Glossary
Glossary of Terms
-
Context Window The amount of text (tokens) an AI model can consider at once. Large models like Claude 2 have ~100K token windows , enabling more project content to be included. • 65
-
Vibe Coding A fast, iterative coding style leveraging AI assistants for generating code, where the developer "goes with the flow" of AI suggestions, making adjustments on the fly rather than following a strict upfront design . • 63
-
Context Engineering The practice of assembling and supplying all relevant background information to AI systems dynamically, ensuring they have the necessary knowledge (project history, requirements, prior decisions, etc.) to perform tasks effectively . • 1 3
-
MCP (Model Context Protocol) An open protocol by Anthropic for connecting AI assistants to external data sources and tools (like filesystems, databases, web search) in a secure, standardized way . It allows AI models to fetch context or perform actions via "servers" exposing functionalities. • 19 22
-
Cursor AI An AI-powered code editor similar to an IDE that uses a context-aware model. It includes features like Notepads (for adding context notes) and a rules file for persistent instructions to the AI . • 18
-
Windsurf (Codeium) An AI code editor with advanced context features such as codebase indexing (Indexing Engine), Cascade (multi-step flows with memory), and Memories (persistent user rules and auto-learned context) . • 66 27
-
Claude Desktop A desktop application interface for Anthropic's Claude AI. Supports large context windows and integrates MCP servers for extended capabilities (like reading local files, accessing APIs). Useful for vibe coders needing Claude's 100k token context and tool integrations . • 61 24
-
Notepads / Rules file In context of Cursor (and analogous to Windsurf's Memories), Notepads are user-created notes providing context on specific topics (e.g., a file containing DB schema info) . The rules file (e.g., .cursorrules ) holds project-wide instructions that the AI should always consider (coding style, technologies in use, etc.) . • 14 18
-
Vector Database (Embedding Store) A database optimized for similarity search of highdimensional vectors. Used in context engineering to store embeddings of text (code, docs) so that relevant pieces can be retrieved via semantic similarity to a query. Examples: Pinecone, Weaviate. •
-
Embedding A numeric representation of text (or code) produced by an AI model, capturing semantic meaning. Embeddings allow similarity comparison; used to find which pieces of context are related to a given input or question. •
-
Literate Programming A paradigm where code and documentation are written together in a narrative, later tangled into compilable code. Ensures context (explanation) is packaged with code. Early attempt by Knuth to improve code understanding. •
-
Architectural Decision Record (ADR) A short document (often one per decision) capturing an important architectural choice, along with context and consequences. Useful in context engineering as they record why things are done a certain way. •
-
CI/CD Continuous Integration / Continuous Deployment. In our context, CI can be leveraged to continuously update context artifacts (like regenerating documentation, updating search indexes) as part of the development pipeline. •
-
Prompt Template A predefined format for AI prompts that ensures all necessary context fields are included. Helps users consistently provide context (e.g., a template that has sections for context, problem, constraints, etc.). •
-
Hallucination (AI) When an AI model produces information or code that is not grounded in reality or provided context (essentially "making things up"). Often due to lack of context or trying to fill gaps from its general training. •
-
Pair Programming Two programmers working together at one workstation. By analogy, with AI, the developer and AI pair program, so techniques ensure both "partners" share context (the •
human must feed the AI context it lacks, and vice versa the AI can remind the human of context found in code/docs).
Flow (in Tao of Code context) – The concept of being adaptive and fluid. In code/AI terms, it implies the development process (and context within it) should be flexible, continuous, and unobstructed – much like water flowing . Emphasizes adaptability in context management. • 54 67
Bibliography & Resources
Comprehensive bibliography and resources for context engineering. Academic papers, documentation, community discussions, and tools for implementing AI-assisted development.
Implementation Checklist
Practical step-by-step checklist for implementing context engineering in vibe coding projects. From defining artifacts to team alignment and monitoring.