Deep Research

Problem

When a topic spans multiple domains, sources are fragmented, and claims conflict, plain search tends to return scattered material without clear citations, recency, or evidence weighting. This skill targets systematic research and evidence tracking: it converts a broad request into verifiable sub-questions and produces a report that can be audited.

Workflow

• Five phases: planning, source gathering, cross-validation, report generation, self-review, delivery.
• Question decomposition: uses 5W1H to split the topic into 3–5 sub-questions, then defines Chinese and English keywords and source types.
• Multi-source search: for each sub-question, it performs at least 3 independent searches, aiming for 8–12 distinct sources from papers, official data, industry reports, and reputable media, while recording URL, date, and key claims.
• Cross-validation: major claims need at least 2 independent sources; conflicting perspectives are compared with evidence strength and recency noted.
• Report output: defaults to Markdown with numbered [1] citations, tables for structured data, and support for user-provided local files as primary sources.

Boundaries

It fits research summaries, technology or industry trend analysis, and policy or competitive research. It is not a substitute for real-time trading decisions, unauthorized analysis of sensitive data, or professional legal and medical advice. For very broad topics, split the work into sub-reports; when sources conflict, judge by evidence quality instead of adopting a single narrative.