Deep Research

Problem Being Solved

A typical search answer often stops at keyword lookup plus summary stitching: one angle, few sources, predictions presented as facts, and little context about year, scope, sample, or evidence chain. That makes market analysis, competitive analysis, industry research, and paper/open-source reviews look complete but hard to use for decisions.

How the Skill Works

vadeepresearch turns research into an executable workflow: it parses Topic, Scope, Depth, Time Range, and Output, then sets source and angle requirements by quick, standard, or deep level. Key steps include:

• Broad exploration: build a topic map and a list of research angles.
• Targeted deep dive: generate queries per angle and read high-value source pages or key sections.
• Multi-source validation: cover official, academic, industry, news, and open-source sources, then mark credibility.
• Evidence extraction: normalize findings into fields such as Source, Claim, Data, Context, and Confidence.
• Conflict handling: present inconsistent figures side by side instead of forcing one definitive conclusion.
• Report output: default to Markdown, then prompt for PDF, Word, PPT, or other follow-up formats.

It also recommends using helper scripts such as orchestrator.py, research_engine.py, analysis_engine.py, and report_generator.py for orchestration, retrieval, analysis, and persistence, but only when those tools are actually available in the runtime. Hard-coding unavailable services is not acceptable.

Boundaries and Caveats

Best for tasks that need external facts, recent data, citations, and structured reporting; not for pure rewriting, translation, polishing of supplied text, or requests that explicitly forbid retrieval. Keep key facts double-sourced, annotate year/geography/sample/statistical scope, do not let weak fresh sources override authoritative older ones, and fall back to manual analysis with a clear limitation note if a script fails.