Core Concepts
Agent
Agent
The main AI assistant that processes requests and executes tools
Context Exhaustion
Context Exhaustion
When conversation history fills context window, limiting the ability to reference earlier content
Context Window
Context Window
Amount of conversation history the AI can reference
MCP (Model Context Protocol)
MCP (Model Context Protocol)
System for extending Verdent with external tools
Subagent
Subagent
Specialized AI agent with isolated context for specific tasks (Explorer, Verifier, Code-reviewer)
Tool
Tool
Function the agent can execute (file_read, bash, glob, etc.)
Execution Modes
Auto-Run Mode
Auto-Run Mode
Tools execute automatically without approval prompts
Manual Accept Mode
Manual Accept Mode
User approves each tool execution before running
Plan Mode
Plan Mode
Agent creates plan before executing changes
Skip Permissions
Skip Permissions
Run without permission prompts (advanced)
Files & Configuration
AGENTS.md
AGENTS.md
Project-specific team standards file (project root, version controlled)
mcp.json
mcp.json
MCP server configuration (
~/.verdent/mcp.json)plan_rules.md
plan_rules.md
Plan Mode output customization (
~/.verdent/plan_rules.md)user_rules.md
user_rules.md
Global personal preferences file (
~/.verdent/user_rules.md)Workspace
Workspace
Current VS Code project directory where Verdent operates
Tool-Specific Terms
Glob Pattern
Glob Pattern
File search pattern using wildcards (
**/*.ts, src/**/*.js)Line Range
Line Range
Specific lines to read from large files (start_line, max_lines)
Regex
Regex
Regular expression for advanced text searching
Replace All
Replace All
file_edit flag to replace multiple identical strings
Timeout
Timeout
Maximum time for bash commands: 120 seconds (2 minutes, hard limit)
Subagent Terms
Flexible Policy
Flexible Policy
Subagent can activate automatically based on request patterns
Invocation Policy
Invocation Policy
Rules for when subagent activates (strict vs flexible)
Isolated Context
Isolated Context
Subagent’s separate context window (doesn’t consume main context)
Strict Policy
Strict Policy
Subagent only runs when explicitly @-mentioned
System Prompt
System Prompt
Instructions defining subagent’s behavior and expertise
MCP Terms
Global Config
Global Config
User-wide MCP configuration (
~/.verdent/mcp.json)MCP Server
MCP Server
External tool/service integrated via Model Context Protocol
Plugin
Plugin
Pre-built MCP integration for specific tools or services
Workspace Config
Workspace Config
Project-specific MCP configuration (
.verdent/mcp.json)Performance Terms
Delegation
Delegation
Routing tasks to subagents to preserve main context
Parallel Execution
Parallel Execution
Multiple subagents running simultaneously
Strategic Reading
Strategic Reading
Using line ranges and search to minimize context usage
Task Coordination
Task Coordination
Main agent managing multiple subagent operations