Skip to content

terms.md

Latest commit

 

History

History
107 lines (79 loc) · 2.75 KB

terms.md

File metadata and controls

107 lines (79 loc) · 2.75 KB
marp true
theme tessl-cheatsheet
paginate false
size 1400px 1200px

Coding Agents: Essential Terms

What is a Coding Agent?

AI-powered tool that autonomously writes, edits, and executes code. Unlike chatbots, agents take actions: read files, run commands, and iterate on solutions.

LLM (Large Language Model)

The AI brain powering the agent. Trained on code/text to understand and generate code.

Model

Specific version of an LLM with defined capabilities.

claude-sonnet-4-20250514
gpt-4o-2024-08-06
gemini-2.5-pro

Context Window

Max tokens the model processes at once. Larger = more code context. Ranges 8K to 1M+ tokens.

Tokens

How LLMs measure text. ~4 characters = 1 token. Code uses more tokens than prose.

Agent Capabilities

Capability Description
Read Access files in codebase
Write Create and modify files
Execute Run shell commands, tests
Search Find code patterns
Browse Fetch web documentation
MCP Connect to external tools

Key Concepts

System Prompt

Hidden instructions defining agent behavior. Sets tone, capabilities, constraints.

Tool Use / Function Calling

How agents take actions. Model outputs structured calls that runtime executes.

Agentic Loop

The core cycle that powers autonomous work:

think → act → observe → repeat

Context Management

Fitting relevant code into limited context via summarization, chunking, semantic search.

Working with Agents

Prompting Best Practices

  • Be specific about what you want
  • Provide relevant context upfront
  • Break complex tasks into steps
  • Let the agent ask clarifying questions

Rules Files

Project-specific instructions that persist across sessions:

CLAUDE.md    # Claude Code
GEMINI.md    # Gemini CLI
AGENTS.md    # OpenAI Codex
OPENCODE.md  # OpenCode

MCP (Model Context Protocol)

Standard for connecting agents to external data sources and tools (databases, APIs, services).

Common Patterns

Pattern Description
Plan-then-Execute Create plan, get approval, implement
Iterative Refinement Change, test, fix, repeat
Code Review Mode Analyze without modifying

Popular Models

Model Provider Strengths
Claude Sonnet/Opus Anthropic Coding, reasoning
GPT-4o / o1 OpenAI Broad knowledge
Gemini 2.5 Google Long context
DeepSeek R1 DeepSeek Open weights

Pricing Terms

  • Input tokens: What you send to the model
  • Output tokens: What the model generates
  • Cached tokens: Reused context (cheaper)
Tessl · Coding Agents Terms · __VERSION__