writing-credibility-auditor

Overview

This skill performs a deep credibility audit on any user-supplied piece of writing. It applies four independent analytical lenses — logical fallacy detection, unsupported claim identification, weasel word scanning, and misleading statistics recognition — then synthesises the findings into a structured, actionable Credibility Report. No external APIs or search tools are required. All analysis is powered by the agent's own reasoning applied against the precise detection frameworks defined in this skill. Target users include researchers, journalists, students, editors, and anyone who needs to critically evaluate the quality of an argument before acting on, publishing, or citing it.

When to use this skill

Trigger on any of these user intents:

• "Check this article / essay / report for logical fallacies"
• "Audit this text for credibility / fact-check this writing"
• "Find unsupported claims in this passage"
• "Does this use weasel words or vague language?"
• "Are these statistics misleading or being misused?"
• "Review the reasoning in this argument"
• "Is this source trustworthy / well-cited?"
• "Flag any citation problems in this text"
• "Check if correlation is being confused with causation here"
• User pastes a block of text and asks "What's wrong with this?" or "Is this solid?"

Do NOT trigger this skill for:

• Requests to generate or write new content (use a writing skill instead)
• Plagiarism checks or copyright detection
• Stylistic or grammar-only editing requests
• Sentiment analysis or tone-only reviews
• Fact verification that requires live web search (this skill audits reasoning structure and language, not live factual accuracy)
• Summarisation tasks where no critical review is requested

Instructions

Step 0 — Intake and scope confirmation

1. Confirm the user has provided a text to audit. If not, ask:

   "Please paste the text you'd like me to audit for credibility."

2. If the text exceeds approximately 2 000 words, inform the user:

   "This text is long. I'll audit it thoroughly — it may take a moment. If you'd like, you can also specify a section to focus on."

3. Proceed without further clarification unless the user specifies a focus area.

Step 1 — Logical fallacy scan (apply all 12 definitions)

Read the text carefully and check for each of the following fallacies. For every instance found, note the exact excerpt, the fallacy name, and a one-sentence explanation of why it qualifies.

F-01 · Ad Hominem Definition: Attacking the person making an argument rather than the argument itself. Detection pattern: The text dismisses or attacks a claim by criticising the character, background, motives, or identity of its source rather than evaluating the evidence.