Glossary¶

Orchestrator Core Concepts¶

Agent¶

Definition: An AI program capable of autonomously executing tasks
Characteristics: Goal-oriented, autonomous decision-making, environment-aware
Classification:
Hermes Agent: Master Agent, responsible for strategy formulation
Codex Agent: Execution Agent, responsible for code implementation
Worker Agent: Worker Agent, responsible for specific tasks

Orchestrator¶

Definition: A control system that coordinates multiple Agents working together
Core Functions: Task allocation, resource management, fault recovery, progress monitoring
Architecture Patterns:
Dual-Agent Loop: Hermes + Codex alternating work
Orchestrator-Worker: Hierarchical management structure
Hub-and-Spoke: Central辐射 (hub-and-spoke) structure

Orchestration¶

Definition: The process of decomposing complex tasks into subtasks and executing them in logical order
Key Elements:
Task Decomposition
Dependency Management
Concurrency Control
Fault Tolerance

Architecture Terminology¶

Session Management¶

Definition: Managing the lifecycle of Agent runtime sessions
Implementation Methods:
tmux sessions: Linux terminal multiplexing
systemd services: System daemons
Process groups: Collections of related processes

Isolation Mechanism¶

Definition: Mechanisms that prevent different Agents or tasks from interfering with each other
Types:
Process Isolation: Separate process spaces
File System Isolation: Independent file views
Network Isolation: Independent network namespaces
Resource Isolation: CPU, memory limits

Communication Mechanism¶

Definition: How information is passed between Agents
Five Communication Patterns:
Bracket-Paste: Terminal clipboard interaction
Mail System: Structured message passing
File Coordination: Information exchange via files
MCP Memory: Model Context Protocol
Copy-Paste Handoff: Explicit information passing

Fault Tolerance and Recovery Terminology¶

Watchdog¶

Definition: A daemon that monitors system health status
Hierarchy:
L1: Process-level Watchdog (detects process liveness)
L2: Function-level Watchdog (detects functional correctness)
L3: System-level Watchdog (detects overall operation)
L4: Business-level Watchdog (detects business logic)

Fault Recovery¶

Definition: The automatic repair process after a system failure
Recovery Strategies:
Fast Crash Protection: Immediate restart
Progressive Remediation: Gradually escalating handling intensity
Self-Scheduling Chain: Automatically scheduling recovery tasks

Rate Limit¶

Definition: A mechanism that controls the frequency of API calls
Handling Strategies:
Exponential backoff: Gradually increasing wait time
Request batching: Processing requests in batches
Caching mechanism: Reducing repeated calls

Knowledge Management Terminology¶

LEARNINGS.md¶

Definition: A document recording system runtime experience and lessons learned
Content Structure:
Timestamp: When the event occurred
Problem Description: Fault or phenomenon description
Root Cause Analysis: Deep-rooted cause analysis
Solution: Effective resolution method
Prevention Measures: Methods to avoid recurrence

Mulch Knowledge Base¶

Definition: A structured knowledge management system
Characteristics:
Version control: Git-managed
Categorical organization: Topic-based classification
Search functionality: Quick retrieval
Update mechanism: Automatic synchronization

MCP (Model Context Protocol)¶

Definition: Model Context Protocol, used for knowledge sharing between Agents
Functions:
Memory storage: Persisting Agent memories
Context passing: Runtime information sharing
Tool integration: External tool invocation

Deployment and Operations Terminology¶

systemd Services¶

Definition: Linux system daemon manager
Key Features:
Auto-start: Automatically runs at system startup
Crash recovery: Automatically restarts after process crash
Resource limits: CPU, memory, and other resource controls
Status monitoring: Real-time service status

tmux Sessions¶

Definition: Terminal multiplexer supporting multiple terminal sessions
Role in the Orchestrator:
Session persistence: Terminal sessions remain uninterrupted
Session recovery: Reconnection after disconnection
Multiplexing: Managing multiple Agents from a single terminal

Self-Check Mechanism¶

Definition: A mechanism for the system to self-inspect and verify
Check Types:
Health Check: System running status
Functional Check: Whether core functions are normal
Data Check: Data integrity verification
Configuration Check: Configuration file validity

Design Pattern Terminology¶

Self-Scheduling¶

Definition: A mechanism for Agents to autonomously arrange subsequent tasks
Implementation Methods:
Timed wakeup: Background process periodically wakes up
Event-triggered: Specific events trigger scheduling
Condition-triggered: Scheduling when conditions are met

Iron Law Guard¶

Definition: A guardian mechanism that enforces system constraints
How It Works:
Periodic checking: Verifying whether constraints have been modified
Automatic recovery: Restoring modified files from Git
Warning notifications: Sending warning messages to Agents

Progressive Remediation¶

Definition: A recovery strategy that gradually escalates handling intensity
Recovery Levels:
Level 0: Warning (may just be thinking)
Level 1: Nudge (may be waiting for input)
Level 2: Triage (may be truly stuck)
Level 3: Kill (confirmed unrecoverable)

Architecture Pattern Terminology¶

Dual-Agent Loop¶

Definition: A pattern where two Agents alternate their work
Workflow:
Agent A formulates strategy
Agent B executes tasks
Agent A evaluates results
Agent B adjusts execution
Repeat the cycle

Orchestrator-Worker¶

Definition: A hierarchical management architecture pattern
Role Division:
Orchestrator: Strategy formulation, task allocation
Worker: Specific execution, status feedback

Hub-and-Spoke¶

Definition: A hub-and-spoke architecture pattern
Structural Characteristics:
Hub: Central coordinator
Spoke: Peripheral execution nodes
Communication: All nodes communicate through the Hub

Tools and Technology Terminology¶

Git Worktree¶

Definition: Git's branch isolation mechanism
Application in the Orchestrator:
Concurrent development: Working on multiple branches simultaneously
Isolated environments: Independent development on different branches
Quick switching: Fast switching between branches

ZFC Health Check¶

Definition: Zero Fault Correction health check mechanism
Check Dimensions:
Z (Zero): Zero fault target
F (Fault): Fault detection
C (Correction): Fault correction

Adaptive Scheduling¶

Definition: Dynamically adjusting scheduling strategies based on system state
Scheduling Strategies:
Load-based: Adjusting based on task volume
Priority-based: Task priority ranking
Resource-based: Scheduling based on resource availability

Performance Optimization Terminology¶

Resource Utilization¶

Definition: The efficiency of system resource usage
Key Metrics:
CPU utilization: Computing resource usage rate
Memory utilization: Memory usage efficiency
Disk I/O: Disk read/write efficiency
Network bandwidth: Network usage efficiency

Throughput¶

Definition: The number of tasks completed per unit of time
Optimization Methods:
Batch processing: Processing tasks in batches
Parallelization: Executing tasks in parallel
Caching mechanism: Reducing repeated computation

Response Time¶

Definition: The time from request to response
Components:
Network latency: Transmission time
Processing time: Computation time
Wait time: Queue waiting time

Security Terminology¶

Session Isolation¶

Definition: Data isolation between different sessions
Implementation Methods:
Independent environments: Each session has an independent environment
Access control: Access permission control
Data encryption: Sensitive data encryption

File Locking¶

Definition: A mechanism to prevent concurrent file access
Lock Types:
Mutex lock: Only one process can access at a time
Shared lock: Multiple processes can read simultaneously
Conditional lock: Locking only when conditions are met

Access Control¶

Definition: A mechanism that controls user or process access to resources
Control Methods:
Role-Based Access Control (RBAC)
Attribute-Based Access Control (ABAC)
Discretionary Access Control (DAC)

Monitoring and Alerting Terminology¶

Metrics Monitoring¶

Definition: Real-time monitoring of system runtime metrics
Monitoring Metrics:
System metrics: CPU, memory, disk
Application metrics: Response time, error rate
Business metrics: Task completion rate, success rate

Alert Threshold¶

Definition: The critical metric value that triggers an alert
Threshold Types:
Static threshold: Fixed value
Dynamic threshold: Adjusted based on historical data
Trend threshold: Based on change trends

Log Aggregation¶

Definition: Collecting and centrally managing system logs
Functions:
Log collection: Collecting logs from multiple sources
Log storage: Centralized storage and management
Log analysis: Searching and analyzing logs

Extensibility Terminology¶

Plugin Architecture¶

Definition: An architecture that supports dynamic loading of functional modules
Characteristics:
Hot-pluggable: Loading/unloading at runtime
Standardized interfaces: Unified plugin interface
Version management: Plugin version control

Microservices Architecture¶

Definition: An architecture that decomposes a system into multiple independent services
Advantages:
Independent deployment: Services can be deployed independently
Technology heterogeneity: Different services can use different technologies
Elastic scaling: Can scale independently

Horizontal Scaling¶

Definition: Scaling a system by adding more nodes
Implementation Methods:
Load balancing: Distributing requests across multiple nodes
Data sharding: Distributed data storage
Stateless design: Making services stateless

Summary¶

This glossary covers key terminology related to AI Agent orchestrator core concepts, architecture patterns, technical implementations, and more. These terms form the foundation for understanding and implementing orchestrator systems. It is recommended to continuously deepen your understanding of these concepts through learning and practice.

As technology evolves, this glossary will be continuously updated and refined to reflect the latest technology trends and practical experience.