# Technical Glossary of AI-Assisted Programming

A reference guide to core architectural terms used across modern AI coding agents and research papers:

### 1. Abstract Syntax Tree (AST)
A hierarchical, tree-structured representation of source code syntax produced by a compiler parser. Tools like Tree-sitter build ASTs to extract exact function signatures, class boundaries, and variable scopes without executing code.

### 2. Fill-in-the-Middle (FIM)
A pre-training objective where segments of code are extracted and appended to the end of a prompt with special sentinel tokens (`<PRE>`, `<SUF>`, `<MID>`). This enables foundation models to complete code in the middle of an existing file (infilling) rather than merely appending code at the bottom.

### 3. Agent-Computer Interface (ACI)
A specialized abstraction layer formalized by SWE-agent designed to make shell environments, directory listings, and file viewports easily steerable by language models, avoiding terminal stdout bloat.

### 4. Repository Map (Repo Map)
A condensed token budget representation of an entire repository's architecture. Typically generated by computing personalized PageRank over the AST symbol call graph to identify the top-$K$ most relevant definitions.

### 5. Reflexion / Verbal Reinforcement Learning
An agentic self-debugging architecture where a model receives execution tracebacks from a failing test suite, generates a verbal critique of its own mistake, and uses that reflection to refine subsequent patches.