an equivocation** on the word \"intelligent.\" P1 uses \"intelligent\" in a rich sense (general, flexible, understanding). The conclusion inherits that rich sense — but all the evidence established was a benchmark score, which at most supports \"intelligent\" in a thin, task-specific sense. The key term shifts meaning between premise and conclusion. This is why benchmark-driven \"it's intelligent now\" claims feel stronger than they are: the word does double duty.\n\n**Argument B smuggles a definitional move** on \"AI\" and on \"doom/utopia\" — the terms are left vague enough that any outcome can be sorted into one bucket, which is what makes the dilemma feel exhaustive.\n\n## Step 4 — Cognitive pass\n\n**Argument C is powered by the availability heuristic** (Tversky & Kahneman, 1974): a dramatic, emotionally salient demo is easy to recall and therefore feels representative of typical performance. The vivid case crowds out the invisible base rate of failures that were never posted.\n\n**Argument A leans on base-rate neglect.** Even if lab leaders are somewhat better calibrated than outsiders, the base rate of confident near-term timeline predictions in the history of AI that did *not* come true is high. A forecast should update on that base rate, not just on the forecaster's seat.\n\n## Step 5 — Rhetorical-trap pass\n\n**Argument A is appeal to authority (*ad verecundiam*).** Note the disciplined version of this finding: a lab CEO *is* a domain expert, and their testimony is legitimate *evidence* about, say, what a model can do today. The fallacy is not \"they're an executive so ignore them.\" The fallacy is (a) treating a prediction as *proof*, and (b) the conflict-of-interest overhang — the same person is raising capital, recruiting, and setting expectations, so their timeline is also a business artifact. Expertise about present capability does not transfer into authority over a multi-year forecast.\n\n**Argument B is a false dilemma (false dichotomy).** \"Doom or utopia\" presents two extremes as the only options, when the outcome space plainly includes a wide middle: uneven diffusion, sector-by-sector disruption, muddling through, partial gains with real harms. Once one horn is knocked down, the argument rushes to the other — but the disjunction was never exhaustive, so knocking down one horn establishes nothing about the other.\n\n## Step 6 — Judge the argument, not the moves\n\nEach argument *fails to establish its conclusion*. That is the verdict — and it is **not** the claim that the conclusions are false:\n\n- Transformative AI genuinely *might* be near (A) — the CEO's prediction just doesn't prove it.\n- One of doom/utopia *could* occur (B) — the dilemma just doesn't force it.\n- Models genuinely *may* do X reliably (C) — one demo just doesn't show it.\n- The system *may* be intelligent in some sense (D) — the benchmark just doesn't entail it.\n\n## Step 7 — Fallacy-fallacy check\n\nThe temptation here is strong and worth naming: it is easy to shout \"appeal to authority!\" and treat the CEO as rebutted, or \"affirming the consequent!\" and treat the model as *proven* unintelligent. Both would be the fallacy fallacy — using a flaw in the argument to assert the negation of the conclusion. Every finding above specifies *which inference step* fails and *why*, not merely a label. That is the line between an audit and a dismissal.\n\n## Step 8 — Output: the Fallacy Audit\n\n```\nArgument A: CEO predicts near-term transformative AI / they see the frontier / therefore it's near\n  Structural: —\n  Linguistic: —\n  Cognitive: base-rate neglect (history of failed AI timelines)\n  Rhetorical: appeal to authority — expert on present capability ≠ authority on multi-year forecast; conflict of interest\n  Verdict: fails to establish | still open — needs track record of calibrated forecasts, not a single seat\n  Repair: aggregate multiple independent forecasters; score them against past calibration; treat timelines as probability distributions\n\nArgument B: AI → doom or utopia / one is implausible / therefore the other\n  Structural: —\n  Linguistic: vague terms make the dilemma feel exhaustive\n  Cognitive: —\n  Rhetorical: false dichotomy — ignores the large middle of outcomes\n  Verdict: fails to establish | still open — enumerate the full outcome space, assign probabilities to each\n  Repair: replace binary with a distribution over scenarios; argue each on its own evidence\n\nArgument C: one viral demo of task X / therefore reliable at X\n  Structural: hasty generalization from a filtered sample of one\n  Linguistic: —\n  Cognitive: availability heuristic — vivid demo feels typical\n  Rhetorical: —\n  Verdict: fails to establish | still open — needs pass rate over many unfiltered trials\n  Repair: measure success rate on a held-out, contamination-checked test set; report variance and failure modes\n\nArgument D: if intelligent then passes B / passed B / therefore intelligent\n  Structural: affirming the consequent (invalid form)\n  Linguistic: equivocation on \"intelligent\" (rich sense vs. thin task sense)\n  Cognitive: —\n  Rhetorical: —\n  Verdict: fails to establish | still open — define \"intelligent\" operationally, test generalization beyond B, rule out benchmark contamination\n  Repair: use held-out tasks the model wasn't optimized for; check for train/test leakage; specify which capability the benchmark actually measures\n\nFallacy-fallacy check: no conclusion is asserted false — each is left open pending real evidence.\n```\n\n---\n\nWhy this is the right modern companion to the Linda problem: Tversky-Kahneman showed the fallacies live inside individual cognition even under lab control. The AI debate shows the same fallacies scaled up to a public argument moving at social-media speed, wrapped in real expertise and real stakes — precisely the conditions (\"speed is the sophist's friend,\" \"supported entirely by authority\") the skill's *When to Use* section flags. The audit does not tell you whether transformative AI is near. It tells you that none of these four popular arguments has earned that conclusion yet — and exactly what evidence would.\n\n*Sources: Tversky, A., & Kahneman, D. (1974), \"Judgment under Uncertainty: Heuristics and Biases,\" Science 185(4157), 1124–1131, https://doi.org/10.1126/science.185.4157.1124 (availability heuristic; base-rate neglect). Aristotle, Sophistical Refutations, c. 350 BCE — affirming the consequent, equivocation, false cause, hasty generalization, https://classics.mit.edu/Aristotle/sophist_refut.html . Stanford Encyclopedia of Philosophy, \"Fallacies,\" https://plato.stanford.edu/entries/fallacies/ (formal vs. informal fallacies; false dilemma; ad verecundiam). On benchmark contamination and saturation as a documented measurement problem in this period: Stanford HAI, AI Index Report 2024, https://aiindex.stanford.edu/report/ .*\n\nFile v1.0.5:examples/tversky-kahnemans-linda-problem-1983.md\n\n# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky-Kahneman showed that for the modern cognitive fallacies, the *default* state of human reasoning is to commit them.\n\n**Second**, the fallacy survived **knowing the rule**. Subjects who could state the conjunction axiom in the abstract still violated it in the concrete. This is the deep finding: knowing what a fallacy is does not, by itself, prevent committing one. Only a **procedure** — a check applied at the point of judgment — does. That is exactly what this skill is.\n\n**Third**, the fallacy was **invisible from the inside**. Subjects rated T∧F more probable than T and did not experience cognitive dissonance. The wrong answer *felt right*. This is the deepest reason the fallacy filter has to be external — a checklist, a written audit, a colleague asking — rather than introspective. Your mind cannot reliably catch its own conjunction fallacies. It can be trained to *suspect* them in patterns (\"is this story making a more specific claim feel more likely?\") and route to the explicit check.\n\n**Fourth**, the case shows the **right and wrong use of fallacy-naming**. The right use: identify the move, repair the inference, judge the underlying claim on its actual evidence (what *is* the probability Linda is a bank teller? you'd need actual base rates). The wrong use: announce \"conjunction fallacy!\" and consider Linda's profile rebutted. The fallacy invalidates the *intuition*; it does not establish that Linda *isn't* a feminist bank teller.\n\nThe arc from Aristotle's 350 BCE taxonomy to Tversky-Kahneman's 1983 experiment is the arc this skill teaches. Aristotle gave us the **language** for the moves; Tversky-Kahneman gave us the **evidence** that the moves are not optional features of bad arguers but mandatory features of human cognition. Both are required. The taxonomy without the empirical map underestimates the threat. The empirical map without the taxonomy has no vocabulary to name what just happened.\n\nFile v1.0.5:skill-card.md\n\n## Description:\n\nGuides an agent through a structured fallacy audit that separates premises from conclusions, checks structural, linguistic, cognitive, and rhetorical reasoning errors, and explains what evidence would repair the argument.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[deciqai](https://clawhub.ai/user/deciqai)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nExternal users and developers use this skill to help an agent audit persuasive arguments for reasoning errors, including authority, emotion, popularity, analogy, cognitive bias, and AI-hype claims. It is most useful when the conclusion is not directly settled by checking data and the user needs a careful explanation of which inference fails.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The skill can make an agent more formal and critical in discussions where a lightweight response would be more appropriate.\n\nMitigation: Use the skill when the argument's reasoning quality matters, and avoid activating it for casual low-stakes conversation.\n\nRisk: Fallacy labels can be misused as dismissals rather than as explanations of which inference fails.\n\nMitigation: Require each finding to identify the argument location, the failed inference, and what evidence would actually support the conclusion.\n\nRisk: Private or sensitive examples could be copied into the skill file during customization.\n\nMitigation: Keep private examples out of the skill artifact unless the publisher intentionally approves including them.\n\n## Reference(s):\n\n- [ClawHub Skill Page](https://clawhub.ai/deciqai/skills/logical-fallacies)\n- [Logical Fallacies Sources](references/sources.md)\n- [Tversky and Kahneman Linda Problem Example](examples/tversky-kahnemans-linda-problem-1983.md)\n- [AI Hype Discourse Fallacies Example](examples/ai-hype-discourse-fallacies-2024-2026.md)\n- [Aristotle Sophistical Refutations](https://classics.mit.edu/Aristotle/sophist_refut.html)\n- [Stanford Encyclopedia of Philosophy: Aristotle's Logic](https://plato.stanford.edu/entries/aristotle-logic/)\n- [Tversky and Kahneman 1983 Conjunction Fallacy Paper](https://doi.org/10.1037/0033-295X.90.4.293)\n- [Tversky and Kahneman 1974 Heuristics and Biases Paper](https://doi.org/10.1126/science.185.4157.1124)\n- [Stanford Encyclopedia of Philosophy: Fallacies](https://plato.stanford.edu/entries/fallacies/)\n- [Stanford HAI AI Index Report](https://aiindex.stanford.edu/report/)\n- [Bender et al. 2021 Stochastic Parrots](https://doi.org/10.1145/3442188.3445922)\n- [deciqAI Logical Fallacies Metadata](https://www.deciqai.com/s/logical-fallacies.json)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, analysis, guidance]\n\n**Output Format:** [Markdown fallacy audit with premises, findings, verdict, and repair guidance]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [Markdown-only reasoning guide; no executable code, install hooks, credential use, or tool calls.]\n\n## Skill Version(s):\n\n1.0.5 (source: server-resolved release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v1.0.4: 6 files, 14892 bytes\n\nFiles: examples/ai-hype-discourse-fallacies-2024-2026.md (10010b), examples/tversky-kahnemans-linda-problem-1983.md (4966b), references/sources.md (3043b), skill-card.md (3053b), SKILL.md (8975b), _meta.json (136b)\n\nFile v1.0.4:SKILL.md\n\n---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low.\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: `critical-thinking` audits evidence quality and framing; `first-principles` attacks premises; `mece` catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n- You're weighing an AI hype or AI-adoption claim (\"a lab CEO said it's near,\" \"it passed the benchmark so it's intelligent,\" \"doom vs. utopia\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not advance until user responds]**\n\n## The Process\n\nRun the **Fallacy Audit** in four passes — structure, language, cognition, rhetoric — then judge.\n\n1. **State the argument cleanly.** Rewrite as premises → conclusion. If you cannot, first finding: it's an assertion dressed as an argument.\n2. **Structural pass.** Begging the question, affirming the consequent, denying the antecedent, false cause (*post hoc*), hasty generalization, accident, complex question, ignoratio elenchi.\n3. **Linguistic pass.** Equivocation (key term shifts meaning), amphiboly, composition/division (part↔whole), accent/figure of speech.\n4. **Cognitive pass.** Conjunction fallacy (P(A∧B) > P(A) from representativeness), base-rate neglect, availability, anchoring — see `anchoring`.\n5. **Rhetorical-trap pass.** Ad hominem, appeal to authority (exception: expert in own domain), appeal to popularity (weak prior only), appeal to emotion (evidence vs. substitute), false dichotomy, straw man, slippery slope, tu quoque.\n6. **Judge the argument, not the moves.** A fallacy means the inference fails — not that the conclusion is false.\n7. **Fallacy-fallacy check.** If you can't articulate *why this instance fails*, you have a dismissal, not a finding.\n8. **Output:** for each fallacy — (a) which and where, (b) what the inference fails to establish, (c) what would actually support the conclusion.\n\n### Output: the Fallacy Audit\n\n```\nArgument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>\n```\n\n*→ Method in Action: [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md)*\n*→ 2026 lens: [Four Fallacies in the AI Debate (2024–2026)](examples/ai-hype-discourse-fallacies-2024-2026.md)*\n\n## Pack: Common Fallacy Patterns by Domain\n\n- **Startup/business:** survivorship bias (hasty generalization on filtered data), appeal to authority (\"Sequoia said X\"), anecdote as evidence, false dichotomy (\"raise now or die\")\n- **Policy/public debate:** undocumented slippery slope, straw man, ad hominem, appeal to consequences\n- **Internal arguments (most expensive):** conjunction fallacy, confirmation bias, sunk cost, optimism bias — see `expected-value-and-kelly` and `probabilistic-thinking`\n- **Statistical/data:** base-rate neglect, Texas sharpshooter (target drawn after the fact), cherry-picking, regression to mean confused with causation\n\n## Applying It Well\n\nRun the filter on your *own* arguments first. The ones that survive are worth more than the ones you flag in opponents. Fallacy-detection at speed is suspicious — it usually means surface pattern-matching, not an actual audit. Slow is good.\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I named the fallacy, so I refuted the argument\" | A fallacious argument doesn't make the conclusion false — it fails to *establish* it. Refuting an argument ≠ refuting a claim. |\n| [D] \"Citing an expert is appeal to authority\" | Expert testimony in the expert's actual domain is legitimate evidence. The fallacy is treating a citation as *proof* or citing outside their domain. |\n| [D] \"Any analogy is a false analogy\" | Reasoning by analogy is often valid. The fallacy is asserting similarity on the relevant dimensions without showing it. |\n| [D] Spotting fallacies only in arguments you disagree with | If the filter has a personal valence, it is broken. Run it on your own most-loved arguments. |\n| [D] \"No formal fallacy → the argument is sound\" | Formal fallacies are a small fraction. Most modern errors are informal. Run all four passes. |\n| [D] \"Knowing the conjunction fallacy means I won't commit it\" | Tversky-Kahneman 1983 showed otherwise. Only the explicit check prevents the error. |\n| [D] \"I caught the fallacy quickly, so I'm good at this\" | Speed signals surface pattern-matching, not an actual audit. Slow is good. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Fallacy label with no articulation of which inference step fails and why\n- Audit found fallacies only in arguments the auditor already disagreed with\n- Only the structural pass ran; linguistic, cognitive, or rhetorical skipped\n- No distinction between \"the argument fails\" and \"the conclusion is false\"\n- Output is a list of labels with no \"what would actually support this claim\" section\n\n## Verification\n\n- [ ] Argument rewritten as premises → conclusion (or flagged as not an argument)\n- [ ] All four passes run: structural, linguistic, cognitive, rhetorical\n- [ ] Each fallacy specifies exact location and exact failure\n- [ ] Fallacy-fallacy check performed — no findings are dismissals dressed as fallacies\n- [ ] Verdict separates \"argument fails\" from \"conclusion is false\"\n- [ ] Output names what evidence would actually support the claim\n- [ ] At least one cognitive fallacy (modern empirical category) considered\n- [ ] Filter applied to arguments the auditor agrees with, not only opponents'\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 223 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/logical-fallacies** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\nFile v1.0.4:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.4\",\n  \"publishedAt\": 1783679213800\n}\n\nFile v1.0.4:references/sources.md\n\n# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n- Stanford Institute for Human-Centered AI (HAI), *AI Index Report* (annual, 2024 edition and later). Documents the measurement problems behind AI capability claims — benchmark saturation and data contamination (test items leaking into training data), which is what makes \"it passed the benchmark, therefore it's intelligent\" an instance of affirming the consequent rather than a valid inference. https://aiindex.stanford.edu/report/\n- Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). \"On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?\" *Proceedings of FAccT '21*, 610–623. A durable reference for the equivocation-on-\"intelligence\" and hasty-generalization patterns in AI discourse — arguing that fluent output can be mistaken for understanding. https://doi.org/10.1145/3442188.3445922\n\nFile v1.0.4:examples/ai-hype-discourse-fallacies-2024-2026.md\n\n# Method in Action: Four Fallacies in the 2024–2026 AI Debate\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nThe Linda problem shows the fallacy filter under laboratory conditions. This example runs the same **Fallacy Audit** on messy public discourse — the argument about artificial intelligence as it circulated across roughly 2024–2025. This is the environment the skill was built for: claims that *sound* authoritative, move fast, and carry high stakes, where naming the fallacy is only useful if you can also say what would actually settle the question.\n\nWe audit four representative arguments, each a real pattern that recurred across essays, interviews, and social threads in this period. No individual quote is reconstructed verbatim below; each argument is stated as a clean paraphrase of a widely-circulated *pattern* of reasoning, then run through the process.\n\n---\n\n## Step 1 — State each argument cleanly (premises → conclusion)\n\n**Argument A (authority).**\n- P1: The CEO of a leading AI lab says transformative AI is a few years away.\n- P2: They run the lab and see the frontier models first.\n- C: Therefore transformative AI is a few years away.\n\n**Argument B (false dilemma).**\n- P1: AI leads either to catastrophe (\"doom\") or to radical abundance (\"utopia\").\n- P2: The doom scenario is implausible / the utopia scenario is implausible (whichever the speaker rejects).\n- C: Therefore the other outcome is what we should expect.\n\n**Argument C (hasty generalization).**\n- P1: A single demo went viral showing a model doing task X impressively.\n- C: Therefore models can now do X (and tasks like X) reliably.\n\n**Argument D (affirming the consequent).**\n- P1: If a system is intelligent, it will pass benchmark B.\n- P2: This system passed benchmark B.\n- C: Therefore this system is intelligent.\n\nAll four are genuine arguments (premises and a conclusion), so none collapses at Step 1 into \"an assertion dressed as an argument.\" Good — that means the work is in the passes.\n\n## Step 2 — Structural pass\n\n**Argument D is a textbook formal fallacy: affirming the consequent.** The form is \"If P then Q; Q; therefore P.\" That is invalid: Q can be true for reasons unrelated to P. Passing benchmark B is consistent with intelligence *and* with narrow pattern-matching, benchmark contamination (test items leaking into training data), or overfitting to the benchmark's format. The inference fails because P1 only licenses the reverse direction (intelligent → passes B), not (passes B → intelligent). Benchmark saturation across this period — models scoring very high on tests that older models failed — is exactly the observation that makes the invalid direction tempting.\n\n**Argument C is hasty generalization.** One vivid, curated, possibly cherry-picked instance is generalized to reliable capability across a class of tasks. A viral demo is a maximally-filtered sample: the impressive run is the one that got posted. The inference from \"did X once, on camera\" to \"does X reliably\" ignores variance, failure rate, and selection.\n\n## Step 3 — Linguistic pass\n\n**Argument D also hides an equivocation** on the word \"intelligent.\" P1 uses \"intelligent\" in a rich sense (general, flexible, understanding). The conclusion inherits that rich sense — but all the evidence established was a benchmark score, which at most supports \"intelligent\" in a thin, task-specific sense. The key term shifts meaning between premise and conclusion. This is why benchmark-driven \"it's intelligent now\" claims feel stronger than they are: the word does double duty.\n\n**Argument B smuggles a definitional move** on \"AI\" and on \"doom/utopia\" — the terms are left vague enough that any outcome can be sorted into one bucket, which is what makes the dilemma feel exhaustive.\n\n## Step 4 — Cognitive pass\n\n**Argument C is powered by the availability heuristic** (Tversky & Kahneman, 1974): a dramatic, emotionally salient demo is easy to recall and therefore feels representative of typical performance. The vivid case crowds out the invisible base rate of failures that were never posted.\n\n**Argument A leans on base-rate neglect.** Even if lab leaders are somewhat better calibrated than outsiders, the base rate of confident near-term timeline predictions in the history of AI that did *not* come true is high. A forecast should update on that base rate, not just on the forecaster's seat.\n\n## Step 5 — Rhetorical-trap pass\n\n**Argument A is appeal to authority (*ad verecundiam*).** Note the disciplined version of this finding: a lab CEO *is* a domain expert, and their testimony is legitimate *evidence* about, say, what a model can do today. The fallacy is not \"they're an executive so ignore them.\" The fallacy is (a) treating a prediction as *proof*, and (b) the conflict-of-interest overhang — the same person is raising capital, recruiting, and setting expectations, so their timeline is also a business artifact. Expertise about present capability does not transfer into authority over a multi-year forecast.\n\n**Argument B is a false dilemma (false dichotomy).** \"Doom or utopia\" presents two extremes as the only options, when the outcome space plainly includes a wide middle: uneven diffusion, sector-by-sector disruption, muddling through, partial gains with real harms. Once one horn is knocked down, the argument rushes to the other — but the disjunction was never exhaustive, so knocking down one horn establishes nothing about the other.\n\n## Step 6 — Judge the argument, not the moves\n\nEach argument *fails to establish its conclusion*. That is the verdict — and it is **not** the claim that the conclusions are false:\n\n- Transformative AI genuinely *might* be near (A) — the CEO's prediction just doesn't prove it.\n- One of doom/utopia *could* occur (B) — the dilemma just doesn't force it.\n- Models genuinely *may* do X reliably (C) — one demo just doesn't show it.\n- The system *may* be intelligent in some sense (D) — the benchmark just doesn't entail it.\n\n## Step 7 — Fallacy-fallacy check\n\nThe temptation here is strong and worth naming: it is easy to shout \"appeal to authority!\" and treat the CEO as rebutted, or \"affirming the consequent!\" and treat the model as *proven* unintelligent. Both would be the fallacy fallacy — using a flaw in the argument to assert the negation of the conclusion. Every finding above specifies *which inference step* fails and *why*, not merely a label. That is the line between an audit and a dismissal.\n\n## Step 8 — Output: the Fallacy Audit\n\n```\nArgument A: CEO predicts near-term transformative AI / they see the frontier / therefore it's near\n  Structural: —\n  Linguistic: —\n  Cognitive: base-rate neglect (history of failed AI timelines)\n  Rhetorical: appeal to authority — expert on present capability ≠ authority on multi-year forecast; conflict of interest\n  Verdict: fails to establish | still open — needs track record of calibrated forecasts, not a single seat\n  Repair: aggregate multiple independent forecasters; score them against past calibration; treat timelines as probability distributions\n\nArgument B: AI → doom or utopia / one is implausible / therefore the other\n  Structural: —\n  Linguistic: vague terms make the dilemma feel exhaustive\n  Cognitive: —\n  Rhetorical: false dichotomy — ignores the large middle of outcomes\n  Verdict: fails to establish | still open — enumerate the full outcome space, assign probabilities to each\n  Repair: replace binary with a distribution over scenarios; argue each on its own evidence\n\nArgument C: one viral demo of task X / therefore reliable at X\n  Structural: hasty generalization from a filtered sample of one\n  Linguistic: —\n  Cognitive: availability heuristic — vivid demo feels typical\n  Rhetorical: —\n  Verdict: fails to establish | still open — needs pass rate over many unfiltered trials\n  Repair: measure success rate on a held-out, contamination-checked test set; report variance and failure modes\n\nArgument D: if intelligent then passes B / passed B / therefore intelligent\n  Structural: affirming the consequent (invalid form)\n  Linguistic: equivocation on \"intelligent\" (rich sense vs. thin task sense)\n  Cognitive: —\n  Rhetorical: —\n  Verdict: fails to establish | still open — define \"intelligent\" operationally, test generalization beyond B, rule out benchmark contamination\n  Repair: use held-out tasks the model wasn't optimized for; check for train/test leakage; specify which capability the benchmark actually measures\n\nFallacy-fallacy check: no conclusion is asserted false — each is left open pending real evidence.\n```\n\n---\n\nWhy this is the right modern companion to the Linda problem: Tversky-Kahneman showed the fallacies live inside individual cognition even under lab control. The AI debate shows the same fallacies scaled up to a public argument moving at social-media speed, wrapped in real expertise and real stakes — precisely the conditions (\"speed is the sophist's friend,\" \"supported entirely by authority\") the skill's *When to Use* section flags. The audit does not tell you whether transformative AI is near. It tells you that none of these four popular arguments has earned that conclusion yet — and exactly what evidence would.\n\n*Sources: Tversky, A., & Kahneman, D. (1974), \"Judgment under Uncertainty: Heuristics and Biases,\" Science 185(4157), 1124–1131, https://doi.org/10.1126/science.185.4157.1124 (availability heuristic; base-rate neglect). Aristotle, Sophistical Refutations, c. 350 BCE — affirming the consequent, equivocation, false cause, hasty generalization, https://classics.mit.edu/Aristotle/sophist_refut.html . Stanford Encyclopedia of Philosophy, \"Fallacies,\" https://plato.stanford.edu/entries/fallacies/ (formal vs. informal fallacies; false dilemma; ad verecundiam). On benchmark contamination and saturation as a documented measurement problem in this period: Stanford HAI, AI Index Report 2024, https://aiindex.stanford.edu/report/ .*\n\nFile v1.0.4:examples/tversky-kahnemans-linda-problem-1983.md\n\n# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky-Kahneman showed that for the modern cognitive fallacies, the *default* state of human reasoning is to commit them.\n\n**Second**, the fallacy survived **knowing the rule**. Subjects who could state the conjunction axiom in the abstract still violated it in the concrete. This is the deep finding: knowing what a fallacy is does not, by itself, prevent committing one. Only a **procedure** — a check applied at the point of judgment — does. That is exactly what this skill is.\n\n**Third**, the fallacy was **invisible from the inside**. Subjects rated T∧F more probable than T and did not experience cognitive dissonance. The wrong answer *felt right*. This is the deepest reason the fallacy filter has to be external — a checklist, a written audit, a colleague asking — rather than introspective. Your mind cannot reliably catch its own conjunction fallacies. It can be trained to *suspect* them in patterns (\"is this story making a more specific claim feel more likely?\") and route to the explicit check.\n\n**Fourth**, the case shows the **right and wrong use of fallacy-naming**. The right use: identify the move, repair the inference, judge the underlying claim on its actual evidence (what *is* the probability Linda is a bank teller? you'd need actual base rates). The wrong use: announce \"conjunction fallacy!\" and consider Linda's profile rebutted. The fallacy invalidates the *intuition*; it does not establish that Linda *isn't* a feminist bank teller.\n\nThe arc from Aristotle's 350 BCE taxonomy to Tversky-Kahneman's 1983 experiment is the arc this skill teaches. Aristotle gave us the **language** for the moves; Tversky-Kahneman gave us the **evidence** that the moves are not optional features of bad arguers but mandatory features of human cognition. Both are required. The taxonomy without the empirical map underestimates the threat. The empirical map without the taxonomy has no vocabulary to name what just happened.\n\nFile v1.0.4:skill-card.md\n\n## Description: <br>\nHelps an agent audit persuasive arguments for structural, linguistic, cognitive, and rhetorical fallacies before judging what the argument actually establishes. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nEmployees, external reviewers, and agents use this skill to turn a concrete argument into premises and conclusion, audit it across formal, linguistic, cognitive, and rhetorical passes, and state what evidence would repair or settle the claim. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill can make an agent more critical of arguments, including AI-related claims, and may overstate uncertainty if used where the conclusion can be checked directly. <br>\nMitigation: Use it for inference quality, not fact lookup; verify empirical claims against data before relying on the audit. <br>\nRisk: Fallacy labels can become dismissals when the agent does not identify the exact failed inference. <br>\nMitigation: Require the audit to restate premises and conclusion, identify where the inference fails, and name what evidence would support the claim. <br>\n\n\n## Reference(s): <br>\n- [Sources - logical-fallacies](references/sources.md) <br>\n- [Tversky and Kahneman Linda Problem example](examples/tversky-kahnemans-linda-problem-1983.md) <br>\n- [AI discourse fallacies example](examples/ai-hype-discourse-fallacies-2024-2026.md) <br>\n- [Aristotle, Sophistical Refutations](https://classics.mit.edu/Aristotle/sophist_refut.html) <br>\n- [Stanford Encyclopedia of Philosophy, Aristotle's Logic](https://plato.stanford.edu/entries/aristotle-logic/) <br>\n- [Tversky and Kahneman 1983, Extensional versus intuitive reasoning](https://doi.org/10.1037/0033-295X.90.4.293) <br>\n- [Tversky and Kahneman 1974, Judgment under Uncertainty](https://doi.org/10.1126/science.185.4157.1124) <br>\n- [Stanford Encyclopedia of Philosophy, Fallacies](https://plato.stanford.edu/entries/fallacies/) <br>\n- [Stanford HAI AI Index Report](https://aiindex.stanford.edu/report/) <br>\n- [Bender et al. 2021, On the Dangers of Stochastic Parrots](https://doi.org/10.1145/3442188.3445922) <br>\n- [ClawHub skill page](https://clawhub.ai/deciqai/skills/logical-fallacies) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Text, Markdown, Guidance] <br>\n**Output Format:** [Markdown structured as a fallacy audit] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Produces argument restatements, findings by audit pass, a fallacy-fallacy check, verdict, and repair guidance.] <br>\n\n## Skill Version(s): <br>\n1.0.4 (source: release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.3: 5 files, 9675 bytes\n\nFiles: examples/tversky-kahnemans-linda-problem-1983.md (4966b), references/sources.md (2208b), skill-card.md (2813b), SKILL.md (8714b), _meta.json (136b)\n\nFile v1.0.3:SKILL.md\n\n---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low.\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: `critical-thinking` audits evidence quality and framing; `first-principles` attacks premises; `mece` catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not advance until user responds]**\n\n## The Process\n\nRun the **Fallacy Audit** in four passes — structure, language, cognition, rhetoric — then judge.\n\n1. **State the argument cleanly.** Rewrite as premises → conclusion. If you cannot, first finding: it's an assertion dressed as an argument.\n2. **Structural pass.** Begging the question, affirming the consequent, denying the antecedent, false cause (*post hoc*), hasty generalization, accident, complex question, ignoratio elenchi.\n3. **Linguistic pass.** Equivocation (key term shifts meaning), amphiboly, composition/division (part↔whole), accent/figure of speech.\n4. **Cognitive pass.** Conjunction fallacy (P(A∧B) > P(A) from representativeness), base-rate neglect, availability, anchoring — see `anchoring`.\n5. **Rhetorical-trap pass.** Ad hominem, appeal to authority (exception: expert in own domain), appeal to popularity (weak prior only), appeal to emotion (evidence vs. substitute), false dichotomy, straw man, slippery slope, tu quoque.\n6. **Judge the argument, not the moves.** A fallacy means the inference fails — not that the conclusion is false.\n7. **Fallacy-fallacy check.** If you can't articulate *why this instance fails*, you have a dismissal, not a finding.\n8. **Output:** for each fallacy — (a) which and where, (b) what the inference fails to establish, (c) what would actually support the conclusion.\n\n### Output: the Fallacy Audit\n\n```\nArgument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>\n```\n\n*→ Method in Action: [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md)*\n\n## Pack: Common Fallacy Patterns by Domain\n\n- **Startup/business:** survivorship bias (hasty generalization on filtered data), appeal to authority (\"Sequoia said X\"), anecdote as evidence, false dichotomy (\"raise now or die\")\n- **Policy/public debate:** undocumented slippery slope, straw man, ad hominem, appeal to consequences\n- **Internal arguments (most expensive):** conjunction fallacy, confirmation bias, sunk cost, optimism bias — see `expected-value-and-kelly` and `probabilistic-thinking`\n- **Statistical/data:** base-rate neglect, Texas sharpshooter (target drawn after the fact), cherry-picking, regression to mean confused with causation\n\n## Applying It Well\n\nRun the filter on your *own* arguments first. The ones that survive are worth more than the ones you flag in opponents. Fallacy-detection at speed is suspicious — it usually means surface pattern-matching, not an actual audit. Slow is good.\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I named the fallacy, so I refuted the argument\" | A fallacious argument doesn't make the conclusion false — it fails to *establish* it. Refuting an argument ≠ refuting a claim. |\n| [D] \"Citing an expert is appeal to authority\" | Expert testimony in the expert's actual domain is legitimate evidence. The fallacy is treating a citation as *proof* or citing outside their domain. |\n| [D] \"Any analogy is a false analogy\" | Reasoning by analogy is often valid. The fallacy is asserting similarity on the relevant dimensions without showing it. |\n| [D] Spotting fallacies only in arguments you disagree with | If the filter has a personal valence, it is broken. Run it on your own most-loved arguments. |\n| [D] \"No formal fallacy → the argument is sound\" | Formal fallacies are a small fraction. Most modern errors are informal. Run all four passes. |\n| [D] \"Knowing the conjunction fallacy means I won't commit it\" | Tversky-Kahneman 1983 showed otherwise. Only the explicit check prevents the error. |\n| [D] \"I caught the fallacy quickly, so I'm good at this\" | Speed signals surface pattern-matching, not an actual audit. Slow is good. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Fallacy label with no articulation of which inference step fails and why\n- Audit found fallacies only in arguments the auditor already disagreed with\n- Only the structural pass ran; linguistic, cognitive, or rhetorical skipped\n- No distinction between \"the argument fails\" and \"the conclusion is false\"\n- Output is a list of labels with no \"what would actually support this claim\" section\n\n## Verification\n\n- [ ] Argument rewritten as premises → conclusion (or flagged as not an argument)\n- [ ] All four passes run: structural, linguistic, cognitive, rhetorical\n- [ ] Each fallacy specifies exact location and exact failure\n- [ ] Fallacy-fallacy check performed — no findings are dismissals dressed as fallacies\n- [ ] Verdict separates \"argument fails\" from \"conclusion is false\"\n- [ ] Output names what evidence would actually support the claim\n- [ ] At least one cognitive fallacy (modern empirical category) considered\n- [ ] Filter applied to arguments the auditor agrees with, not only opponents'\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 164 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/logical-fallacies** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\nFile v1.0.3:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.3\",\n  \"publishedAt\": 1783508892814\n}\n\nFile v1.0.3:references/sources.md\n\n# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n\nFile v1.0.3:examples/tversky-kahnemans-linda-problem-1983.md\n\n# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky-Kahneman showed that for the modern cognitive fallacies, the *default* state of human reasoning is to commit them.\n\n**Second**, the fallacy survived **knowing the rule**. Subjects who could state the conjunction axiom in the abstract still violated it in the concrete. This is the deep finding: knowing what a fallacy is does not, by itself, prevent committing one. Only a **procedure** — a check applied at the point of judgment — does. That is exactly what this skill is.\n\n**Third**, the fallacy was **invisible from the inside**. Subjects rated T∧F more probable than T and did not experience cognitive dissonance. The wrong answer *felt right*. This is the deepest reason the fallacy filter has to be external — a checklist, a written audit, a colleague asking — rather than introspective. Your mind cannot reliably catch its own conjunction fallacies. It can be trained to *suspect* them in patterns (\"is this story making a more specific claim feel more likely?\") and route to the explicit check.\n\n**Fourth**, the case shows the **right and wrong use of fallacy-naming**. The right use: identify the move, repair the inference, judge the underlying claim on its actual evidence (what *is* the probability Linda is a bank teller? you'd need actual base rates). The wrong use: announce \"conjunction fallacy!\" and consider Linda's profile rebutted. The fallacy invalidates the *intuition*; it does not establish that Linda *isn't* a feminist bank teller.\n\nThe arc from Aristotle's 350 BCE taxonomy to Tversky-Kahneman's 1983 experiment is the arc this skill teaches. Aristotle gave us the **language** for the moves; Tversky-Kahneman gave us the **evidence** that the moves are not optional features of bad arguers but mandatory features of human cognition. Both are required. The taxonomy without the empirical map underestimates the threat. The empirical map without the taxonomy has no vocabulary to name what just happened.\n\nFile v1.0.3:skill-card.md\n\n## Description: <br>\nGuides an agent through a structured fallacy audit that rewrites arguments as premises and conclusions, checks structural, linguistic, cognitive, and rhetorical failure modes, and separates weak inference from false conclusions. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers, analysts, and other external users use this skill to evaluate persuasive arguments, identify the exact inference step that fails, and determine what evidence would actually support the claim. It can also coach novice users step by step when they do not yet have a concrete argument. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: Users may paste private or sensitive real-world arguments into shared notes while recording observed examples. <br>\nMitigation: Avoid including confidential, personal, or proprietary details in shared audit notes; redact examples before storing or sharing them. <br>\nRisk: Fallacy labels can be used as dismissive rhetoric instead of a real analysis of the inference. <br>\nMitigation: Require each finding to identify the exact inference that fails, explain why it fails, and state what evidence would support the conclusion. <br>\n\n\n## Reference(s): <br>\n- [Sources - logical-fallacies](references/sources.md) <br>\n- [Tversky and Kahneman Linda Problem Example](examples/tversky-kahnemans-linda-problem-1983.md) <br>\n- [Aristotle, Sophistical Refutations](https://classics.mit.edu/Aristotle/sophist_refut.html) <br>\n- [Stanford Encyclopedia of Philosophy: Aristotle's Logic](https://plato.stanford.edu/entries/aristotle-logic/) <br>\n- [Tversky and Kahneman (1983), Extensional versus Intuitive Reasoning](https://doi.org/10.1037/0033-295X.90.4.293) <br>\n- [Tversky and Kahneman (1974), Judgment under Uncertainty](https://doi.org/10.1126/science.185.4157.1124) <br>\n- [Stanford Encyclopedia of Philosophy: Fallacies](https://plato.stanford.edu/entries/fallacies/) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Analysis, Markdown, Guidance] <br>\n**Output Format:** [Markdown fallacy audit with premises, findings, verdict, and repair guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Text-only reasoning output; no code execution, external tools, or API calls are required.] <br>\n\n## Skill Version(s): <br>\n1.0.3 (source: server release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.2: 5 files, 9559 bytes\n\nFiles: examples/tversky-kahnemans-linda-problem-1983.md (4966b), references/sources.md (2208b), skill-card.md (2494b), SKILL.md (8821b), _meta.json (136b)\n\nFile v1.0.2:SKILL.md\n\n---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low.\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: `critical-thinking` audits evidence quality and framing; `first-principles` attacks premises; `mece` catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not advance until user responds]**\n\n## The Process\n\nRun the **Fallacy Audit** in four passes — structure, language, cognition, rhetoric — then judge.\n\n1. **State the argument cleanly.** Rewrite as premises → conclusion. If you cannot, first finding: it's an assertion dressed as an argument.\n2. **Structural pass.** Begging the question, affirming the consequent, denying the antecedent, false cause (*post hoc*), hasty generalization, accident, complex question, ignoratio elenchi.\n3. **Linguistic pass.** Equivocation (key term shifts meaning), amphiboly, composition/division (part↔whole), accent/figure of speech.\n4. **Cognitive pass.** Conjunction fallacy (P(A∧B) > P(A) from representativeness), base-rate neglect, availability, anchoring — see `anchoring`.\n5. **Rhetorical-trap pass.** Ad hominem, appeal to authority (exception: expert in own domain), appeal to popularity (weak prior only), appeal to emotion (evidence vs. substitute), false dichotomy, straw man, slippery slope, tu quoque.\n6. **Judge the argument, not the moves.** A fallacy means the inference fails — not that the conclusion is false.\n7. **Fallacy-fallacy check.** If you can't articulate *why this instance fails*, you have a dismissal, not a finding.\n8. **Output:** for each fallacy — (a) which and where, (b) what the inference fails to establish, (c) what would actually support the conclusion.\n\n### Output: the Fallacy Audit\n\n```\nArgument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>\n```\n\n*→ Method in Action: [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md)*\n\n## Pack: Common Fallacy Patterns by Domain\n\n- **Startup/business:** survivorship bias (hasty generalization on filtered data), appeal to authority (\"Sequoia said X\"), anecdote as evidence, false dichotomy (\"raise now or die\")\n- **Policy/public debate:** undocumented slippery slope, straw man, ad hominem, appeal to consequences\n- **Internal arguments (most expensive):** conjunction fallacy, confirmation bias, sunk cost, optimism bias — see `expected-value-and-kelly` and `probabilistic-thinking`\n- **Statistical/data:** base-rate neglect, Texas sharpshooter (target drawn after the fact), cherry-picking, regression to mean confused with causation\n\n## Applying It Well\n\nRun the filter on your *own* arguments first. The ones that survive are worth more than the ones you flag in opponents. Fallacy-detection at speed is suspicious — it usually means surface pattern-matching, not an actual audit. Slow is good.\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I named the fallacy, so I refuted the argument\" | A fallacious argument doesn't make the conclusion false — it fails to *establish* it. Refuting an argument ≠ refuting a claim. |\n| [D] \"Citing an expert is appeal to authority\" | Expert testimony in the expert's actual domain is legitimate evidence. The fallacy is treating a citation as *proof* or citing outside their domain. |\n| [D] \"Any analogy is a false analogy\" | Reasoning by analogy is often valid. The fallacy is asserting similarity on the relevant dimensions without showing it. |\n| [D] Spotting fallacies only in arguments you disagree with | If the filter has a personal valence, it is broken. Run it on your own most-loved arguments. |\n| [D] \"No formal fallacy → the argument is sound\" | Formal fallacies are a small fraction. Most modern errors are informal. Run all four passes. |\n| [D] \"Knowing the conjunction fallacy means I won't commit it\" | Tversky-Kahneman 1983 showed otherwise. Only the explicit check prevents the error. |\n| [D] \"I caught the fallacy quickly, so I'm good at this\" | Speed signals surface pattern-matching, not an actual audit. Slow is good. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Fallacy label with no articulation of which inference step fails and why\n- Audit found fallacies only in arguments the auditor already disagreed with\n- Only the structural pass ran; linguistic, cognitive, or rhetorical skipped\n- No distinction between \"the argument fails\" and \"the conclusion is false\"\n- Output is a list of labels with no \"what would actually support this claim\" section\n\n## Verification\n\n- [ ] Argument rewritten as premises → conclusion (or flagged as not an argument)\n- [ ] All four passes run: structural, linguistic, cognitive, rhetorical\n- [ ] Each fallacy specifies exact location and exact failure\n- [ ] Fallacy-fallacy check performed — no findings are dismissals dressed as fallacies\n- [ ] Verdict separates \"argument fails\" from \"conclusion is false\"\n- [ ] Output names what evidence would actually support the claim\n- [ ] At least one cognitive fallacy (modern empirical category) considered\n- [ ] Filter applied to arguments the auditor agrees with, not only opponents'\n\n---\n\n*Part of **deciqAI Knowledge Skills** — 163 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/skills/logical-fallacies?utm_source=clawhub&utm_medium=marketplace&utm_campaign=knowledge-skills&utm_content=logical-fallacies** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*\n\nFile v1.0.2:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.2\",\n  \"publishedAt\": 1783471979849\n}\n\nFile v1.0.2:references/sources.md\n\n# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n\nFile v1.0.2:examples/tversky-kahnemans-linda-problem-1983.md\n\n# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky-Kahneman showed that for the modern cognitive fallacies, the *default* state of human reasoning is to commit them.\n\n**Second**, the fallacy survived **knowing the rule**. Subjects who could state the conjunction axiom in the abstract still violated it in the concrete. This is the deep finding: knowing what a fallacy is does not, by itself, prevent committing one. Only a **procedure** — a check applied at the point of judgment — does. That is exactly what this skill is.\n\n**Third**, the fallacy was **invisible from the inside**. Subjects rated T∧F more probable than T and did not experience cognitive dissonance. The wrong answer *felt right*. This is the deepest reason the fallacy filter has to be external — a checklist, a written audit, a colleague asking — rather than introspective. Your mind cannot reliably catch its own conjunction fallacies. It can be trained to *suspect* them in patterns (\"is this story making a more specific claim feel more likely?\") and route to the explicit check.\n\n**Fourth**, the case shows the **right and wrong use of fallacy-naming**. The right use: identify the move, repair the inference, judge the underlying claim on its actual evidence (what *is* the probability Linda is a bank teller? you'd need actual base rates). The wrong use: announce \"conjunction fallacy!\" and consider Linda's profile rebutted. The fallacy invalidates the *intuition*; it does not establish that Linda *isn't* a feminist bank teller.\n\nThe arc from Aristotle's 350 BCE taxonomy to Tversky-Kahneman's 1983 experiment is the arc this skill teaches. Aristotle gave us the **language** for the moves; Tversky-Kahneman gave us the **evidence** that the moves are not optional features of bad arguers but mandatory features of human cognition. Both are required. The taxonomy without the empirical map underestimates the threat. The empirical map without the taxonomy has no vocabulary to name what just happened.\n\nFile v1.0.2:skill-card.md\n\n## Description: <br>\nGuides agents through a structured fallacy audit of arguments, covering classical fallacy patterns and modern cognitive errors. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nExternal users, developers, and agents use this skill to examine persuasive arguments, identify where an inference fails, and separate weak reasoning from the truth of the underlying conclusion. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The skill influences how an agent critiques arguments and could over-label rhetoric as a fallacy. <br>\nMitigation: Require each finding to identify the exact inference failure and run the fallacy-fallacy check before accepting the critique. <br>\nRisk: A user may treat a fallacious argument as proof that the conclusion is false. <br>\nMitigation: Keep the verdict separate from the conclusion and state what evidence would actually support or settle the claim. <br>\n\n\n## Reference(s): <br>\n- [Sources - logical-fallacies](references/sources.md) <br>\n- [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md) <br>\n- [Aristotle, Sophistical Refutations](https://classics.mit.edu/Aristotle/sophist_refut.html) <br>\n- [Stanford Encyclopedia of Philosophy: Aristotle's Logic](https://plato.stanford.edu/entries/aristotle-logic/) <br>\n- [Tversky & Kahneman (1983), Extensional versus Intuitive Reasoning](https://doi.org/10.1037/0033-295X.90.4.293) <br>\n- [Tversky & Kahneman (1974), Judgment under Uncertainty](https://doi.org/10.1126/science.185.4157.1124) <br>\n- [Stanford Encyclopedia of Philosophy: Fallacies](https://plato.stanford.edu/entries/fallacies/) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Text, Markdown, Guidance] <br>\n**Output Format:** [Markdown fallacy audit with structured findings and verdict] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [No code execution; produces reasoning guidance and audit text.] <br>\n\n## Skill Version(s): <br>\n1.0.2 (source: server release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.1: 5 files, 9645 bytes\n\nFiles: examples/tversky-kahnemans-linda-problem-1983.md (4966b), references/sources.md (2208b), skill-card.md (2803b), SKILL.md (8747b), _meta.json (136b)\n\nFile v1.0.1:SKILL.md\n\n---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low.\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: [`critical-thinking`](../critical-thinking/SKILL.md) audits evidence quality and framing; [`first-principles`](../first-principles/SKILL.md) attacks premises; [`mece`](../mece/SKILL.md) catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not advance until user responds]**\n\n## The Process\n\nRun the **Fallacy Audit** in four passes — structure, language, cognition, rhetoric — then judge.\n\n1. **State the argument cleanly.** Rewrite as premises → conclusion. If you cannot, first finding: it's an assertion dressed as an argument.\n2. **Structural pass.** Begging the question, affirming the consequent, denying the antecedent, false cause (*post hoc*), hasty generalization, accident, complex question, ignoratio elenchi.\n3. **Linguistic pass.** Equivocation (key term shifts meaning), amphiboly, composition/division (part↔whole), accent/figure of speech.\n4. **Cognitive pass.** Conjunction fallacy (P(A∧B) > P(A) from representativeness), base-rate neglect, availability, anchoring — see [`anchoring`](../anchoring/SKILL.md).\n5. **Rhetorical-trap pass.** Ad hominem, appeal to authority (exception: expert in own domain), appeal to popularity (weak prior only), appeal to emotion (evidence vs. substitute), false dichotomy, straw man, slippery slope, tu quoque.\n6. **Judge the argument, not the moves.** A fallacy means the inference fails — not that the conclusion is false.\n7. **Fallacy-fallacy check.** If you can't articulate *why this instance fails*, you have a dismissal, not a finding.\n8. **Output:** for each fallacy — (a) which and where, (b) what the inference fails to establish, (c) what would actually support the conclusion.\n\n### Output: the Fallacy Audit\n\n```\nArgument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>\n```\n\n*→ Method in Action: [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md)*\n\n## Pack: Common Fallacy Patterns by Domain\n\n- **Startup/business:** survivorship bias (hasty generalization on filtered data), appeal to authority (\"Sequoia said X\"), anecdote as evidence, false dichotomy (\"raise now or die\")\n- **Policy/public debate:** undocumented slippery slope, straw man, ad hominem, appeal to consequences\n- **Internal arguments (most expensive):** conjunction fallacy, confirmation bias, sunk cost, optimism bias — see [`expected-value-and-kelly`](../expected-value-and-kelly/SKILL.md) and [`probabilistic-thinking`](../probabilistic-thinking/SKILL.md)\n- **Statistical/data:** base-rate neglect, Texas sharpshooter (target drawn after the fact), cherry-picking, regression to mean confused with causation\n\n## Applying It Well\n\nRun the filter on your *own* arguments first. The ones that survive are worth more than the ones you flag in opponents. Fallacy-detection at speed is suspicious — it usually means surface pattern-matching, not an actual audit. Slow is good.\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I named the fallacy, so I refuted the argument\" | A fallacious argument doesn't make the conclusion false — it fails to *establish* it. Refuting an argument ≠ refuting a claim. |\n| [D] \"Citing an expert is appeal to authority\" | Expert testimony in the expert's actual domain is legitimate evidence. The fallacy is treating a citation as *proof* or citing outside their domain. |\n| [D] \"Any analogy is a false analogy\" | Reasoning by analogy is often valid. The fallacy is asserting similarity on the relevant dimensions without showing it. |\n| [D] Spotting fallacies only in arguments you disagree with | If the filter has a personal valence, it is broken. Run it on your own most-loved arguments. |\n| [D] \"No formal fallacy → the argument is sound\" | Formal fallacies are a small fraction. Most modern errors are informal. Run all four passes. |\n| [D] \"Knowing the conjunction fallacy means I won't commit it\" | Tversky-Kahneman 1983 showed otherwise. Only the explicit check prevents the error. |\n| [D] \"I caught the fallacy quickly, so I'm good at this\" | Speed signals surface pattern-matching, not an actual audit. Slow is good. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Fallacy label with no articulation of which inference step fails and why\n- Audit found fallacies only in arguments the auditor already disagreed with\n- Only the structural pass ran; linguistic, cognitive, or rhetorical skipped\n- No distinction between \"the argument fails\" and \"the conclusion is false\"\n- Output is a list of labels with no \"what would actually support this claim\" section\n\n## Verification\n\n- [ ] Argument rewritten as premises → conclusion (or flagged as not an argument)\n- [ ] All four passes run: structural, linguistic, cognitive, rhetorical\n- [ ] Each fallacy specifies exact location and exact failure\n- [ ] Fallacy-fallacy check performed — no findings are dismissals dressed as fallacies\n- [ ] Verdict separates \"argument fails\" from \"conclusion is false\"\n- [ ] Output names what evidence would actually support the claim\n- [ ] At least one cognitive fallacy (modern empirical category) considered\n- [ ] Filter applied to arguments the auditor agrees with, not only opponents'\n\n---\n\n*Part of **deciqAI Knowledge Skills** — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.*\n\nFile v1.0.1:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.1\",\n  \"publishedAt\": 1783463293788\n}\n\nFile v1.0.1:references/sources.md\n\n# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n\nFile v1.0.1:examples/tversky-kahnemans-linda-problem-1983.md\n\n# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky-Kahneman showed that for the modern cognitive fallacies, the *default* state of human reasoning is to commit them.\n\n**Second**, the fallacy survived **knowing the rule**. Subjects who could state the conjunction axiom in the abstract still violated it in the concrete. This is the deep finding: knowing what a fallacy is does not, by itself, prevent committing one. Only a **procedure** — a check applied at the point of judgment — does. That is exactly what this skill is.\n\n**Third**, the fallacy was **invisible from the inside**. Subjects rated T∧F more probable than T and did not experience cognitive dissonance. The wrong answer *felt right*. This is the deepest reason the fallacy filter has to be external — a checklist, a written audit, a colleague asking — rather than introspective. Your mind cannot reliably catch its own conjunction fallacies. It can be trained to *suspect* them in patterns (\"is this story making a more specific claim feel more likely?\") and route to the explicit check.\n\n**Fourth**, the case shows the **right and wrong use of fallacy-naming**. The right use: identify the move, repair the inference, judge the underlying claim on its actual evidence (what *is* the probability Linda is a bank teller? you'd need actual base rates). The wrong use: announce \"conjunction fallacy!\" and consider Linda's profile rebutted. The fallacy invalidates the *intuition*; it does not establish that Linda *isn't* a feminist bank teller.\n\nThe arc from Aristotle's 350 BCE taxonomy to Tversky-Kahneman's 1983 experiment is the arc this skill teaches. Aristotle gave us the **language** for the moves; Tversky-Kahneman gave us the **evidence** that the moves are not optional features of bad arguers but mandatory features of human cognition. Both are required. The taxonomy without the empirical map underestimates the threat. The empirical map without the taxonomy has no vocabulary to name what just happened.\n\nFile v1.0.1:skill-card.md\n\n## Description: <br>\nHelps an agent audit persuasive arguments by rewriting premises and conclusions, checking structural, linguistic, cognitive, and rhetorical fallacies, and separating failed inference from the truth of the conclusion. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nDevelopers, analysts, writers, and decision-makers use this skill to test whether an argument actually supports its conclusion. It is suited for reviewing persuasive claims, debate moves, business reasoning, policy arguments, and self-auditing before decisions. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The agent may name a fallacy without proving which inference step fails, turning analysis into dismissal. <br>\nMitigation: Require the structured audit fields to identify the fallacy, location, failed inference, and what evidence would actually support the claim. <br>\nRisk: The skill's reasoning style may not fit every workflow or decision context. <br>\nMitigation: Review the generated audit for fit with the user's domain and verify empirical claims against data when the conclusion can be checked directly. <br>\n\n\n## Reference(s): <br>\n- [Primary sources for logical-fallacies](references/sources.md) <br>\n- [Method in Action: Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md) <br>\n- [Aristotle, Sophistical Refutations](https://classics.mit.edu/Aristotle/sophist_refut.html) <br>\n- [Stanford Encyclopedia of Philosophy: Aristotle's Logic](https://plato.stanford.edu/entries/aristotle-logic/) <br>\n- [Tversky and Kahneman 1983: Extensional versus intuitive reasoning](https://doi.org/10.1037/0033-295X.90.4.293) <br>\n- [Tversky and Kahneman 1974: Judgment under Uncertainty](https://doi.org/10.1126/science.185.4157.1124) <br>\n- [Stanford Encyclopedia of Philosophy: Fallacies](https://plato.stanford.edu/entries/fallacies/) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Text, Markdown, Guidance] <br>\n**Output Format:** [Markdown fallacy audit with structured findings and repair guidance] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [May ask stepwise coaching questions before producing an audit when the user is a novice or has not provided a concrete argument.] <br>\n\n## Skill Version(s): <br>\n1.0.1 (source: server release metadata) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>\n\nArchive v1.0.0: 5 files, 9391 bytes\n\nFiles: examples/tversky-kahnemans-linda-problem-1983.md (4966b), references/sources.md (2208b), skill-card.md (2204b), SKILL.md (8747b), _meta.json (136b)\n\nFile v1.0.0:SKILL.md\n\n---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low.\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: [`critical-thinking`](../critical-thinking/SKILL.md) audits evidence quality and framing; [`first-principles`](../first-principles/SKILL.md) attacks premises; [`mece`](../mece/SKILL.md) catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not advance until user responds]**\n\n## The Process\n\nRun the **Fallacy Audit** in four passes — structure, language, cognition, rhetoric — then judge.\n\n1. **State the argument cleanly.** Rewrite as premises → conclusion. If you cannot, first finding: it's an assertion dressed as an argument.\n2. **Structural pass.** Begging the question, affirming the consequent, denying the antecedent, false cause (*post hoc*), hasty generalization, accident, complex question, ignoratio elenchi.\n3. **Linguistic pass.** Equivocation (key term shifts meaning), amphiboly, composition/division (part↔whole), accent/figure of speech.\n4. **Cognitive pass.** Conjunction fallacy (P(A∧B) > P(A) from representativeness), base-rate neglect, availability, anchoring — see [`anchoring`](../anchoring/SKILL.md).\n5. **Rhetorical-trap pass.** Ad hominem, appeal to authority (exception: expert in own domain), appeal to popularity (weak prior only), appeal to emotion (evidence vs. substitute), false dichotomy, straw man, slippery slope, tu quoque.\n6. **Judge the argument, not the moves.** A fallacy means the inference fails — not that the conclusion is false.\n7. **Fallacy-fallacy check.** If you can't articulate *why this instance fails*, you have a dismissal, not a finding.\n8. **Output:** for each fallacy — (a) which and where, (b) what the inference fails to establish, (c) what would actually support the conclusion.\n\n### Output: the Fallacy Audit\n\n```\nArgument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>\n```\n\n*→ Method in Action: [Tversky & Kahneman's Linda Problem (1983)](examples/tversky-kahnemans-linda-problem-1983.md)*\n\n## Pack: Common Fallacy Patterns by Domain\n\n- **Startup/business:** survivorship bias (hasty generalization on filtered data), appeal to authority (\"Sequoia said X\"), anecdote as evidence, false dichotomy (\"raise now or die\")\n- **Policy/public debate:** undocumented slippery slope, straw man, ad hominem, appeal to consequences\n- **Internal arguments (most expensive):** conjunction fallacy, confirmation bias, sunk cost, optimism bias — see [`expected-value-and-kelly`](../expected-value-and-kelly/SKILL.md) and [`probabilistic-thinking`](../probabilistic-thinking/SKILL.md)\n- **Statistical/data:** base-rate neglect, Texas sharpshooter (target drawn after the fact), cherry-picking, regression to mean confused with causation\n\n## Applying It Well\n\nRun the filter on your *own* arguments first. The ones that survive are worth more than the ones you flag in opponents. Fallacy-detection at speed is suspicious — it usually means surface pattern-matching, not an actual audit. Slow is good.\n\n*→ Primary sources: [references/sources.md](references/sources.md)*\n\n## Common Rationalizations\n\n**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**\n\n| Fake move | Reality |\n|---|---|\n| [D] \"I named the fallacy, so I refuted the argument\" | A fallacious argument doesn't make the conclusion false — it fails to *establish* it. Refuting an argument ≠ refuting a claim. |\n| [D] \"Citing an expert is appeal to authority\" | Expert testimony in the expert's actual domain is legitimate evidence. The fallacy is treating a citation as *proof* or citing outside their domain. |\n| [D] \"Any analogy is a false analogy\" | Reasoning by analogy is often valid. The fallacy is asserting similarity on the relevant dimensions without showing it. |\n| [D] Spotting fallacies only in arguments you disagree with | If the filter has a personal valence, it is broken. Run it on your own most-loved arguments. |\n| [D] \"No formal fallacy → the argument is sound\" | Formal fallacies are a small fraction. Most modern errors are informal. Run all four passes. |\n| [D] \"Knowing the conjunction fallacy means I won't commit it\" | Tversky-Kahneman 1983 showed otherwise. Only the explicit check prevents the error. |\n| [D] \"I caught the fallacy quickly, so I'm good at this\" | Speed signals surface pattern-matching, not an actual audit. Slow is good. |\n| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |\n\n## Red Flags\n\n- Fallacy label with no articulation of which inference step fails and why\n- Audit found fallacies only in arguments the auditor already disagreed with\n- Only the structural pass ran; linguistic, cognitive, or rhetorical skipped\n- No distinction between \"the argument fails\" and \"the conclusion is false\"\n- Output is a list of labels with no \"what would actually support this claim\" section\n\n## Verification\n\n- [ ] Argument rewritten as premises → conclusion (or flagged as not an argument)\n- [ ] All four passes run: structural, linguistic, cognitive, rhetorical\n- [ ] Each fallacy specifies exact location and exact failure\n- [ ] Fallacy-fallacy check performed — no findings are dismissals dressed as fallacies\n- [ ] Verdict separates \"argument fails\" from \"conclusion is false\"\n- [ ] Output names what evidence would actually support the claim\n- [ ] At least one cognitive fallacy (modern empirical category) considered\n- [ ] Filter applied to arguments the auditor agrees with, not only opponents'\n\n---\n\n*Part of **deciqAI Knowledge Skills** — open-source thinking skills that make rigor executable for AI agents. Built by deciqAI · https://deciqai.com · Contributions welcome — see the template at the repo root.*\n\nFile v1.0.0:_meta.json\n\n{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.0\",\n  \"publishedAt\": 1782731862214\n}\n\nFile v1.0.0:references/sources.md\n\n# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n\nFile v1.0.0:examples/tversky-kahnemans-linda-problem-1983.md\n\n# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky-Kahneman showed that for the modern cognitive fallacies, the *default* state of human reasoning is to commit them.\n\n**Second**, the fallacy survived **knowing the rule**. Subjects who could state the conjunction axiom in the abstract still violated it in the concrete. This is the deep finding: knowing what a fallacy is does not, by itself, prevent committing one. Only a **procedure** — a check applied at the point of judgment — does. That is exactly what this skill is.\n\n**Third**, the fallacy was **invisible from the inside**. Subjects rated T∧F more probable than T and did not experience cognitive dissonance. The wrong answer *felt right*. This is the deepest reason the fallacy filter has to be external — a checklist, a written audit, a colleague asking — rather than introspective. Your mind cannot reliably catch its own conjunction fallacies. It can be trained to *suspect* them in patterns (\"is this story making a more specific claim feel more likely?\") and route to the explicit check.\n\n**Fourth**, the case shows the **right and wrong use of fallacy-naming**. The right use: identify the move, repair the inference, judge the underlying claim on its actual evidence (what *is* the probability Linda is a bank teller? you'd need actual base rates). The wrong use: announce \"conjunction fallacy!\" and consider Linda's profile rebutted. The fallacy invalidates the *intuition*; it does not establish that Linda *isn't* a feminist bank teller.\n\nThe arc from Aristotle's 350 BCE taxonomy to Tversky-Kahneman's 1983 experiment is the arc this skill teaches. Aristotle gave us the **language** for the moves; Tversky-Kahneman gave us the **evidence** that the moves are not optional features of bad arguers but mandatory features of human cognition. Both are required. The taxonomy without the empirical map underestimates the threat. The empirical map without the taxonomy has no vocabulary to name what just happened.\n\nFile v1.0.0:skill-card.md\n\n## Description: <br>\nHelps an agent audit arguments by rewriting premises and conclusions, checking structural, linguistic, cognitive, and rhetorical fallacy patterns, and separating failed inference from the truth of the underlying claim. <br>\n\nThis skill is ready for commercial/non-commercial use. <br>\n\n## Publisher: <br>\n[deciqai](https://clawhub.ai/user/deciqai) <br>\n\n### License/Terms of Use: <br>\nMIT-0 <br>\n\n\n## Use Case: <br>\nExternal users and developers use this skill to help an agent evaluate persuasive arguments, identify fallacy patterns, and explain what evidence or inference would actually support a conclusion. <br>\n\n### Deployment Geography for Use: <br>\nGlobal <br>\n\n## Known Risks and Mitigations: <br>\nRisk: The agent may be used in authenticated administrative workflows where sensitive actions require explicit intent. <br>\nMitigation: Use authenticated access only when those actions are intended, and keep confirmation and dry-run gates in place. <br>\n\n\n## Reference(s): <br>\n- [ClawHub skill page](https://clawhub.ai/deciqai/skills/logical-fallacies) <br>\n- [Sources](references/sources.md) <br>\n- [Tversky and Kahneman Linda Problem Example](examples/tversky-kahnemans-linda-problem-1983.md) <br>\n- [Aristotle, Sophistical Refutations](https://classics.mit.edu/Aristotle/sophist_refut.html) <br>\n- [Stanford Encyclopedia of Philosophy: Fallacies](https://plato.stanford.edu/entries/fallacies/) <br>\n- [Tversky and Kahneman 1983](https://doi.org/10.1037/0033-295X.90.4.293) <br>\n- [Tversky and Kahneman 1974](https://doi.org/10.1126/science.185.4157.1124) <br>\n\n\n## Skill Output: <br>\n**Output Type(s):** [Analysis, Markdown, Guidance] <br>\n**Output Format:** [Markdown] <br>\n**Output Parameters:** [1D] <br>\n**Other Properties Related to Output:** [Structured fallacy audit with findings, verdict, and repair guidance] <br>\n\n## Skill Version(s): <br>\n1.0.0 (source: server release evidence) <br>\n\n## Ethical Considerations: <br>\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment. <br>","readmeExcerpt":"Skill: Logical Fallacies Owner: deciqai Summary: Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning... Tags: latest:1.0.5 Version history: v1.0.5 | 2026-07-16T18:05:10.879Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/logical-fallacies.json) v1.0.4 | 2026-07-10T10:26:53.800","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"Argument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>"},{"language":"text","snippet":"Argument A: CEO predicts near-term transformative AI / they see the frontier / therefore it's near\n  Structural: —\n  Linguistic: —\n  Cognitive: base-rate neglect (history of failed AI timelines)\n  Rhetorical: appeal to authority — expert on present capability ≠ authority on multi-year forecast; conflict of interest\n  Verdict: fails to establish | still open — needs track record of calibrated forecasts, not a single seat\n  Repair: aggregate multiple independent forecasters; score them against past calibration; treat timelines as probability distributions\n\nArgument B: AI → doom or utopia / one is implausible / therefore the other\n  Structural: —\n  Linguistic: vague terms make the dilemma feel exhaustive\n  Cognitive: —\n  Rhetorical: false dichotomy — ignores the large middle of outcomes\n  Verdict: fails to establish | still open — enumerate the full outcome space, assign probabilities to each\n  Repair: replace binary with a distribution over scenarios; argue each on its own evidence\n\nArgument C: one viral demo of task X / therefore reliable at X\n  Structural: hasty generalization from a filtered sample of one\n  Linguistic: —\n  Cognitive: availability heuristic — vivid demo feels typical\n  Rhetorical: —\n  Verdict: fails to establish | still open — needs pass rate over many unfiltered trials\n  Repair: measure success rate on a held-out, contamination-checked test set; report variance and failure modes\n\nArgument D: if intelligent then passes B / passed B / therefore intelligent\n  Structural: affirming the consequent (invalid form)\n  Linguistic: equivocation on \"intelligent\" (rich sense vs. thin task sense)\n  Cognitive: —\n  Rhetorical: —\n  Verdict: fails to establish | still open — define \"intelligent\" operationally, test generalization beyond B, rule out benchmark contamination\n  Repair: use held-out tasks the model wasn't optimized for; check for train/test leakage; specify which capability the benchmark actually measures\n\nFallacy-fallacy check: no conclusion is asserted"},{"language":"text","snippet":"Argument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>"},{"language":"text","snippet":"Argument A: CEO predicts near-term transformative AI / they see the frontier / therefore it's near\n  Structural: —\n  Linguistic: —\n  Cognitive: base-rate neglect (history of failed AI timelines)\n  Rhetorical: appeal to authority — expert on present capability ≠ authority on multi-year forecast; conflict of interest\n  Verdict: fails to establish | still open — needs track record of calibrated forecasts, not a single seat\n  Repair: aggregate multiple independent forecasters; score them against past calibration; treat timelines as probability distributions\n\nArgument B: AI → doom or utopia / one is implausible / therefore the other\n  Structural: —\n  Linguistic: vague terms make the dilemma feel exhaustive\n  Cognitive: —\n  Rhetorical: false dichotomy — ignores the large middle of outcomes\n  Verdict: fails to establish | still open — enumerate the full outcome space, assign probabilities to each\n  Repair: replace binary with a distribution over scenarios; argue each on its own evidence\n\nArgument C: one viral demo of task X / therefore reliable at X\n  Structural: hasty generalization from a filtered sample of one\n  Linguistic: —\n  Cognitive: availability heuristic — vivid demo feels typical\n  Rhetorical: —\n  Verdict: fails to establish | still open — needs pass rate over many unfiltered trials\n  Repair: measure success rate on a held-out, contamination-checked test set; report variance and failure modes\n\nArgument D: if intelligent then passes B / passed B / therefore intelligent\n  Structural: affirming the consequent (invalid form)\n  Linguistic: equivocation on \"intelligent\" (rich sense vs. thin task sense)\n  Cognitive: —\n  Rhetorical: —\n  Verdict: fails to establish | still open — define \"intelligent\" operationally, test generalization beyond B, rule out benchmark contamination\n  Repair: use held-out tasks the model wasn't optimized for; check for train/test leakage; specify which capability the benchmark actually measures\n\nFallacy-fallacy check: no conclusion is asserted"},{"language":"text","snippet":"Argument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>"},{"language":"text","snippet":"Argument: Premise 1 / Premise 2 / Conclusion\nStructural findings: <fallacy, which inference, why it fails>\nLinguistic findings: <fallacy, which term, what shifts>\nCognitive findings: <fallacy, why intuition misfires>\nRhetorical-trap findings: <fallacy, what move replaced an argument>\nFallacy-fallacy check: <any dismissals dressed as findings?>\nVerdict: argument <fails/partly/holds> | conclusion <still open — what evidence would settle it>\nRepair: <what inference or evidence would actually support the claim>"}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\nname: logical-fallacies\ndescription: \"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning?', an argument relies entirely on authority/emotion/popularity, or you're about to decide based on a single analogy. Do NOT activate when: the conclusion is already verifiable empirically (just check the data); casual conversation where rigor is socially expensive and stakes are low. More: deciqai.com/c/logical-fallacies\"\n---\n\n# Logical Fallacies\n\n## Overview\n\nA fallacy is an argument that looks like it works but doesn't. The test is not whether the conclusion is true — it's whether the *inference* from premises to conclusion is valid. This skill covers two layers: the **classical taxonomy** (Aristotle's 13, c. 350 BCE — verbal and structural errors) and the **modern cognitive map** (Tversky-Kahneman 1983 — errors competent reasoners commit automatically before any sophist arrives).\n\nComposes with neighbors: `critical-thinking` audits evidence quality and framing; `first-principles` attacks premises; `mece` catches decomposition errors that masquerade as false-dichotomy or composition fallacies.\n\n## When to Use\n\n- An argument feels persuasive but you cannot articulate why\n- A claim is supported entirely by authority, popularity, emotion, or anecdote\n- You're about to decide based on a single argument or analogy\n- A debate is moving fast (\"everyone knows Y\") — speed is the sophist's friend\n- You catch yourself reasoning emotionally (\"this has to be true because…\")\n- You're weighing an AI hype or AI-adoption claim (\"a lab CEO said it's near,\" \"it passed the benchmark so it's intelligent,\" \"doom vs. utopia\")\n\n**When NOT to use:** casual small talk with low stakes; conclusion is empirically verifiable (just check the data); you're tempted to name a fallacy to dismiss an opponent rather than find truth (that is itself the fallacy fallacy).\n\n## Coaching Novices (Adaptive Front Door)\n\n- **Engine mode:** user has a concrete argument → run The Process directly.\n- **Coach mode:** user signals unfamiliarity or has no concrete case → guide step by step.\n\nIn Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.\n\n1. One-line what-it-is: some arguments *sound* right but don't earn their conclusion — this is a checklist for finding that gap, including in your own thinking.\n2. Check fit against When to Use / When NOT to use. If data can answer it, say so.\n3. Elicit their real argument — ask for a concrete case (something someone said, an article they're suspicious of). > **[WAIT — do not advance until user responds]**\n4. Walk through the Audit one pass per turn: pose the question, wait for their answer, surface what they missed. > **[WAIT — do not advance until user responds]**\n5. Close by naming the one fallacy *they* found and what changes about the conclusion now that they've seen it. > **[WAIT — do not a"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn754b8sk22s8c6gjxt02bftbn88q7ye\",\n  \"slug\": \"logical-fallacies\",\n  \"version\": \"1.0.5\",\n  \"publishedAt\": 1784225110879\n}"},{"path":"references/sources.md","content":"# Sources — logical-fallacies\n\n> *Primary sources for the [logical-fallacies](../SKILL.md) skill.*\n\n- Aristotle, *Sophistical Refutations* (*Peri Sophistikōn Elenchōn*), c. 350 BCE, trans. W. A. Pickard-Cambridge, in *The Complete Works of Aristotle*, Princeton/Bollingen. The founding taxonomy: six linguistic fallacies (equivocation, amphiboly, composition, division, accent, figure of speech) and seven non-linguistic (accident, hasty generalization, ignoratio elenchi, begging the question, false cause, complex question, affirming the consequent). Full text: https://classics.mit.edu/Aristotle/sophist_refut.html\n- Stanford Encyclopedia of Philosophy, *Aristotle's Logic*, §3 \"The Subject of Logic: Syllogisms\" and §6 \"Demonstrations and Demonstrative Sciences\" — for the relationship between fallacy taxonomy and Aristotle's broader logical project. https://plato.stanford.edu/entries/aristotle-logic/\n- Tversky, A., & Kahneman, D. (1983). \"Extensional versus intuitive reasoning: The conjunction fallacy in probability judgment.\" *Psychological Review*, 90(4), 293–315. The Linda problem; the empirical demonstration that competent reasoners (including statistically-trained subjects) systematically commit the conjunction fallacy by substituting representativeness for probability. https://doi.org/10.1037/0033-295X.90.4.293\n- Tversky, A., & Kahneman, D. (1974). \"Judgment under Uncertainty: Heuristics and Biases.\" *Science*, 185(4157), 1124–1131. The earlier programmatic paper introducing representativeness, availability, and anchoring as the cognitive substrate from which many modern fallacies arise. https://doi.org/10.1126/science.185.4157.1124\n- Stanford Encyclopedia of Philosophy, *Fallacies* — a comprehensive modern survey distinguishing formal vs informal fallacies and tracing each through the literature from Aristotle onward. https://plato.stanford.edu/entries/fallacies/\n- Walton, Douglas (2008). *Informal Logic: A Pragmatic Approach* (2nd ed.), Cambridge University Press — the standard modern reference on informal fallacy theory, treating fallacies as misuses of otherwise-legitimate argumentation schemes (e.g., when *ad verecundiam* is and is not legitimate).\n- Stanford Institute for Human-Centered AI (HAI), *AI Index Report* (annual, 2024 edition and later). Documents the measurement problems behind AI capability claims — benchmark saturation and data contamination (test items leaking into training data), which is what makes \"it passed the benchmark, therefore it's intelligent\" an instance of affirming the consequent rather than a valid inference. https://aiindex.stanford.edu/report/\n- Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). \"On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?\" *Proceedings of FAccT '21*, 610–623. A durable reference for the equivocation-on-\"intelligence\" and hasty-generalization patterns in AI discourse — arguing that fluent output can be mistaken for understanding. https://doi.org/10."},{"path":"examples/ai-hype-discourse-fallacies-2024-2026.md","content":"# Method in Action: Four Fallacies in the 2024–2026 AI Debate\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nThe Linda problem shows the fallacy filter under laboratory conditions. This example runs the same **Fallacy Audit** on messy public discourse — the argument about artificial intelligence as it circulated across roughly 2024–2025. This is the environment the skill was built for: claims that *sound* authoritative, move fast, and carry high stakes, where naming the fallacy is only useful if you can also say what would actually settle the question.\n\nWe audit four representative arguments, each a real pattern that recurred across essays, interviews, and social threads in this period. No individual quote is reconstructed verbatim below; each argument is stated as a clean paraphrase of a widely-circulated *pattern* of reasoning, then run through the process.\n\n---\n\n## Step 1 — State each argument cleanly (premises → conclusion)\n\n**Argument A (authority).**\n- P1: The CEO of a leading AI lab says transformative AI is a few years away.\n- P2: They run the lab and see the frontier models first.\n- C: Therefore transformative AI is a few years away.\n\n**Argument B (false dilemma).**\n- P1: AI leads either to catastrophe (\"doom\") or to radical abundance (\"utopia\").\n- P2: The doom scenario is implausible / the utopia scenario is implausible (whichever the speaker rejects).\n- C: Therefore the other outcome is what we should expect.\n\n**Argument C (hasty generalization).**\n- P1: A single demo went viral showing a model doing task X impressively.\n- C: Therefore models can now do X (and tasks like X) reliably.\n\n**Argument D (affirming the consequent).**\n- P1: If a system is intelligent, it will pass benchmark B.\n- P2: This system passed benchmark B.\n- C: Therefore this system is intelligent.\n\nAll four are genuine arguments (premises and a conclusion), so none collapses at Step 1 into \"an assertion dressed as an argument.\" Good — that means the work is in the passes.\n\n## Step 2 — Structural pass\n\n**Argument D is a textbook formal fallacy: affirming the consequent.** The form is \"If P then Q; Q; therefore P.\" That is invalid: Q can be true for reasons unrelated to P. Passing benchmark B is consistent with intelligence *and* with narrow pattern-matching, benchmark contamination (test items leaking into training data), or overfitting to the benchmark's format. The inference fails because P1 only licenses the reverse direction (intelligent → passes B), not (passes B → intelligent). Benchmark saturation across this period — models scoring very high on tests that older models failed — is exactly the observation that makes the invalid direction tempting.\n\n**Argument C is hasty generalization.** One vivid, curated, possibly cherry-picked instance is generalized to reliable capability across a class of tasks. A viral demo is a maximally-filtered sample: the impressive run is the one that got posted. The inference from \"did X once, on camera\" to \"does X reli"},{"path":"examples/tversky-kahnemans-linda-problem-1983.md","content":"# Method in Action: Tversky & Kahneman's Linda Problem (1983)\n\n> *Example for the [logical-fallacies](../SKILL.md) skill.*\n\nTo see logical-fallacy detection at full empirical force — not as debate-club gotcha but as an X-ray of how competent minds reason — the cleanest case in the modern literature is **Amos Tversky and Daniel Kahneman's 1983 paper** *\"Extensional versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment\"* (*Psychological Review*, Vol. 90, No. 4, pp. 293–315).\n\nThe experiment is famous, but its construction matters. Subjects were given a personality vignette:\n\n> \"Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297. https://doi.org/10.1037/0033-295X.90.4.293\n\nSubjects then ranked the probability of eight statements about Linda. Two of the statements were:\n\n- (T) \"Linda is a bank teller.\"\n- (T∧F) \"Linda is a bank teller and is active in the feminist movement.\"\n\nThe mathematical fact is **unambiguous**: P(T∧F) ≤ P(T), always, by the basic axioms of probability. The set of \"bank tellers who are feminists\" is a strict subset of the set of \"bank tellers.\" This is not interpretation. It is definition.\n\nThe empirical result was equally unambiguous:\n\n> \"85% of subjects ranked T∧F as more probable than T... The conjunction effect was observed even among graduate students who had completed several courses in probability and statistics.\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 299.\n\nThat second sentence is the dagger. The error was not the property of the untrained; it was committed by **statistically-educated graduate students at Stanford and the University of British Columbia** — people who, if asked the abstract question \"can a conjunction be more probable than its constituents?\", would answer \"of course not.\" Yet the *concrete* form of the question, dressed in a vivid representativeness-triggering personality sketch, broke their judgment systematically.\n\nTversky and Kahneman state the diagnosis precisely:\n\n> \"A conjunction cannot be more probable than one of its constituents. This fundamental rule of probability is violated... because the conjunction (a feminist bank teller) is more representative of Linda's personality than is the more inclusive category (a bank teller).\"\n\n— Tversky & Kahneman, *Psychological Review* 90(4), p. 297.\n\nHere is what the case demonstrates for logical-fallacy detection:\n\n**First**, the fallacy is **mechanical**, not malicious. No one in the experiment was trying to commit a fallacy. They were doing exactly what their minds spontaneously do — substituting the easy question (\"does this story fit my stereotype of Linda?\") for the hard question (\"which set is mathematically larger?\"). Aristotle's Sophists had to *work* to deploy fallacies; Tversky"}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning... Skill: Logical Fallacies Owner: deciqai Summary: Activate when: someone says 'this argument feels off but I can't explain why', 'is this a real argument or just rhetoric?', 'what's wrong with this reasoning... Tags: latest:1.0.5 Version history: v1.0.5 | 2026-07-16T18:05:10.879Z | user Description tail link + agents machine-readable metadata line (deciqai.com/s/logical-fallacies.json) v1.0.4 | 2026-07-10T10:26:53.800","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":2126,"uniquenessScore":48,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-11T13:20:10.514Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-11T13:20:10.514Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-11T16:01:10.104Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}