Agentic AIAI that autonomously takes multiple steps to achieve a goal — reads files, writes code, runs commands, verifies results — rather than just answering questions.Agentic LoopThe cycle an agent repeats: Gather context → Take action → Verify results → Done (or loop back). The human can interrupt at any point.CLI (Command Line Interface)A text-based interface where you type commands. AI-powered CLIs let you talk to an agent in natural language directly from your terminal.Claude CodeAnthropic's agentic CLI tool for full-project coding. Runs in the terminal, understands your codebase, and can read/write/execute autonomously.Context EngineeringThe discipline of giving an AI agent exactly the right information at the right time — no more, no less — to maximize output quality while minimizing token cost.Context WindowThe total amount of text (measured in tokens) an AI model can process at once. Everything — system prompt, conversation history, file contents — competes for this space.CLAUDE.mdAn instruction file loaded at session startup that tells the agent your project conventions, build commands, and architecture rules. Keep under 200 lines.Fallback RotationAutomatically switching to secondary API keys when primary quotas are exceeded — ensuring zero failed queries and no wasted retry tokens.GitHub Copilot CLIGitHub/Microsoft's terminal-based AI assistant focused on shell commands, git workflows, and PR management.HITL (Human-in-the-Loop)A pattern where the human can interrupt, redirect, or stop the agent at any point during its autonomous execution. Press Ctrl+C.HooksAutomated actions triggered by agent events (session start, before/after tool use, on stop). Used for formatting, validation, logging, and security guardrails.MCP (Model Context Protocol)An open standard that lets you expose your data and services as tools an AI agent can call live — turning it from a text generator into an operational assistant.OpenCodeOpen-source agentic coding CLI that supports any model (local or cloud). Key strength: self-hosting and data sovereignty.Permission ModesControls for how much autonomy the agent has: default (ask everything), acceptEdits (auto-allow edits), auto (AI classifies risk), plan (read-only until you approve).Prompt CachingA technique where repeated system instructions are cached so you pay for them once per session rather than once per message — reducing costs significantly.Schema CompressionSending only relevant tables, column types, and row counts to the LLM instead of your entire database schema — reducing a 400-table DB to ~2KB of context.Self-Learning AgentAn agent that improves accuracy over time through a memory layer that accumulates patterns — without model retraining or fine-tuning.Slash CommandsShortcuts typed directly in the agent prompt (e.g. /compact, /init, /deploy, /review) that trigger built-in functionality.Sub-AgentA specialist agent spawned by the main agent to handle a focused task in its own context window, returning a clean summary without polluting your conversation.TokenThe basic unit of text processing for LLMs. Roughly 4 characters or 3/4 of a word. Every token in your prompt costs money and context space.WorktreeA separate git branch + directory that lets you run multiple agent sessions simultaneously without merge conflicts./compactCompresses conversation history to free context-window space. Use when switching domains or after long debugging sessions./initAuto-generates a CLAUDE.md instruction file by reading your codebase — project conventions, build commands, architecture notes.