Agent Discovery System

Progressive Agent Loading

This directory contains specialized agents organized by category for on-demand discovery and loading. Instead of loading all agent definitions upfront, explore categories based on task requirements to reduce context usage.

Agent Categories

📁 codebase/ - Code Analysis & Search

Agents for exploring, analyzing, and understanding codebases.

Use when:

• Finding specific files, functions, or patterns
• Analyzing code structure and relationships
• Understanding implementation details
• Locating usage examples

Available agents:

• locator - Find files and components quickly
• analyzer - Deep analysis of code structure
• pattern-finder - Identify patterns and anti-patterns

📁 documentation/ - Documentation & Thoughts Analysis

Agents for exploring documentation, comments, and architectural thoughts.

Use when:

• Finding architectural decisions
• Exploring development thoughts and rationale
• Analyzing documentation structure
• Understanding design choices

Available agents:

• thoughts-locator - Find specific thoughts/decisions
• thoughts-analyzer - Analyze and synthesize documentation

📁 workflow/ - Workflow Orchestration

Agents for managing complex multi-step workflows.

Use when:

• Orchestrating research tasks
• Creating implementation plans
• Managing development workflows
• Coordinating multi-agent operations

Available agents:

• (To be added based on workflow patterns)

📁 web/ - Web Research & External Resources

Agents for searching and analyzing external resources.

Use when:

• Researching external libraries or frameworks
• Finding documentation online
• Gathering best practices
• Exploring similar implementations

Available agents:

• web-search-researcher - Search and analyze web resources

Usage Pattern

1. Identify task requirements - What type of analysis is needed?
2. Explore relevant category - Load only the category README
3. Load specific agent - Load individual agent definition when needed
4. Execute with filtering - Agents return limited, structured results

Result Filtering Guidelines

All agents follow these filtering principles:

• Return maximum 10 most relevant results by default
• Provide structured summaries, not raw data dumps
• Include relevance scores when applicable
• Support on-demand detailed exploration

Context Efficiency

This progressive discovery system reduces initial context load by:

• 40-60% for complex workflows
• 30-50% for research operations
• 20-30% for simple tasks

Instead of loading ~50KB of agent definitions upfront, load only the 2-5KB needed for specific tasks.