Agentic Setup Examples

This repository illustrates agentic configurations for GitHub Copilot, focusing on multi-agent collaboration and automated workflows.

Research Orchestration System

The current setup implements a sophisticated Research Orchestrator that manages a multi-phase research lifecycle by delegating tasks to sub-agents.

Key Agents

• Orchestrator: Manages the high-level workflow, state, and delegation. It handles task initialization, sequential execution of sub-tasks, and final consolidation.
• Employee: A versatile general-purpose agent used for preliminary research, detailed task execution, and report compilation.
• Reviewer: A quality-control agent focused on verification, anti-hallucination, and minor corrective edits.

Workflow Phases

1. Planning: Landscape survey (prelim-research) followed by a task breakdown (plan-research) into atomic, sequential task files.
2. Execution & Review: A strict loop where the employee researches a specific sub-task and the reviewer audits the result for accuracy and citations.
3. Consolidation: Merging all verified sub-reports into a single authoritative FINAL_REPORT.md or a structured README_RESEARCH.md index.

Features

• Context-Aware File Passing: Agents exchange information via file paths in a dedicated {{taskFolder}} to minimize context window bloat.
• Metadata Headers: All research documents include optimized headers (Keywords & Abstract) for high-speed relevance judging and RAG-style searchability.
• Accountability: Built-in "mission-critical" framing to encourage high integrity and verification.