deep-research

claude-deep-research

Official-docs-first deep research for Claude Code.

---

Skills

Skill	Command	What It Does
deep-research	/deep-research [topic]	Iterative multi-source research producing comprehensive Obsidian reports with full citations
research-mode	/research-mode	Anti-hallucination constraints with source authority hierarchy and citation requirements

Agents

Agent	What It Does
deep-research	Executes the iterative research loop -- locates official docs, surveys secondary sources, tracks state across iterations, writes final report to Obsidian

Install

/plugin marketplace add jl-cmd/claude-deep-research
/plugin install deep-research@claude-deep-research

Or locally:

git clone https://github.com/jl-cmd/claude-deep-research.git
claude --plugin-dir ./claude-deep-research

Official-Docs-First Methodology

The core principle: when researching any external tool, library, API, or protocol, always locate and read the official vendor/creator documentation before consulting secondary sources.

• Official docs are tier 1 -- blog posts, tutorials, and community content are tier 5
• If no official docs exist, that is explicitly stated as a finding
• When sources conflict, official docs win
• For local project code, the codebase itself is the authority

/deep-research -- Iterative Research Pipeline

Two-phase pipeline:

Phase 1 (Interactive): Scopes the research through 1-3 clarifying questions, builds a structured research brief, and identifies known official documentation sources.

Phase 2 (Agent): The deep-research agent executes iterative research:

Phase	Iterations	Focus
Early	1-3	Official vendor/creator docs -- locate, deep read, extract quotes
Middle	4-8	Secondary source deep dives -- fill gaps, cross-reference against official docs
Late	9+	Synthesis and gap-filling -- resolve contradictions, compile report

Sources are tagged [official] or [secondary] throughout. The final report notes whether official documentation was available.

Depth options:

Depth	Iterations	Target Sources
Quick overview	8	5-10
Standard	15	15-20
Exhaustive	25	25+

Output is written to Obsidian with full YAML frontmatter, inline citations, and a numbered bibliography. Falls back to conversation output if Obsidian MCP is unavailable.

/research-mode -- Anti-Hallucination Constraints

Activates three constraints that stay active until explicitly exited:

1. Say "I don't know" -- no credible source, no claim
2. Cite everything -- every claim needs a specific source with URL, named paper, or official docs
3. Direct quotes -- extract actual text before analyzing, ground responses in word-for-word quotes

Plus a Source Authority Hierarchy (highest to lowest):

1. Official vendor/creator documentation
2. Files in the current project
3. Academic papers, named researchers
4. Reputable external sources with URLs
5. Blog posts, tutorials, community content

Prerequisites

• Claude Code 1.0.33+
• For Obsidian output: an Obsidian MCP server connected to Claude Code (optional -- falls back to conversation output)

Plugin Structure

claude-deep-research/
├── .claude-plugin/
│   ├── plugin.json
│   └── marketplace.json
├── skills/
│   ├── deep-research/
│   │   └── SKILL.md
│   └── research-mode/
│       └── SKILL.md
├── agents/
│   └── deep-research.md
├── README.md
├── LICENSE
└── .gitignore

Sources

• Anthropic - Reduce Hallucinations
• Plugin creation docs
• Skills authoring docs

License

MIT