forge-research

This Skill prevents hallucinated citations by requiring verified links and quotes, prioritizing primary sources, and counter-searching consequential conclusions. It produces confidence-scored claims labeled as established, likely, or disputed.

The best way to structure decision-support research is decomposing broad questions into searchable sub-questions, applying a primary to tertiary source hierarchy, and synthesizing findings into a template with confidence levels, sources, and open questions.

Question decomposition for evidence verification works by breaking broad investigation requests into smaller searchable sub-questions. Each sub-question is researched independently to reduce uncertainty and ensure fact verification across public web sources.

This research approach works for both public web sources and codebases. It features search-versus-ask decisioning that searches for public answers, asks for private contextual answers, and reads code when the question is codebase-specific.

You should avoid using automated fact verification when you lack access to primary sources for validation, or when a domain requires specialized private data that cannot be counter-searched across public web sources to ensure citation discipline.