deep-research

Executes a five-step research workflow producing cited final answer files.

1.3k|142|Updated Jun 1, 2026
One-click install
npx skills add https://github.com/MiniMax-AI/minimax-code --skill deep-research-minimax-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: deep-research
Source: https://github.com/MiniMax-AI/minimax-code/tree/main/packages/local-runtime/assets/skills/deep-research
Command: npx skills add https://github.com/MiniMax-AI/minimax-code --skill deep-research-minimax-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Complex, open-ended questions often get shallow one-shot answers that lack verified sources, clear direction, and traceable reasoning. This Skill enforces a disciplined five-step research process so the final answer is grounded in searched, cross-checked evidence with proper citations. ## Core Features & Use Cases - Five-Step Research Pipeline: Sequentially executes background confirmation, direction judgment, deep analysis and planning, plan-driven search and verification, and final writing, with each step gated by an explicit read of its prompt file. - Source-Grounded Final Answers: Requires searched facts with URLs, conflict flagging, and citation markers, and writes the result to a unique final_turn_XXX.md file in a per-run workspace. - Use Case: Ask a question like "What strategy should a specific AI coding tool adopt against its competitors?" and the Skill will gather background facts, lock the actor and perspective, plan the research, verify claims across sources, and deliver a structured cited report. ## Quick Start Ask the agent to run deep research on your question, for example: use deep research to analyze the competitive landscape of open-source coding agents and write a cited report.

Frequently Asked Questions about deep-research

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a deep research task with cited sources?

Invoke the deep-research skill with your question. It executes five ordered steps: background fact gathering, direction judgment, analysis and planning, plan-driven search and verification, then writes a cited final answer to a final_turn_XXX.md file.

What kinds of questions is deep research best suited for?

It suits complex, open-ended questions needing search, cross-checking, and citations, such as market analysis, competitor research, trend judgment, and fact verification. Simple factual questions or code-editing tasks should not use it.

Can I skip steps in the deep research workflow?

No. Each of the five step prompt files must be explicitly read and executed in order. Skipping steps, reading later steps early, or jumping from planning directly to writing is treated as an invalid trace.

Where does the deep research final answer get written?

Each invocation creates a run workspace in a writable scratch location and writes exactly one Markdown file named final_turn_XXX.md, using the smallest unused three-digit number. No intermediate Markdown files are allowed in the workspace.

Does deep research handle time-sensitive questions?

Yes. It judges whether the question depends on current status, recent changes, or latest data, then actively searches and verifies time-sensitive facts, stating the relevant time point in the final answer when it affects the conclusion.