Deep Researcher
A Claude Code plugin for multi-dimensional deep research using parallel subagents. Combines live web search with codebase analysis into synthesized reports with confidence tiers, conflict zones, and styled HTML visualizations.
Introduction
Deep Researcher automates rigorous research inside Claude Code. Instead of spending hours opening dozens of browser tabs or grepping through unfamiliar code, you describe what you need to know — the plugin scopes the question, spawns specialized subagents to investigate each dimension in parallel, cross-verifies findings against source-quality tiers, and delivers a structured report with citations, insight extraction, and flagged disagreements.
It is designed for tech stack evaluations, competitive landscape mapping, pre-build research, and rapid codebase onboarding. You stay in control via human checkpoints at key decision points; the plugin handles the parallel execution, stale-claim detection, and synthesis.
What it does
Deep Researcher investigates a topic across multiple analytical dimensions simultaneously. Each dimension is researched by an independent subagent, then findings are cross-verified, classified by confidence, and assembled into a structured report. The result is a set of Markdown report pages and, optionally, a self-contained HTML report with Mermaid diagrams and interactive navigation.
Components
deep-researcher agent — A multi-role research agent that switches between investigator, synthesizer, and reviewer modes. It performs web searches with source-quality tiering, navigates codebases with ripgrep and LSP, and writes checkpointed findings so nothing is lost to context limits./deep-researcher:brainstorm — Primes and focuses a research topic before the deep dive. Through conversational scoping, it surfaces hidden assumptions, evaluates 2–3 alternative framings, and produces a Research Brief that pre-fills dimension decomposition. This ensures the parallel research hits the right targets instead of scattering across irrelevant tangents./deep-researcher:deep-research — Orchestrates the full seven-phase pipeline: orient → landscape scan → dimension decomposition → parallel deep dive → cross-verification → synthesis → report delivery. Time-boxed subagents, human checkpoints, and automatic resume keep long research sessions manageable./deep-researcher:deep-research-visualize — Converts a completed Markdown report into a styled, self-contained HTML file with Mermaid diagrams, themed typography, and interactive navigation. Four themes cover business, technical, academic, and dark-mode reading contexts.
It can also answer targeted codebase questions — "how does X work", "where is Y configured" — by navigating the current project's source tree directly.
Report Example
Editorial theme, sectioned report
Installation
Prerequisites
- Claude Code (latest version recommended)
- Node.js 22 LTS
- ripgrep (
rg) — required for codebase search. Install with brew install ripgrep (macOS), apt install ripgrep (Linux), or cargo install ripgrep
python3 and perl — used in pipeline validation steps. Both ship with macOS and most Linux distributions. If missing, the plugin loads but some pipeline validation steps will skip
The plugin warns at startup if any of these are missing.
From Marketplace (recommended)
Add this repository as a marketplace in Claude Code, then install the plugin:
# Add the marketplace (once)
claude plugin marketplace add sdkks/deep-researcher-visualized
# Install the plugin
claude plugin install deep-researcher@sdkks-deep-researcher
Or from within a Claude Code session:
/plugin marketplace add sdkks/deep-researcher-visualized
/plugin install deep-researcher@sdkks-deep-researcher
After installing, run /reload-plugins to load the skills and agent.
From Source (for development or manual install)
git clone https://github.com/sdkks/deep-researcher-visualized.git
claude plugin install ./deep-researcher-visualized
Or load it for a single session without installing:
claude --plugin-dir ./deep-researcher-visualized
Usage
/deep-researcher:brainstorm
/deep-researcher:brainstorm <topic-or-idea>
A lightweight conversational exploration that runs before /deep-researcher:deep-research. Use it when you have a topic but aren't sure how to scope it or which angles matter. Brainstorm probes intent one question at a time, proposes 2-3 scope framings with trade-offs, and produces a Research Brief — a structured block you can hand directly to /deep-researcher:deep-research so the Phase 1 orientation dialog is skipped.
Example:
/deep-researcher:brainstorm "local-first databases for a collaborative note-taking app"
