deep-research

This skill is a multi-source deep-research workflow from the ECC collection. It solves the problem of shallow, uncited answers by orchestrating web research through the firecrawl and exa MCP tools into thorough, source-attributed reports. It is designed for topics that require synthesis across many sources — competitive analysis, technology evaluation, market sizing, and due diligence.

The single-file SKILL.md defines a six-step workflow: clarify the goal, decompose the topic into 3–5 sub-questions, execute multi-source searches (firecrawl_search, web_search_exa, web_search_advanced_exa) aiming for 15–30 unique sources, deep-read a handful of key sources in full (firecrawl_scrape, crawling_exa), synthesize into a structured cited report, and deliver either inline or to a file. It documents the MCP tool requirements and where to configure them, offers a report template with executive summary and methodology sections, describes optional parallelization via Claude Code's Task tool, and enforces quality rules: every claim needs a source, cross-reference single-source claims, prefer recent sources, acknowledge gaps, and never hallucinate. A prominent "drift-prone" note warns that MCP tool names and quotas change and should be verified.

The target users are analysts, founders, engineers, and decision-makers who need evidence-backed research with citations. It is a read-oriented workflow: it searches and scrapes public web content through official MCP tooling and writes a report; it performs no destructive actions and exfiltrates no local data. Its capability depends entirely on the user having firecrawl and/or exa MCP servers configured.