Glossary

Dictionary

Plain-language definitions for vibecoding concepts — from context windows to agentic loops.

Agent Skill: A packaged set of instructions, and optionally scripts or reference files, that extends what an AI agent knows how to do — loaded only when relevant.
Agentic Loop: The plan → act → observe → repeat cycle an AI agent runs through to complete a multi-step task without a new prompt at every step.

Chain of Thought: The step-by-step reasoning a model produces on the way to an answer, rather than jumping straight from question to conclusion.
Context Engineering: The practice of deliberately curating what information goes into an agent's context window — and what gets left out — rather than just wording a single prompt …
Context Rot: The tendency for a model's accuracy and focus to degrade as its context window fills up, even though the relevant information is technically still in there.
Context Window: The amount of text — code, conversation, files — a model can 'see' at once when generating a response.

Diff Review: The practice of reading exactly what an AI agent changed, line by line, before accepting it — the core review discipline of vibecoding.

Embedding: A numeric representation of a piece of text that captures its meaning, so a computer can measure how similar two pieces of text are.
Extended Thinking: A mode where a model spends extra computation reasoning through a problem step by step before giving its final answer, often shown to the user as a separate …

Git Worktree: A Git feature that checks out another branch of the same repository into a separate folder, so work on that branch doesn't disturb your main checkout.

Hallucination: When a model confidently produces something false — a function that doesn't exist, a library that isn't installed, a fact that isn't true.
Headless Mode: Running an AI coding agent from a script or command line, without its interactive chat interface, so it can be triggered automatically — for example, in a CI …

MCP (Model Context Protocol): An open standard that lets AI agents connect to external tools and data sources — databases, APIs, file systems — through a common interface.

Plan Mode: A mode where an agent explores and analyzes a codebase but is blocked from editing files or running commands until you review and approve its plan.
Prompt Caching: A feature that lets a model reuse previously processed parts of a prompt — like a long system prompt or file — instead of reprocessing them on every request.
Prompt Engineering: The practice of structuring instructions to an AI model to reliably get the output you actually want.
Prompt Injection: An attack where malicious instructions are hidden inside content a model reads — a webpage, a file, an issue comment — to hijack what the agent does next.

Repo Map: A lightweight summary of a codebase's structure that lets an agent reason about files it hasn't fully read into context.
Retrieval-Augmented Generation (RAG): A technique where a model looks up relevant information from an outside source — docs, code, a database — before answering, instead of relying only on what it …

Sandboxing: Running an agent's commands and file changes inside an isolated environment — a container, VM, or restricted filesystem — so mistakes or malicious actions can't …
Subagent: A secondary AI agent that a main agent spins up to handle part of a task — often with its own clean context — then reports results back to the main …
System Prompt: The standing set of instructions that shapes an AI model's behavior before any user message — tone, rules, tool access, and constraints.

Temperature: A setting that controls how random or deterministic a model's output is — low for consistency, high for variety.
Token: The basic unit of text a language model reads and generates — roughly a word or word-fragment, and the unit context windows are measured in.
Tool Use (Function Calling): The mechanism that lets a model call external functions — read a file, run a command, query an API — instead of only producing text.

Vector Database: A database built to store embeddings and quickly find the ones most similar to a given query — the engine behind most semantic search and RAG systems.
Vibecoding: Building software primarily by directing an AI coding agent in natural language, while still applying engineering judgment to review and steer its output.

YOLO Mode: A setting that lets an AI agent take actions — edits, commands, commits — without asking for approval on each one.