Run evidence-first research investigations in Claude Code with Deepdive
Install Deepdive as a Claude Code skill for structured investigations with plans, source files, claim tracing, dissent checks, and reusable research reports.
npx skills add agentskillexchange/skills --skill run-evidence-first-research-investigations-in-claude-code-with-deepdive
Use Deepdive when a Claude Code operator needs a repeatable research investigation rather than a one-shot answer. The workflow turns a prompt such as investigate a tradeoff or validate this hypothesis into a staged pipeline: reframe the question, choose the research genre, write a plan, search across configured channels, save sources with quotes and provenance, triangulate claims, run adversarial checks, and produce a report that can be revisited later. Invoke this instead of using normal web search or the product chat when the answer must survive audit, comparison, or later decision review. The scope boundary is a Claude Code native research skill with file-backed evidence and verification steps; it is not a generic deep-research product card, a search engine, a citation library, or a broad agent framework.
What this skill actually does
Inputs and prerequisites: Claude Code with Skills enabled, git, local filesystem access, and optional configured search/API sources for richer investigations.
Setup notes: Clone the repository into the Claude Code skills directory with git clone https://github.com/Socialpranker/deepdive.git ~/.claude/skills/deepdive. In Claude Code, invoke it with requests such as investigate a topic, validate a hypothesis, or compare technical options, then review the generated plan, sources, findings, and final report files.
Source and verification boundary: use https://socialpranker.github.io/deepdive/ as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Claude Code workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.