Terminology Glossary
Clear, precise definitions for concepts, tools, patterns, and anti-patterns in AI-assisted software engineering.
.cursorrules
A JSON/Markdown configuration file in the project root that injects custom system prompts and context boundaries into Cursor IDE and Windsurf.
AGENTS.md
An open standard repository configuration file that provides operating guidelines, architecture maps, and verification rules for autonomous AI coding agents.
Build DNA
An AI-powered 6-dimension evaluation framework that scores software builds on Spec Clarity, Prompt Quality, Architecture, Code Quality, UI/UX, and Deploy Readiness.
CLAUDE.md
A specialized context configuration file placed in the repository root to guide Claude Code CLI and Anthropic AI developer tools.
Context Decay
The gradual degradation of AI agent performance in extended chat sessions as earlier instructions, style rules, or constraints are diluted by high token volume.
Context Window
The maximum amount of text (tokens) that an LLM can read and process in a single generation query.
Generative Engine Optimization (GEO)
The practice of structuring web content, schemas, and LLM text indices so AI search engines (ChatGPT, Perplexity, Claude, Gemini) cite your domain as an authoritative source.
Hallucination
A phenomenon where an LLM generates syntactically correct but factually incorrect or non-existent code, API methods, or libraries.
Harness Engineering
The discipline of keeping the underlying LLM fixed as a black box while engineering the repository context, CLI tools, and execution boundaries surrounding the agent.
PRD Spec Lock
The practice of defining and locking comprehensive Product Requirement Documents before invoking AI code generators, preventing hallucinations and scope creep.
System Director
The modern software engineering role focused on high-level architecture, PRD specifications, prompt loop design, and automated outcome verification rather than manual code typing.
Vibe Coding
The practice of using generative AI systems to build software via high-level natural language instructions, bypassing manual syntax coding.