OpenSkillIndex← back to search
claude-code

DeepTutor deep-research

Multi-agent research reports on any topic.

install
source · Clone the upstream repo
git clone https://github.com/HKUDS/DeepTutor
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/HKUDS/DeepTutor "$T" && mkdir -p ~/.claude/skills && cp -r "$T/deeptutor/tutorbot/skills/deep-research" ~/.claude/skills/hkuds-deeptutor-deep-research && rm -rf "$T"
manifest: deeptutor/tutorbot/skills/deep-research/SKILL.md
source content

Deep Research

Use the 

exec
tool to invoke DeepTutor's multi-agent research pipeline for comprehensive topic analysis.

When to Use

  • User needs a research report or literature review
  • User wants a comparison of approaches, frameworks, or technologies
  • User asks for a learning path or structured overview of a field
  • The topic requires aggregating information from multiple sources

Command

deeptutor run deep_research "<topic>" --format json -l <lang> --config-json '<json>'

Config JSON Fields

FieldValuesDescription
mode
notes
, 
report
, 
comparison
, 
learning_path
Output format
depth
quick
, 
standard
, 
deep
Research thoroughness
sources
["kb", "web", "papers"]
Information sources to use

Examples

Standard report from web + papers:

deeptutor run deep_research "Recent advances in attention mechanisms" --format json -l en --config-json '{"mode":"report","depth":"deep","sources":["papers","web"]}'

Quick comparison:

deeptutor run deep_research "Rust vs Go for backend services" --format json -l en --config-json '{"mode":"comparison","depth":"quick","sources":["web"]}'

Learning path from knowledge base:

deeptutor run deep_research "Machine learning fundamentals" --format json -l zh --config-json '{"mode":"learning_path","depth":"standard","sources":["kb"]}' --kb ml-textbook

KB + web hybrid:

deeptutor run deep_research "Agentic RAG vs traditional RAG" --format json -l zh --config-json '{"mode":"comparison","depth":"deep","sources":["kb","web"]}' --kb my-papers

Important

  • All three config fields (
    mode
    , 
    depth
    , 
    sources
    ) are required.     Always use 
    --config-json
    to pass them together.
  • depth=deep
    can take several minutes — use 
    timeout=300
    or higher with the 
    exec
    tool.
  • Always pass 
    --format json
    to get parseable NDJSON output.
  • Parse lines with 
    "type": "content"
    and concatenate for the full report.