The vocabulary of AI agent configuration.

Plain definitions for the terms HubBound uses, and for the wider vocabulary teams run into when they try to standardize how their AI agents work.

Written to be quoted

Every entry is a self-contained definition, not marketing copy.

Core concepts first

Artifact, Kit and distribution — the three nouns the product runs on.

The wider ecosystem

MCP, context engineering, prompt versioning and the terms around them.

HubBound core concepts

AI Artifact: An AI Artifact is a versioned, reusable unit of AI agent configuration. It can be a prompt, a skill, a rule, a hook, a subagent or an MCP server. Every Artifact has an owner, a version history and a visibility scope, and any Artifact can be installed on its own or bundled into a Kit. Artifacts are to AI agents what packages are to code: the smallest thing worth sharing, versioning and depending on.
Kit: A Kit is a named, versioned bundle of AI Artifacts installed with a single command. A Kit is to Artifacts what a package is to modules — the unit a team actually distributes. Instead of asking a new developer to copy nine files out of a wiki, you hand them one Kit and their environment matches the rest of the team in seconds.
Distribution: Distribution is the act of assigning a Kit to a team or to an entire organization. It is the layer public registries do not have: because HubBound knows who belongs to which team, a Kit can reach the frontend group without reaching everyone. Scoped distribution is the main reason an internal registry beats a public one inside a company.
Organization: An organization is the top-level tenant in HubBound. It owns private Artifacts, private Kits and its members, and it is the boundary that visibility rules are enforced against. Anything marked private is visible only to members of the organization that owns it.
Team: A team is a group of people inside an organization — frontend, backend, platform, product. Teams are the target of distribution and the unit of measurement: Kits are sent to teams, and AI usage metrics are reported per team as well as across the whole organization.
Marketplace: The marketplace is where Artifacts and Kits are browsed, installed and forked. Public entries are visible to everyone; private entries are visible only inside the organization that owns them. Both live in the same interface, so a team searches one place regardless of whether the answer is internal or community-built.
Fork: Forking creates an independent copy of an Artifact or Kit that a team can modify without affecting the original. It is how a team adopts a community standard and then adapts it — the same operation, and the same reasoning, as forking a repository.

The six Artifact types

Prompt: A prompt is a reusable instruction given to an AI model or agent. In HubBound a prompt is an Artifact, which means it has a version history and an owner: improving a prompt improves it for everyone who installed it, instead of improving it only in the file of the person who wrote it.
Skill: A skill is a packaged procedure an AI agent can invoke to perform repeatable work — run a migration checklist, write a changelog entry, prepare a release. Where a prompt is an instruction, a skill is a capability the agent can call by name.
Rule: A rule is a convention or constraint an AI agent must always follow. Rules are what files like CLAUDE.md, .cursorrules and AGENTS.md contain. As Artifacts they live in one shared, versioned place instead of being copied into every developer's dotfiles and drifting apart.
MCP server: An MCP server is a program that exposes tools and data to an AI agent through the Model Context Protocol. It is how an agent reaches your issue tracker, your database or your internal APIs. Distributing MCP servers as Artifacts means a team connects to the same systems the same way, rather than each developer configuring their own.
Subagent: A subagent is a specialized AI agent that a main agent delegates a focused task to, with its own instructions, tools and model — a code reviewer, a test writer, a migration agent. Subagents keep the main agent's context clean by moving narrow work into a separate one.
Hook: A hook is an automation that fires around an AI agent's lifecycle: before a tool call, after an edit, on session start. Hooks are where a team's policy becomes enforcement rather than a document — the point at which a convention can actually block something.

The wider ecosystem

Model Context Protocol (MCP): The Model Context Protocol is an open standard, introduced by Anthropic in 2024, for connecting AI models to external tools and data sources. It defines how an AI client discovers and calls capabilities exposed by an MCP server, which is what makes a tool integration portable across AI applications rather than specific to one.
Registry: A registry is a system of record for versioned, installable artifacts, along with the metadata needed to find and depend on them. npm is a registry for JavaScript packages; Docker Hub is a registry for container images; HubBound is a registry for AI agent capabilities.
Prompt management: Prompt management is the practice of storing, versioning, reviewing and distributing prompts as shared assets rather than as text pasted into individual editors. It becomes necessary the moment more than one person depends on a prompt working the same way.
Prompt version control: Prompt version control is applying versioning to prompts and other AI configuration: every change produces a new version, previous versions remain retrievable, and consumers choose when to move. It answers the question teams hit first — which version of this prompt is the one that actually worked.
Context engineering: Context engineering is the practice of deciding what information an AI agent has available at the moment it acts — instructions, rules, tools, retrieved documents and conversation history. It is the discipline that grew out of prompt engineering once agents gained tools and memory, and the prompt stopped being the only lever.
Agentic AI: Agentic AI describes systems where a model plans and executes multi-step work using tools, rather than only producing text in response to a single request. AI coding agents such as Claude Code, Cursor and Codex are the form most engineering teams meet first.
AI governance: AI governance, in an engineering context, is the set of controls that determine which AI capabilities are available to whom, how they are configured, and what evidence exists about how they were used. It is a distribution and measurement problem before it is a policy document.
Configuration drift: Configuration drift is the gradual divergence of setups that were meant to be identical. In AI tooling it appears when every developer configures rules, hooks and MCP servers on their own machine: what works for one person does not exist for the next, and onboarding starts from zero every time.