Glossary — Every Term in Plain Language
Quick definitions with pointers to the file that teaches each concept.
| Term | Plain meaning | Learn more |
|---|---|---|
| Agentic | The AI doesn’t just answer — it acts: reads files, runs commands, checks its own work, repeats until done | 00 |
| Agentic loop | The repeat cycle: gather context → take action → verify | 00, 06 |
| Session | One saved conversation tied to a project folder; resumable | 08 |
| Context window | Claude’s working memory for a session (~200K tokens) — everything it can “see” right now | 09 |
| Token | The unit text is measured (and billed) in; ~¾ of a word on average | 09, 41 |
| Compaction | Replacing old conversation with a summary to free context space | 09 |
| Prompt caching | API reuses the unchanged start of each request so long sessions stay fast/cheap | 54 |
| Checkpoint | Automatic snapshot of files before each change; /rewind restores | 08 |
| CLAUDE.md | A markdown file Claude reads at session start — your project’s standing instructions | 17 |
| Auto memory | Notes Claude writes for itself across sessions (MEMORY.md + topic files) | 17 |
| Permission mode | How much Claude may do without asking: default / acceptEdits / plan / bypass | 14 |
| Plan mode | Read-only mode: Claude explores and proposes a plan you approve before edits | 10 |
| Sandbox | OS-enforced walls around what Bash commands can touch (files/network) | 16 |
| Tool | A capability Claude can invoke: Read, Edit, Bash, WebFetch, Grep… | 07 |
| Skill | A reusable instruction package invoked as /name or auto-triggered | 20–22 |
| Hook | Your shell command that runs automatically on events (before/after tools, on stop) | 23–24 |
| MCP (Model Context Protocol) | Open standard for plugging external tools/data (GitHub, DBs, Slack) into Claude | 25–26 |
| Plugin | Shareable bundle of skills/hooks/MCP/agents installed from a marketplace | 27 |
| Subagent | A helper Claude with its own context window that does a task and reports back | 29 |
| Agent team | Multiple full Claude sessions with a lead, shared tasks, and peer messaging | 32 |
| Workflow (dynamic) | A script (written by Claude) that orchestrates dozens–hundreds of subagents | 50 |
| Agent view | One-screen manager for background sessions (claude agents) | 50 |
| Worktree | Git feature giving each parallel session its own checkout of the repo | 31 |
| Headless / non-interactive | claude -p "..." — run once, print, exit; the scripting building block | 33 |
| Routine | Scheduled cloud automation that runs even with your machine off | 34 |
| Goal | /goal <condition> — Claude keeps working until an evaluator confirms it’s met | 49 |
| Advisor | A stronger model Claude consults at decision points | 49 |
| Effort | How much reasoning the model applies (low→high, ultracode) | 13 |
| Fast mode | /fast — Opus with faster output, premium billing | 13 |
| opusplan | Opus brains for planning, Sonnet hands for execution | 10 |
| Extended thinking | ”think / think hard / ultrathink” — visible extra reasoning budget | 10 |
| Teleport | claude --teleport — pull a web/mobile session into your terminal | 38 |
| Remote Control | Drive your local session from phone/browser | 38 |
| Channel | MCP server that pushes events (Telegram, webhooks) into a running session | 39 |
| Computer use | Claude controls real apps on your screen (click/type/screenshot) | 48 |
| Artifact | A live web page published from your session to a private claude.ai URL | 51 |
| Deep link | claude-cli:// URL that opens a pre-filled local session | 51 |
| Output style | System-prompt persona/format change (Explanatory, Learning, custom) | 52 |
| Ultrareview | Deep multi-agent cloud code review (/code-review ultra) | 36 |
| Ultraplan | Deep multi-agent cloud planning | 10 |
| Ultracode | Effort level where Claude auto-orchestrates workflows | 50 |
| Agent SDK | Python/TS library to build your own agents on Claude Code’s engine | 40 |
| Dev container | Docker-based isolated environment for safe autonomous runs | 53 |
| Managed settings | Admin-enforced configuration users can’t override | 56 |
Official glossary: https://code.claude.com/docs/en/glossary
🧠 Quick Recall
Section titled “🧠 Quick Recall”- Trick: “Word blocking you? Glossary first — every entry links to its lesson.”
👨🏫 Teach It
Section titled “👨🏫 Teach It”- Do: use it as a vocab quiz bank at the start of each session.
- They’ll trip on: memorizing terms without doing the linked file’s hands-on.
Learning path: ⬅ 56-enterprise-deployment.md · Index · 🏁 You’ve reached the end — revisit 45-capstone-projects.md to close any open capstones.
Written by Fenil Patel