Specialist Agent Overview

A meta-agent that provides an overview of the Open Deep Research Team multi-agent system, orchestrating specialized research agents through hierarchical coordination to produce comprehensive, academic-quality research on complex topics.

When to Use This Agent

Choose Agent Overview when:

• Understanding the architecture of multi-agent research systems
• Configuring a team of specialized research agents for complex queries
• Learning how orchestration, coordination, and synthesis agents work together
• Setting up a hierarchical agent workflow for deep research tasks
• Evaluating which specialized agents to deploy for a given research question

Consider alternatives when:

• Running a single-agent research task (use a research analyst agent)
• Doing code research rather than domain research (use a codebase explorer)
• Generating reports without deep research (use a report generator agent)

Quick Start

Copy
# .claude/agents/specialist-agent-overview.yml
name: Research Team Overview
model: claude-sonnet-4-20250514
tools:
  - Read
  - Write
  - Bash
  - Glob
  - Grep
prompt: |
  You are the orchestrator of a multi-agent research team. Coordinate
  specialized agents for comprehensive research: query clarification,
  brief generation, research coordination, analysis, fact-checking,
  synthesis, and report generation. Route tasks to the right agent.

Example invocation:

Copy
claude --agent specialist-agent-overview "Coordinate a deep research
  project on the current state of quantum computing for drug discovery,
  covering recent breakthroughs, key companies, and timeline estimates"

Core Concepts

Research Team Hierarchy

User Query
    ↓
Query Clarifier → refines ambiguous questions
    ↓
Research Brief Generator → creates structured research plan
    ↓
Research Coordinator → assigns tasks to specialists
    ↓
┌────────────┬──────────────┬─────────────────┐
│ Researcher │ Researcher   │ Technical        │
│ (Domain 1) │ (Domain 2)   │ Researcher       │
└─────┬──────┘──────┬───────┘────────┬─────────┘
      ↓             ↓               ↓
  Fact Checker (validates claims across all findings)
      ↓
  Research Synthesizer (combines findings into coherent narrative)
      ↓
  Report Generator (produces final deliverable)

Agent Responsibilities

Agent	Role	Input	Output
Query Clarifier	Disambiguate and scope	Raw user query	Refined query with constraints
Brief Generator	Structure the research	Refined query	Research plan with objectives
Coordinator	Assign and sequence tasks	Research plan	Task assignments per agent
Research Analyst	Investigate specific topics	Task assignment	Raw findings with sources
Technical Researcher	Deep-dive technical details	Task assignment	Technical analysis
Fact Checker	Verify claims and sources	All findings	Validated/flagged findings
Synthesizer	Combine and reconcile findings	Validated findings	Unified narrative
Report Generator	Format final output	Synthesized content	Structured report

Orchestration Flow

Sequential phases with parallel research:

Phase 1 (Sequential): Clarify → Brief → Coordinate
Phase 2 (Parallel):   Researcher A ┐
                       Researcher B ├── simultaneous investigation
                       Researcher C ┘
Phase 3 (Sequential): Fact-check → Synthesize → Report

Configuration

Parameter	Description	Default
team_size	Number of research agents	3-5
research_depth	Depth level (quick, standard, deep)	Standard
fact_check_rigor	Fact-checking thoroughness	Medium
output_format	Final report format	Markdown
parallel_research	Enable parallel investigation	true
max_iterations	Maximum research-review cycles	3
source_requirements	Minimum sources per finding	2

Best Practices

1. Start every research project with query clarification. Ambiguous queries produce unfocused research. The Query Clarifier transforms "tell me about AI in healthcare" into "analyze the current adoption of machine learning models for diagnostic imaging in US hospitals, focusing on FDA-approved systems, accuracy benchmarks, and implementation barriers reported since 2022." This specificity directs all downstream agents toward relevant findings.

2. Assign researchers based on domain expertise, not workload balance. Each researcher agent should cover a distinct domain aspect. For "quantum computing in drug discovery," assign one researcher to computational chemistry methods, another to hardware/qubit developments, and a third to commercial applications and partnerships. Overlapping assignments create redundant findings and conflicting perspectives.

3. Run fact-checking across all findings, not just suspicious claims. Confirmation bias affects AI agents too. Findings that seem obviously true are sometimes wrong or outdated. The Fact Checker should validate every significant claim, verify source credibility, check publication dates, and cross-reference between researchers' outputs. Unchecked findings undermine the entire report's credibility.

4. Synthesize before formatting—these are separate concerns. The Synthesizer reconciles conflicting findings, identifies themes, and creates a coherent narrative. The Report Generator formats this narrative into a structured document with proper sections, citations, and visual elements. Combining these steps produces reports that look polished but contain unresolved contradictions.

5. Limit research cycles to prevent diminishing returns. Set a maximum of three research-review-refine cycles. The first cycle covers the broad landscape. The second fills gaps identified during synthesis. The third addresses specific questions from stakeholder review. Additional cycles rarely produce proportional value and delay delivery. Document remaining questions as "areas for further investigation."

Common Issues

Research agents produce redundant findings across topics. This happens when task assignments overlap. The Coordinator should create mutually exclusive research scopes with clear boundaries. If "AI hardware" and "AI software" overlap at the compiler/optimization layer, assign that overlap explicitly to one researcher and exclude it from the other's scope.

Fact-checking bottlenecks the entire pipeline. Fact-checking every claim from multiple researchers takes time. Prioritize fact-checking by claim importance: verify central arguments and quantitative claims first, leave minor contextual details for a second pass if time permits. Use tiered fact-checking: quick source verification for established facts, deep investigation for novel or surprising claims.

Final report lacks coherent narrative despite good research. The Synthesizer is the most critical and most difficult agent role. Combining five researchers' findings into a unified story requires identifying the overarching narrative, not just concatenating sections. Define the report's thesis before synthesizing: what's the one key takeaway? Organize all findings to support, complicate, or contextualize that central insight.