Deep Research Example

A copy-and-customize research pattern for multi-source synthesis with citations preserved. Installable version: the you-research skill.

View as MarkdownOpen in Claude

A portable Agent Skill that decides when a user needs thorough web research and routes retrieval through the standard You.com MCP server. Use it for open-ended briefings, deep dives, and cited synthesis across multiple sources.


What You’ll Build

A SKILL.md package that teaches an agent when a request needs multi-source research instead of a single lookup. The skill sends the user’s question and relevant thread context to the standard You.com MCP server’s you-research tool and preserves inline citations in the final answer.

When It Fires

The task needs synthesis across several sources, not one fact. Trigger signals: “research”, “briefing”, “deep dive”, “literature on”, open-ended “what’s known about”. For single-fact lookups, a plain search is the better fit.


Prerequisites

Install the Skill

  1. Connect your agent harness to the You.com MCP server—see the MCP server guide.
  2. Save the skill below as SKILL.md in your harness’s skills directory (or download it).
  3. The agent loads the skill and invokes you-research whenever the trigger fires.
Download SKILL.md

Grab the ready-to-use skill file.

The Skill

SKILL.md
---
name: deep-research
description: Use when the user needs a thorough, multi-source answer with citations - "research X", "give me a briefing on Y", "what's the state of Z". Not for single-fact lookups; use a plain search for those.
---
# Deep research (You.com)
## When This Fires
The task needs synthesis across several sources, not one fact. Signals:
"research", "briefing", "deep dive", "literature on", open-ended "what's
known about".
## How to Retrieve
1. Call the standard You.com MCP server's `you-research` tool.
2. Pass the user's question plus any relevant context already in the thread.
3. Return the synthesized answer with inline citations preserved. Never strip
sources.

Example Output

# Deep research: agentic browser automation
## Summary
Agentic browser automation is moving from brittle scripts toward goal-directed
agents that can inspect pages, use tools, recover from DOM changes, and cite
the sources behind their conclusions.
Sources:
[1] Vendor docs - browser automation architecture
[2] Recent coverage - enterprise adoption patterns

Next Steps