How to use the Scientific Critical Thinking agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder of instructions and reference files that an AI agent can load when a task matches the skill's description. This skill is for analysing research rather than producing it. Its SKILL.md tells the agent to fix the research question and unit, extract the reported evidence before judging it, assess design and analysis, choose the right appraisal framework, and write a critique that points to specific locations in a paper.

• Methodology critique: design, controls, confounding, and whether the method can answer the question asked.
• Bias detection: selection, measurement, publication and cognitive biases.
• Statistical evaluation: power, multiplicity, p-value misuse and effect sizes.
• Evidence quality: study hierarchy, replication and strength of inference.
• Logical fallacies that recur in scientific argument, and claim evaluation that separates what was shown from what is asserted.

Install the scientific-critical-thinking skill

Download the skill's ZIP from this page and unzip it. You can install it in one of two ways. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. Either way, the skill folder should contain SKILL.md and the references/ folder.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need a special command. Ask in plain language about a study, a claim or an evidence question. The skill's description covers experimental design validity, biases and confounders, statistical interpretation, evidence grading (GRADE, Cochrane Risk of Bias) and teaching critical analysis. Paste or attach the paper, or give the exact result you want checked, and say what you want the agent to judge.

• "Critically appraise this trial. Does the design support the causal claim in the abstract?"
• "Grade the certainty of the evidence for this outcome using GRADE."
• "Check whether this study's statistics support its conclusion. Point to the specific tables or Methods sections."
• "Explain the main biases in this cohort study and which ones could change the result."
• "Help me build a critical analysis exercise for my students using this paper."

What you need and what the skill does on your machine

The skill lists no required dependencies. It does not run scripts and does not use the network for its analytical guidance, so it works from the text you provide and its reference files. Its allowed tools are Read, Write and Edit.

One optional feature is different. The skill can produce illustrative diagrams through a separate scientific-schematics skill, which needs an OPENROUTER_API_KEY and outbound access to OpenRouter. Figures are only created when you explicitly ask for one.

• Requirements listed: none.
• Runs scripts: no, for the core analytical skill.
• Uses network: no, for the core analytical skill.
• Optional figures: need OPENROUTER_API_KEY and internet access to OpenRouter, and send your prompt to that third-party service.

Safety notes before you rely on the output

The optional diagram command in the skill is run from the repository root with OPENROUTER_API_KEY set. The skill notes that its paid generation was not exercised in testing, so check the output and review every generated label. Do not include unpublished or sensitive details in a figure prompt unless sending them to OpenRouter is appropriate for your project.

The skill's reference examples are teaching material, not validated advice. Do not use the output for medical or patient-specific decisions. The skill also cannot fully verify literature sources without network access, so check any citation yourself before you publish it.

• Do not use it for medical or patient-specific decisions.
• Verify any citations it suggests; it cannot fully check its sources offline.
• Review generated figures before use, and keep sensitive material out of image prompts.
• The risk-of-bias guidance points to current Cochrane and tool versions; confirm them against the primary sources.

Tips for better results

Give the agent the exact result you care about, including the outcome, comparison and time point. The skill's own guidance says risk-of-bias judgments should be made per result, not as one score for the whole paper, so a precise request produces a more useful critique. Ask for the reasoning behind each concern, since the skill is designed to report location, observation, consequence and a feasible remedy.

• Name the specific outcome and comparison you want assessed.
• Ask the agent to separate fatal flaws from minor limitations.
• Ask for quoted passages or table references so you can check each point.
• For a formal peer review report, use the separate peer-review skill instead.