Critical analysis

Skill SylphxAI/skills/skills/critical-analysis

Analyze an uncertain material question while multiple credible hypotheses or explanations remain, using assumptions, disconfirming evidence, base rates, Bayesian updates, and calibrated conclusions. Use for root-cause, architecture, policy, or ambiguous-evidence analysis. Do not use for fixed-predicate candidate acceptance or as the final owner of a durable option decision.From its SKILL.md

Install

npx -y skills add SylphxAI/skills --skill critical-analysis

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

24 days oldThe repository was created 24 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.1 KB, 538 tokens by cl100k_base, as published. Nobody here has run it

Critical Analysis

Investigate uncertainty without confusing fluency, consensus, or exhaustive prose with truth. Read references/critical-analysis-method.md for the method and templates. Read references/bayesian-evidence-updates.md when evidence should update ranked hypotheses quantitatively or semi-quantitatively.

Method

Frame the exact question, decision relevance, boundary, and deadline.
Separate observations, inferences, assumptions, and unknowns.
Generate the smallest decision-relevant bounded set of materially distinct hypotheses or contributing explanations, including the status quo and a credible opposing explanation; retain missing-hypothesis risk as residual uncertainty.
Identify evidence expected under each hypothesis and prioritize evidence that discriminates between them.
Search for disconfirmation, missing causes, base rates, incentives, survivorship, selection effects, and reversible alternatives.
Run the risk-matched challenge method: assumptions check, premortem, competing-hypothesis matrix, devil's advocate, or independent perspective.
Update the ranking and express a calibrated conclusion, alternatives still alive, and what would change the answer.

Do not mechanically enumerate remote possibilities. Include a possibility only when it could change the conclusion, action, risk floor, or evidence plan.

Output

Produce a Critical Analysis Brief:

Question and stakes
Observed facts and evidence quality
Competing hypotheses or explanations
Discriminating evidence
Strongest case for and against the leading conclusion
Assumptions and failure paths
Conclusion and calibrated confidence
What would change the conclusion

Boundaries

This skill diagnoses and challenges; decision-quality-standard owns final material option selection and durable decision tradeoffs.
Use design-space-exploration to generate materially different solution options before selection.
Use evidence-and-claims-standard to adjudicate whether an individual claim is supportable.
Use a domain skill for technical, commercial, security, legal, or product requirements.
Multiple agents may provide independent perspectives, but agreement is not evidence and disagreement is not automatically uncertainty.

Routing examples

Use for “critical think why agents keep doing half a migration,” “compare the credible causes,” “challenge this architecture thesis,” and “what evidence would distinguish these explanations?” Do not route a settled implementation task with a known cause and acceptance test.