AI coding glossary
The terms that come up when you work with coding agents, each opening with a short definition. We explain the idea in general first, then how it shows up in AgentVera.
- Agent Skills Agent Skills are folders of instructions, scripts and reference files that teach an agent how to do a particular job. Each skill centres on a SKILL.md file, and the agent loads a skill only when the task at hand calls for it.
- Agentic coding Agentic coding is a way of building software where you hand tasks to an AI agent and direct its planning, code changes and testing instead of writing every line yourself. The developer’s job shifts from typing code to describing work, reviewing it and verifying it.
- AI coding agent An AI coding agent is a tool that takes a software task in plain language and carries it out step by step: it reads the code, edits files, runs terminal commands and checks the result itself. Claude Code, Codex and Gemini CLI are well-known examples.
- Claude Code hooks Claude Code hooks are shell commands that run automatically at set points in the agent’s lifecycle, such as before a tool runs or when a turn ends. Because they run every time instead of relying on the model, they are a dependable way to enforce rules and add checks.
- Context window A context window is the maximum amount of text a language model can take into account in a single request, measured in tokens. System instructions, tool definitions, conversation history, files the agent read and command output all have to fit inside it.
- Git worktree A git worktree is an extra working folder attached to the same git repository. Each folder has a different branch checked out, so you can work on two things at once without cloning the repo again, and keep parallel coding agents out of each other’s way.
- Human-in-the-loop Human-in-the-loop means an automated AI system pauses at critical steps and waits for a person to approve, correct or decide. In coding agents it usually takes the form of permission prompts before editing files, running commands or publishing anything.
- MCP (Model Context Protocol) MCP (Model Context Protocol) is an open protocol that lets AI applications connect to external tools, data sources and services in a standard way. Anthropic introduced it in November 2024, and today Claude Code, Codex and many other coding agents can use MCP servers.
- Multi-agent orchestration Multi-agent orchestration means splitting work across several AI agents and managing their order, communication and how their results come together. Steps like research, planning, coding and review go to different agents, which run one after another or in parallel.
- Prompt caching Prompt caching means the model provider stores the unchanged opening part of a request and reuses it on later requests instead of processing it again. Input read from the cache is much cheaper and faster than regular input.
- Prompt injection Prompt injection is an attack where instructions hidden in text a language model reads steer it away from its real task. For coding agents, those instructions can come from a web page, an issue, a document or tool output, and can make the agent take actions you never asked for.
- Subagent A subagent is a helper agent that a main coding agent starts to hand off a specific job. It has its own context window, instructions and tool permissions, and when it finishes it returns only a summary, so the main conversation doesn’t fill up with detail.
- Token A token is the basic unit of text a language model works with: a word, part of a word or a punctuation mark. A model’s context limit, speed and price are all measured in tokens.
- Usage limits Usage limits are caps on how much you can use a subscription-based AI tool within a given period. Claude and Codex subscriptions combine a 5-hour usage window with weekly limits.
- Vibe coding Vibe coding is a way of programming where you describe what you want in plain language and accept the AI’s code without reading it closely, judging only whether the result works. Andrej Karpathy coined the term in February 2025.
- Visual regression testing Visual regression testing compares screenshots of a user interface against previously approved baseline images to catch unintended visual changes. It finds bugs that code tests miss, like a shifted button, overflowing text or a broken mobile layout.
See these ideas at work.
Download AgentVera for free, add a project folder and run Claude Code, Codex and other agents side by side.