Research Executor

Overview

The Research Executor is a thin wrapper skill that delegates research execution to the research-orchestrator-agent. It validates inputs, prepares the execution context, and invokes the autonomous orchestrator agent to handle the complete 7-phase deep research workflow.

When to Use

• User provides a structured research prompt (from question-refiner)
• Need to execute systematic research with multiple agents
• Require comprehensive report with verified citations
• Research involves 3+ subtopics requiring parallel investigation

Core Responsibilities

1. Input Validation: Verify structured prompt completeness
2. Agent Invocation: Deploy research-orchestrator-agent with proper context
3. Progress Monitoring: Track agent execution and report status
4. Result Delivery: Return final research package to user

Architecture (Post-Refactoring)

User Request
     ↓
research-executor skill (this skill - thin wrapper)
     ↓
research-orchestrator-agent (autonomous agent)
     ↓
├── Phase 1: Question Refinement
├── Phase 2: Research Planning
├── Phase 3: Multi-Agent Deployment
├── Phase 4: Source Triangulation
├── Phase 5: Knowledge Synthesis
├── Phase 6: Quality Assurance
└── Phase 7: Output Generation

Key Change: All orchestration logic has been moved to research-orchestrator-agent. This skill only handles:

• Input validation
• Agent deployment
• Error handling at skill level

Quick Start

Execute research using structured prompt:
[STRUCTURED_PROMPT]

The executor will:
1. Validate prompt structure
2. Invoke research-orchestrator-agent
3. Monitor progress
4. Return results from RESEARCH/[topic]/

Input Requirements

Required: Structured research prompt with:

• TASK: Clear research objective
• CONTEXT: Background and significance
• SPECIFIC_QUESTIONS: 3-7 concrete sub-questions
• KEYWORDS: Search terms
• CONSTRAINTS: Timeframe, geography, sources
• OUTPUT_FORMAT: Deliverable specifications

Optional:

• Research type (deep/quick/custom)
• Quality threshold (default: 8.0)
• Max agents (default: 8)
• Token budget per agent (default: 15k)

Output Structure

RESEARCH/[topic]/
├── README.md
├── executive_summary.md
├── full_report.md
├── data/
│   ├── statistics.md
│   └── ontology/
├── sources/
│   ├── bibliography.md
│   └── source_quality_table.md
├── research_notes/
│   └── agent_findings_summary.md
└── appendices/
    ├── methodology.md
    └── limitations.md

Error Handling

Error Code	Description	Action
E001	Incomplete structured prompt	Request missing fields
E002	Agent deployment failed	Retry with fallback config
E003	Agent execution timeout	Report partial results
E004	Quality threshold not met	Trigger refinement (max 2 attempts)

Safety Limits

Limit	Value	Enforced By
Max parallel agents	8	research-orchestrator-agent
Max research time	90 minutes	research-orchestrator-agent
Min quality score	8.0	research-orchestrator-agent
Max token per agent	15,000	research-orchestrator-agent

Integration with Agents

Primary Agent: research-orchestrator-agent

• Handles all 7 phases
• Manages agent deployment
• Enforces quality gates
• Coordinates synthesis and validation

Supporting Agents (invoked by orchestrator):

• got-agent: For complex research optimization
• synthesizer-agent: For findings aggregation
• red-team-agent: For quality validation
• ontology-scout-agent: For domain reconnaissance
• Multiple research agents (web, academic, verification)

Key Features

• Simplified Design: ~95% of logic moved to orchestrator agent
• Backwards Compatible: Same interface for users
• Better Error Recovery: Agent-level autonomy improves resilience
• Clearer Separation: Skill = invocation, Agent = execution

Examples

See examples.md for usage scenarios.

Detailed Instructions

See instructions.md for implementation guide.