Use this skill when the user asks for critical thinking (including naming it or directing use/apply/run with obvious misspellings; decisive) or wants to eval...
---
name: critical-thinking
description: >
Use this skill when the user asks for critical thinking (including naming it
or directing use/apply/run with obvious misspellings; decisive) or wants to
evaluate a claim, argument, plan, or belief: clarify assertions, weigh evidence,
surface assumptions, test reasoning for gaps or fallacies, scan biases,
consider alternatives, and stress-test conclusions—whether they phrase it
plainly ("red team", "devil's advocate", "what am I missing", steel/straw
man, bias scan) or indirectly (decision-quality review, epistemic
calibration). Skip for execution-only tasks with no evaluative angle, or when
they only want wording, tone, layout, or open-ended brainstorming with no
request to audit reasons, assumptions, or evidence.
license: MIT
metadata:
author: ysskrishna
version: "2026.5.18"
---
# Critical Thinking
**Critical thinking** is disciplined inquiry that keeps **description** separate from **evaluation**: surface assumptions, weigh evidence, test logic, consider alternatives, then state a proportionate conclusion. If the conventional view is well-supported, say so — this is inquiry, not contrarianism by default.
**How to run it with this skill:** one phase per clearly headed section, always in this order: Clarify → Information → Assumptions → Reasoning → Alternatives → Conclusion. Always include **Conclusion** unless the user explicitly stops the whole review early.
---
## Setup (run before starting)
In one short block:
1. **Focus** — the specific claim, proposal, or question under review
2. **Pass** — Clarify → Information → Assumptions → Reasoning → Alternatives → Conclusion (fixed sequence; state this line once in the setup block so the user sees the path)
If essential context is missing, ask at most 3 questions in one message, then proceed. Note any remaining gaps or working guesses in plain language (no bracket tags in Setup).
---
## The Phases
### Clarify
Restate the target in one precise sentence. Separate **factual** vs **normative** claims. Name success criteria if a decision is involved.
### Information
What evidence exists? Each bullet starts with **`[CITED]`** or **`[MISSING]`**:
- **`[CITED]`** — a traceable basis (user text, repo, doc, link, study, etc.); in the same bullet, name the basis and one line on strength or limits (no extra strength tags).
- **`[MISSING]`** — no traceable basis yet for that point, or evidence was requested but not available.
### Assumptions
List tacit premises. For each: **Assumption:** … — **If false:** …
When the **Focus** mixes **is** and **should**, surface **value / normative** premises too (e.g. **Value premise:** … — **If rejected:** …) alongside factual assumptions where it clarifies the chain.
### Reasoning
Trace the argument chain. Flag **leaps**, **circular** patterns, **correlation vs causation**, and **missing steps**. No new factual assertions here — only structure. If a premise is needed but was never established in **Information**, do **not** assert it as true; label it as an **ungrounded premise** (structural gap only). When **values** and **evidence** both do work in the chain, show which links depend on which.
**Bias and fallacy pass (compact):** add a short sub-list — only items that apply; omit the rest rather than padding.
- **Biases to scan:** confirmation; anchoring; survivorship; undue authority; sunk cost — plus any other bias clearly relevant to the case.
- **Fallacies to name if present** (tie each to the chain above): ad hominem; straw man; false dichotomy; slippery slope; hasty generalization; begging the question.
If none apply, state that plainly in one line.
### Alternatives
Credible competing explanations, plans, or frames. Do not collapse into debate rhetoric; keep alternatives plausible.
### Conclusion
1. **Judgment** — answer the **Focus** directly; when factual and normative claims were both in play, separate **what follows from the cited evidence** from **what depends on value premises** (short clauses are enough). Close with **one sentence in plain language** on how strong the case is given `[CITED]` vs `[MISSING]` evidence.
2. **What would change the judgment** — concrete falsifiers or new data; phrase relative to the **Focus** (e.g. the claim-holder’s view, a named third party, or *this assessment* when the review is impersonal).
---
## Execution Rules
1. Run phases in one response unless the user requests step-by-step pacing.
2. Never merge **Information** and **Reasoning** in the same bullet block.
3. Do not smuggle new unsupported facts into **Conclusion**; only synthesize prior phases.
4. If the user is emotionally fused with a position, name it neutrally and continue the phase plan.
5. Be intellectually honest: acknowledge strong opposing evidence and uncertainty where the phases support it.
---
## Checklist (verify before responding)
- [ ] Setup block: **Focus** and stated **Pass** (fixed sequence)
- [ ] Each phase is its own section in canonical order (Clarify through Conclusion)
- [ ] Information: each bullet starts with `[CITED]` (basis + limits in-bullet) or `[MISSING]`
- [ ] Assumptions use **Assumption** / **If false** pairs; **Value premise** / **If rejected** when the Focus mixes facts and shoulds
- [ ] Reasoning references only what earlier phases established; flags **ungrounded premises** where needed; bias/fallacy pass done or explicitly "none identified"
- [ ] Conclusion: judgment (evidence vs values when both apply, plus one plain sentence on strength of case from Information), falsifiers phrased for the **Focus**
don't have the plugin yet? install it then click "run inline in claude" again.
separated intent, inputs, and procedure into explicit Implexa components; mapped the original six-phase structure into numbered procedure steps with clear input-output pairs; added decision points for pacing, missing context, early halts, and emotional fusion; documented the output contract as a section-by-section checklist; clarified outcome signals so the user knows the skill worked; expanded bias and fallacy guidance to say "omit the rest rather than padding" to enforce selectivity.
critical thinking is disciplined inquiry that keeps description separate from evaluation: surface assumptions, weigh evidence, test logic, consider alternatives, then state a proportionate conclusion. if the conventional view is well-supported, say so. this is inquiry, not contrarianism by default.
use this skill when you need to evaluate a claim, proposal, plan, or belief systematically. critical thinking separates what is known from what is assumed, tests reasoning for gaps and fallacies, and explores competing explanations. apply this when the user explicitly asks for red-teaming, devil's advocate review, bias scans, steel-manning or straw-manning, or when they frame a decision-quality or epistemic calibration request. skip this skill for execution-only work (writing copy, designing layouts, brainstorming without evaluation), or when the user wants feedback on tone and wording alone with no request to audit assumptions or evidence.
no external APIs, databases, or connections required. this skill runs on evidence, reasoning, and analysis you perform in real time.
run the six phases below in a single response unless the user requests step-by-step pacing. each phase has explicit inputs and outputs.
input: the user's claim or question, any provided context.
steps:
output: a one-paragraph setup block with focus restated, pass sequence declared, and gaps named.
input: the focus statement from setup.
steps:
output: one to three sentences, clearly labeled, with factual vs normative split visible.
input: evidence provided by the user, your knowledge, cited sources.
steps:
[CITED] or [MISSING].[CITED] items: name the basis (user text, repo, doc, link, study, expert testimony, etc.) and one line on strength or limits (no extra strength tags; keep it factual).[MISSING] items: state plainly that no traceable basis exists yet or was requested but not available.output: a bulleted list with [CITED] or [MISSING] tags and basis notes for each claim.
input: the focus, information phase output, and the chain of reasoning.
steps:
output: a numbered or bulleted list of assumption pairs, value premises separated when they apply.
input: the focus, information phase, and assumptions.
steps:
output: a section tracing the argument chain, flagging structural gaps and leaps, followed by a compact bias and fallacy pass (only applicable items, or "none identified").
input: the focus, information, assumptions, and reasoning.
steps:
output: two to four numbered alternatives, each in one to two sentences, with a brief note on what would confirm or refute it.
input: all prior phases.
steps:
[CITED] to [MISSING] evidence and the solidity of the reasoning chain.output: a judgment statement, value-vs-evidence split if needed, one plain strength sentence, and falsifiers section.
if the user requests step-by-step pacing: pause after each phase and wait for the user to signal "next phase" rather than running all six in one response.
if essential context is missing: ask up to 3 clarifying questions in the setup block, then proceed with the rest of the phases using reasonable working assumptions. do not halt the entire review.
if the user halts the review early (e.g. "stop after Assumptions"): end at that phase. do not proceed to Reasoning or Conclusion unless asked.
if the user is emotionally fused with a position: name the position neutrally in the setup and continue the phase plan without editorial comment. intellectual honesty requires naming strong opposing evidence even when the user has skin in the game.
if the focus mixes factual and normative claims: surface value premises explicitly in the Assumptions phase so the chain separates "what the evidence shows" from "what follows if we adopt this value."
if a claim rests on an ungrounded premise (needed but never established in Information): flag it in Reasoning as a structural gap, do not assert it as true in Conclusion.
if no biases or fallacies apply: state that plainly in one line in the Reasoning phase rather than padding the list.
each skill run produces a response with these sections in order:
[CITED] (basis + limits) or [MISSING] tags.all sections are labeled with clear headers. no facts are smuggled into Conclusion that were not grounded in prior phases. no [CITED] claims lack a stated basis. no assumptions lack an "If false" consequence.
the user knows the skill worked when:
[CITED] with a named basis and one limit, or [MISSING] with no ambiguity.[CITED] (basis + limits in-bullet) or [MISSING]credits: original skill design by ysskrishna (clawhub). enriched and standardized to Implexa quality.