OpenSkillIndex← back to search
claude-code

DeepTutor deep-solve

Multi-stage problem solving via DeepTutor (plan → reason → write).

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 "$T/deeptutor/tutorbot/skills/deep-solve" ~/.claude/skills/hkuds-deeptutor-deep-solve && rm -rf "$T"
manifest: deeptutor/tutorbot/skills/deep-solve/SKILL.md
source content

Deep Solve

Use the 

exec
tool to invoke DeepTutor's multi-stage solver for complex math, science, or engineering problems.

When to Use

  • User asks to prove something, derive a formula, or solve a hard problem
  • The problem requires structured multi-step reasoning
  • A direct answer would be insufficient or unreliable

Command

deeptutor run deep_solve "<problem description>" --format json -l <lang>

Options

FlagDescription
-l <lang>
Response language: 
en
or 
zh
-t rag
Enable RAG to ground in knowledge base
-t web_search
Enable web search
-t code_execution
Enable code verification
-t reason
Enable dedicated reasoning
--kb <name>
Knowledge base to use with RAG
--config detailed_answer=false
Return a concise answer

Examples

Prove a theorem:

deeptutor run deep_solve "Prove that for any positive integer n, n³ - n is divisible by 6" --format json -l en

Solve with knowledge base:

deeptutor run deep_solve "Derive the Euler-Lagrange equation" --format json -l en -t rag --kb physics-textbook

Solve with code verification:

deeptutor run deep_solve "Find all roots of x⁴ - 5x² + 4 = 0" --format json -l en -t code_execution

Important

  • Deep solve can take over a minute — use 
    timeout=300
    with the 
    exec
    tool.
  • The 
    --format json
    flag outputs NDJSON stream events. Parse the 
    content
    field from lines with 
    "type": "content"
    to get the final answer. Concatenate all content events for the full response.