Skip to content

11. Glossary and FAQ - HugAI v2 Specification

This section provides definitions of key terms and answers frequently asked questions about the HugAI v2 protocol.

11.1. Glossary of Terms

  • Agent: Autonomous entity (software) that executes tasks, collaborates, and communicates via the protocol.
  • Governance: Mechanisms and policies to supervise, approve, and audit actions of agents and humans.
  • Capability: Specific functionality or task that an agent can announce and execute (e.g., code-generation, testing).
  • Workspace: Set of folders or resources operated on by an agent.
  • Approval: Process of human review and authorization for critical actions.
  • Audit: Detailed record of actions, decisions, and events for traceability and compliance.
  • Message: JSON structure exchanged between agents, humans, and tools following the protocol.
  • Extension: Addition of new messages, capabilities, or workflows to the base protocol.
  • Integration: Connection with external tools (CI/CD, analysis, etc.) via standard or custom messages.
  • Human-in-the-loop: Model where human intervention is mandatory at critical points.
  • Supervised: Model where human intervention is optional or at defined points.
  • Autonomous: Model where agents act without human intervention.

11.2. Frequently Asked Questions (FAQ)

What is HugAI v2 and what is it for?

It is an open protocol to orchestrate AI agents and humans in development and automation processes, with governance and traceability.

How is the protocol extended?

By adding new types of messages, capabilities, or agents following the JSON-RPC scheme and documenting the extension.

What security mechanisms exist?

Authentication of agents and humans, message validation, auditing, use of encrypted channels, and secure secret management.

Can I integrate my own tools?

Yes, via tool/integration messages or custom extensions, always following best practices for validation and security.

What happens if an agent receives an unknown message?

It should ignore it (if not critical) or respond with a standard error, maintaining robustness and compatibility.

How is configuration managed?

Through files (JSON/YAML), environment variables, and runtime parameters, following the documented best practices.

What governance levels exist?

Autonomous, supervised, and human-in-the-loop, according to the required degree of human intervention.

How are actions audited?

Using the audit/log message to record actor, action, timestamp, details, and result in protected systems.

Is the protocol compatible with JSON-RPC?

Yes, it is based on JSON-RPC 2.0 for message structure and extensibility.

Note: For additional questions, suggestions, or contributions, consult the official documentation or contact the protocol maintainers.