Research & Source Credibility Gate skill

AI-generated research often suffers from hallucinations, circular citations, and outdated data. This skill acts as an adversarial auditor for research briefs, report drafts, and AI outputs. It does not just summarize content; it aggressively challenges claims by checking source support, recency, and framing. By categorizing findings into TRUST, VERIFY, or REJECT, it identifies exactly where a document's credibility breaks down.

What it does

• Claim extraction isolates discrete numerical, causal, and superlative statements for individual scrutiny.
• Citation auditing verifies if cited sources actually support the specific nuances of a claim, such as geography or population size.
• Recency assessment flags time-sensitive data like market figures or regulations that may be based on stale evidence.
• Framing analysis detects cherry-picked examples, missing denominators, or correlation-causation fallacies.
• Actionable verification provides specific steps to resolve evidence gaps rather than just pointing out problems.

How it works

1. Intake the document and identify the target audience, date sensitivity, and available source links.
2. Audit the material using the adversarial TRUST/VERIFY/REJECT framework to categorize every major claim.
3. Generate a structured report featuring exact quotes, severity ratings, and a source quality table.
4. Execute the priority verification checklist to finalize high-stakes sections of the research.

Frameworks & tools

This skill is framework-agnostic and designed for LLMs with browsing capabilities or RAG workflows. It excels when provided with raw text, Markdown reports, or lists of URLs.

Why this beats prompting it yourself

Standard prompts often suffer from confirmation bias, where the AI tries to justify the user's text. This skill is hard-coded for adversarial behavior, requiring specific evidentiary proof before a claim is marked as TRUST. It enforces a strict taxonomy that prevents the agent from making vague "this looks correct" statements.

Use cases

• Auditing AI-generated blog posts for hallucinated statistics or broken links.
• Reviewing white papers for one-sided framing and missing counter-arguments.
• Fact-checking market research reports before they reach executive stakeholders.
• Validating technical documentation against original primary source specifications.

Known limitations

Does not guarantee exhaustive fact-checking or replace subject matter expertise. Accuracy is limited by the accessibility of cited URLs and the quality of the LLM's browsing tool.