agentleFS
TheGoat395TheGoat395/Codex-Skills/skills/critical-thinking/SKILL.md

critical-thinking

Skill136 starsChanged 5 months agoMITView on GitHub

Judge claims through evidence and alternatives.

22 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 20 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Consequential judgment or live competing explanations: read challenge protocol; consequential work also needs verification and confidence.

    line 15
  2. Distinctive:

    Argument audit: Do the reasons support the conclusion?

    line 37
  3. Very specific:

    Use evidence-reconciliation when source authority, version, or contradiction is central.

    line 60
  4. Distinctive:

    Stop when the conclusion survives the strongest material countercase, decisive claims have appropriate evidence, remaining uncertainty is …

    line 76

“Common” is measured: kept by at least 6 in 10 of the 20 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Critical Thinking

Produce a defensible judgment, not reflexive agreement, reflexive skepticism, or a catalog of fallacy names. Critical thinking combines disposition, task-specific analysis, argument reconstruction, evidence evaluation, inference, explanation, and metacognitive monitoring.

Route the actual judgment

Reconstruct the exact position fairly: conclusion, evidence, assumptions, scope and rivals. Test the strongest alternative with discriminating evidence; do not invent doubt or average incompatible claims.

  • Argument, explanation, causal, forecast, normative, decision or source audit with material reasoning: read argument review protocol; consequential or multi-branch arguments also need argument and evidence mapping.
  • Consequential judgment or live competing explanations: read challenge protocol; consequential work also needs verification and confidence. Correlated self-review is not proof.

Bounded claim audits use the core checks directly. Formal entailment belongs to logical-thinking; missing evidence to deep-research; source conflicts to evidence-reconciliation; assertion calibration can use honesty if separately installed. Without it, use the confidence and delivery rules below; the public collection has no required dependency on that skill.

Hold the thinking disposition

For material judgments:

  • seek the most accurate answer even when it conflicts with the initial framing;
  • remain open to alternatives without manufacturing false balance;
  • work systematically enough that decisive claims and assumptions are visible;
  • distinguish confidence in the method from confidence in the conclusion;
  • suspend judgment when the evidence cannot discriminate between live alternatives;
  • revise when better evidence or a successful counterexample requires it.

This is an operating discipline, not a personality performance. Do not add debate or doubt to a routine answer when it cannot improve correctness.

Classify the reasoning task

Identify the central task before choosing checks:

  • Claim audit: Is a factual statement supported?
  • Argument audit: Do the reasons support the conclusion?
  • Explanation audit: Which account best explains the observations?
  • Causal audit: Does the evidence support association, intervention, or counterfactual claims?
  • Forecast audit: What outcome is most probable, under which conditions?
  • Normative audit: Which values, duties, or standards control the judgment?
  • Decision audit: What action best fits the objective, uncertainty, cost, and reversibility?
  • Source audit: What does documentation or another authority actually establish?

Select and adapt only the reasoning modules that fit the task. A fixed checklist applied to every problem creates noise and can hide the decisive issue.

Evaluate evidence at the claim level

For each decisive claim, determine:

  • what observation would support or weaken it;
  • whether the inspected evidence addresses that exact observation;
  • source authority, directness, independence, currency, method, scope, and incentives;
  • whether missing evidence is expected and diagnostic or merely unavailable;
  • whether the evidence supports the conclusion itself, a premise, or only plausibility;
  • what stronger evidence could realistically be obtained.

Quality and diagnostic value matter more than source count. Several retellings of one origin are one evidentiary line. A controlling document may settle a rule without proving implementation; an observation may establish behavior without proving the governing rule.

Use evidence-reconciliation when source authority, version, or contradiction is central. Use deep-research when decisive evidence has not yet been gathered.

Calibrate and stop

Choose a judgment state that matches the evidence:

  • Established within scope
  • Well supported
  • Plausible but not discriminated
  • Unresolved
  • Weakly supported
  • Contradicted
  • Not assessable from available evidence

State the scope and decisive reason. Numeric confidence is optional and must be tied to assumptions or calibration evidence, not decorative precision.

Stop when the conclusion survives the strongest material countercase, decisive claims have appropriate evidence, remaining uncertainty is visible, and further work is unlikely to change the judgment enough to justify its cost.

Deliver

Lead with the judgment. For consequential work, report the decisive support, weakest material assumption, strongest alternative, verification performed, confidence state, and what would change the result. Show the argument map or evidence table when it improves inspection, but do not dump private scratch reasoning or manufacture a long debate.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.