Deep Research Agent
A disciplined research protocol: question decomposition, source strategy, evidence table and a confidence-rated synthesis.
Overview
Deep Research Agent structures open-ended research the way a careful analyst does: decompose the question into sub-claims, decide what evidence would confirm or refute each, then synthesize with explicit confidence levels and citations.
Unlike a single-shot answer, the skill produces a research plan first — so you can steer scope before the expensive work happens. It maintains an evidence table separating what is established, contested and unknown.
Designed for agentic platforms (LangGraph, Dify) that can attach browsing or retrieval tools, but equally useful as a prompt scaffold in GPTs and Claude Code.
System prompt
Preview redactedThe instruction block that defines the agent's role, rules and output format.
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████User prompt template
Fill in the variables and send it to run the skill: {{variables}}
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████
████████████████████████████████████████████████████████████████Variables
| Name | Required | Description | Example |
|---|---|---|---|
| {{question}} | Yes | The research question, as specific as possible. | Which vector DB offers the best p99 latency under 10M embeddings at $500/mo? |
| {{depth}} | Optional | Effort level: quick scan, standard, or exhaustive. | standard |
| {{sources_hint}} | Optional | Known starting sources, date ranges, or constraints. | prefer 2025+ benchmarks; ignore vendor-sponsored posts |
Full skill configuration
Redacted previewThe standard .aiskill file in JSON and YAML.
{
"format": "aiskill/v1",
"skill": {
"name": "Deep Research Agent",
"slug": "deep-research-agent",
"version": "2.2.0",
"description": "A disciplined research protocol: question decomposition, source strategy, evidence table and a confidence-rated synthesis.",
"category": "Research",
"tags": [
"research",
"agent",
"synthesis",
"citations"
],
"platforms": [
"chatgpt",
"claude-code",
"dify",
"langgraph"
],
"tested_environment": "GPT-5 with browsing, Claude Sonnet 4.5, LangGraph 0.2+ with retriever tools",
"license": "CC-BY-4.0"
},
"prompts": {
"system": "████████████████████████████████████████████████████████████████████████████████████████████████",
"user_template": "████████████████████████████████████████████████████████████████████████████████████████████████"
},
"variables": [
{
"name": "question",
"description": "The research question, as specific as possible.",
"required": true,
"example": "Which vector DB offers the best p99 latency under 10M embeddings at $500/mo?"
},
{
"name": "depth",
"description": "Effort level: quick scan, standard, or exhaustive.",
"required": false,
"example": "standard"
},
{
"name": "sources_hint",
"description": "Known starting sources, date ranges, or constraints.",
"required": false,
"example": "prefer 2025+ benchmarks; ignore vendor-sponsored posts"
}
],
"install": {
"steps": [
{
"title": "Copy the system prompt",
"detail": "Copy the full protocol prompt from the skill page."
},
{
"title": "Deploy on an agentic platform",
"detail": "Best with tools: enable browsing in a custom GPT, or attach a web-retrieval tool in Dify / LangGraph."
},
{
"title": "Approve the plan",
"detail": "Always review the Phase 1 plan and prune scope before letting research run."
},
{
"title": "Check the evidence table",
"detail": "Verify cited sources before external use of the synthesis."
}
]
},
"usage": {
"steps": [
"Write the question with measurable terms instead of vague superlatives.",
"Approve or trim the research plan in Phase 1.",
"Let the agent execute the evidence pass with your tools enabled.",
"Use the Confidence and Unknowns sections to decide follow-up research."
]
},
"metadata": {
"author": "SkillVault",
"source": "https://skillvault.top/skills/deep-research-agent",
"exported_at": "2026-11-03T07:42:34.096Z"
}
}format: aiskill/v1
skill:
name: Deep Research Agent
slug: deep-research-agent
version: 2.2.0
description: >-
A disciplined research protocol: question decomposition, source strategy, evidence table and a
confidence-rated synthesis.
category: Research
tags:
- research
- agent
- synthesis
- citations
platforms:
- chatgpt
- claude-code
- dify
- langgraph
tested_environment: GPT-5 with browsing, Claude Sonnet 4.5, LangGraph 0.2+ with retriever tools
license: CC-BY-4.0
prompts:
system: ████████████████████████████████████████████████████████████████████████████████████████████████
user_template: ████████████████████████████████████████████████████████████████████████████████████████████████
variables:
- name: question
description: The research question, as specific as possible.
required: true
example: Which vector DB offers the best p99 latency under 10M embeddings at $500/mo?
- name: depth
description: "Effort level: quick scan, standard, or exhaustive."
required: false
example: standard
- name: sources_hint
description: Known starting sources, date ranges, or constraints.
required: false
example: prefer 2025+ benchmarks; ignore vendor-sponsored posts
install:
steps:
- title: Copy the system prompt
detail: Copy the full protocol prompt from the skill page.
- title: Deploy on an agentic platform
detail: >-
Best with tools: enable browsing in a custom GPT, or attach a web-retrieval tool in Dify /
LangGraph.
- title: Approve the plan
detail: Always review the Phase 1 plan and prune scope before letting research run.
- title: Check the evidence table
detail: Verify cited sources before external use of the synthesis.
usage:
steps:
- Write the question with measurable terms instead of vague superlatives.
- Approve or trim the research plan in Phase 1.
- Let the agent execute the evidence pass with your tools enabled.
- Use the Confidence and Unknowns sections to decide follow-up research.
metadata:
author: SkillVault
source: https://skillvault.top/skills/deep-research-agent
exported_at: "2026-11-03T07:42:34.096Z"
Installation guide
- 1
Copy the system prompt
Copy the full protocol prompt from the skill page.
- 2
Deploy on an agentic platform
Best with tools: enable browsing in a custom GPT, or attach a web-retrieval tool in Dify / LangGraph.
- 3
Approve the plan
Always review the Phase 1 plan and prune scope before letting research run.
- 4
Check the evidence table
Verify cited sources before external use of the synthesis.
How to use
- Write the question with measurable terms instead of vague superlatives.
- Approve or trim the research plan in Phase 1.
- Let the agent execute the evidence pass with your tools enabled.
- Use the Confidence and Unknowns sections to decide follow-up research.
Platform compatibility
ChatGPT GPTs
OpenAI
Paste the system prompt into the GPT "Instructions" field and keep the user template as your starter message.
Official docsClaude Code
Anthropic
Save the exported .aiskill JSON as a skill file in your project or paste the system prompt into CLAUDE.md / a slash-command markdown file.
Official docsDify
LangGenius
Create a new Agent/App, paste the system prompt as the system prompt, and register template variables as input fields.
Official docsLangGraph
LangChain
Load the JSON config into your graph state: use prompts.system for the system message and prompts.user_template as a runnable template.
Official docsPreview gallery
Related skills
bgpt-paper-search
Search scientific papers and retrieve structured experimental data extracted from full-text studies via the BGPT MCP server. Returns 25+ fields per paper including methods, results, sample sizes, quality scores, and conclusions. Use for literature reviews, evidence synthesis, and finding experimental details not available in abstracts alone.
citation-management
Comprehensive citation management for academic research. Search Google Scholar and PubMed for papers, extract accurate metadata, validate citations, and generate properly formatted BibTeX entries. This skill should be used when you need to find papers, verify citation information, convert DOIs to BibTeX, or ensure reference accuracy in scientific writing.
infographics
Create professional infographics using Nano Banana Pro AI with smart iterative refinement. Uses Gemini 3 Pro for quality review. Integrates research-lookup and web search for accurate data. Supports 10 infographic types, 8 industry styles, and colorblind-safe palettes.