AI terms explained in plain words. Every term shows where it comes from, so you can open the original and read the full definition.
Each definition is a short summary in plain words. The original page is the authority: if they differ, follow the original.
52 terms · Checked 2026-10-19
A project instruction file shared by many AI coding agents
A Markdown file of project instructions written for AI coding agents in general, not just Claude. If a repository has AGENTS.md and no CLAUDE.md, Claude Code reads it as the project instructions. When you also keep a CLAUDE.md, import AGENTS.md from it so every tool follows the same rules.
Several Claude Code sessions working together under a team lead
Multiple independent Claude Code sessions coordinated by a team lead, with a shared task list and messages between teammates. Unlike subagents, which stay inside one session and report only to it, each teammate has its own context window and you can talk to any of them directly. The feature is experimental and off by default.
A way of working where the AI reads, runs and changes code by itself
A workflow where the AI reads files, runs commands and makes changes on its own while you watch, redirect it or step away. A chat assistant only answers with text that you must copy and apply yourself. Claude Code is agentic because it has tools that let it act, not just advise.
The program around a model that turns it into a coding agent
The tools, context management and execution environment that turn a language model into a capable coding agent. Claude Code is the harness and Claude is the model inside it. The harness supplies file access, shell commands, permission checks, memory loading and the loop that chains actions together.
The repeating cycle of gathering context, acting and checking results
The cycle Claude works through for every task: gather context, take action, verify the result, and repeat until the job is done. Each tool use returns information that shapes the next step, and you can interrupt the loop at any time to change direction. Hooks, skills and MCP plug into specific phases of this loop.
A live web page Claude Code publishes from your session
A live, interactive web page that Claude Code publishes from your session to a private address on claude.ai. You see the output visually or share it instead of reading terminal text, and the page updates in place when the session publishes again. Artifacts appear in the same gallery as those made in claude.ai conversations.
Notes Claude writes for itself from your corrections and preferences
Notes Claude writes for itself based on your corrections and preferences, stored per git repository under ~/.claude/projects/. The beginning of the MEMORY.md index loads at the start of every session, so lessons carry over. Auto memory is what Claude writes; CLAUDE.md is what you write.
A permission mode where a classifier model approves actions for you
A permission mode where a separate classifier model reviews Claude Code's actions instead of you, so most of them run without asking. Claude Code still asks before actions that match your explicit ask rules. The classifier blocks scope creep, untrusted infrastructure and prompt injection, and it never sees tool results, so hostile text in a file can't steer it.
Starting Claude Code without hooks, skills, plugins or memory
With the --bare flag, Claude Code starts without loading hooks, skills, custom commands, subagents, plugins, MCP servers, auto memory or CLAUDE.md. The only exception is skills in a folder you pass with --add-dir. It is recommended for CI and scripts that must give the same result on every machine.
Ready-made skills that come with Claude Code, such as /code-review
Prompt-based playbooks included with Claude Code, such as /batch, /code-review, /debug and /loop. A built-in command runs fixed logic, but a bundled skill gives Claude a detailed prompt and lets it organize the work. So it can start agents, read files and adapt to your codebase.
An MCP server that pushes outside events into your running session
An MCP server that pushes events into your running session so Claude can react to things that happen while you are away from the terminal. A channel can be two-way: Claude reads an incoming event and replies through the same channel. Telegram, Discord and iMessage are included in the research preview.
A restore point saved at each prompt so you can undo changes
A restore point created each time you send a prompt. Claude Code saves files before every edit, so pressing Esc twice or running /rewind restores the code, the conversation or both to an earlier point. Checkpoints are separate from git and do not track changes made through shell commands.
The folder where Claude Code reads its configuration
The directory where Claude Code reads configuration: settings, hooks, skills, subagents, rules and auto memory. A project keeps .claude/ at its root for project settings, and your personal defaults for every project live in ~/.claude/ in your home folder.
A file of lasting instructions you write for Claude
A Markdown file of persistent instructions you write for Claude, loaded at the start of every session. Put project conventions, architecture notes and "always do X" rules here. It can live in the project, in ~/.claude/ for all your projects, or in an organization policy. All found files are combined, from the broadest to the most specific, and the project-root file is re-read after compaction.
A Claude Code session that runs in the cloud, not on your computer
A Claude Code session that keeps running after you close your laptop, because it runs on cloud infrastructure instead of your machine. You start one from claude.ai/code, the Claude mobile app, the Desktop app with Cloud selected, or claude --cloud. A session in your own terminal or IDE is a local session.
A reusable instruction you run by typing /name
A reusable instruction you invoke by typing /name in the prompt. Built-in commands such as /clear, /model and /compact control the session, and you can add your own as files in .claude/commands/ or install them from a plugin. Skills are now the recommended way to package multi-step commands. The old name "slash command" means the same thing.
Automatically summarizing the conversation when context is almost full
Automatic summarizing of your conversation when the context window nears its limit. Older tool outputs are cleared first, then the conversation is summarized. The project-root CLAUDE.md and auto memory reload from disk afterward, but instructions given only in chat may be lost. Run /compact to do it yourself, for example /compact focus on the API changes.
An MCP server added to your claude.ai account
An MCP server added to your claude.ai account instead of being set up in Claude Code. When you sign in to Claude Code with that account, your connectors appear in /mcp next to the servers you added locally. Organizations can also provide connectors and control each of their tools.
The working memory of a Claude Code session
The working memory of a session. It holds the conversation, file contents, command outputs, CLAUDE.md, auto memory, loaded skills and system instructions. As you work it fills up until compaction summarizes it. Run /context to see what is using the space.
Sending a coding task from your phone to the Desktop app
A task router started from your phone. When you send a coding task from the Claude mobile app, it starts a Claude Code session in the Desktop app, and your prompt goes to the right tool automatically. It is available on certain paid plans.
A setting for how much the model thinks at each step
A setting that controls adaptive reasoning, which lets the model decide whether and how much to think at each step. Higher effort means more thinking tokens and deeper reasoning. Lower effort is faster and cheaper. Only some models support it.
Step-by-step reasoning the model shows before it answers
Visible step-by-step reasoning the model works through before responding. You adjust it with the effort level, or cap thinking tokens with MAX_THINKING_TOKENS on models with a fixed thinking budget. In the terminal, the thinking appears in gray italic text.
The YAML settings block at the very top of a Markdown file
A block of YAML settings at the very top of a Markdown file, between an opening --- line and a closing --- line. Skills, subagents, output styles and rules read their settings from it, such as a skill's description or a subagent's tools. Everything after the closing --- is the instructions, and the opening --- must be the first line of the file.
Your own handler that runs automatically at a set moment
A handler you define that runs automatically at a specific point in Claude Code's lifecycle, such as before a tool runs, after a file edit or when a session starts. It can be a shell command, an HTTP endpoint, an MCP tool, an LLM prompt or a subagent. Hooks are deterministic: they always fire at those points, not when the model decides.
Organization-wide settings that users cannot override
Settings enforced across an organization by IT or DevOps, delivered from Anthropic's servers through the admin console or placed on devices outside ~/.claude. User and project settings cannot override them. They are used for security policies, compliance rules and the same tools across every machine.
An open standard that connects AI tools to outside services
An open standard for connecting AI tools to external data sources and services. MCP servers give Claude new tools for Slack, Jira, databases, browsers and hundreds of other integrations. You connect servers with /mcp or by adding them to .mcp.json.
A program that gives Claude tools, prompts or resources over MCP
A program that gives Claude tools, prompts or resources over MCP. You add one with claude mcp add, in .mcp.json, through a plugin, or as a claude.ai connector. A local server runs as a process started from the command and args fields of its settings, which have nothing to do with the commands you type at the prompt.
Loading MCP tool details only when they are needed
A context-saving mechanism that delays loading MCP tool schemas until they are needed. Only tool names and server instructions load at startup, and Claude fetches a tool's full schema when it decides to use it. This keeps idle MCP servers from using much context.
Running one prompt with claude -p and exiting
A mode that runs a single prompt and exits without an interactive screen, started with -p or --print. It is used in CI, scripts and pipes, and the run is still saved as a session you can resume. It was formerly called headless mode, and the Agent SDK is its Python and TypeScript counterpart.
A setting that changes how Claude responds
A configuration that changes the instructions Claude Code gives Claude, to set the response behavior, tone or format. CLAUDE.md adds project context next to the default instructions. A custom output style adds its own instructions and can leave out the default software engineering instructions.
The basic rule for when Claude Code asks for your approval
The baseline approval behavior for a session. Switch it with Shift+Tab in the CLI or with the mode selector in VS Code, Desktop and claude.ai. The modes are default (shown as Manual), acceptEdits, plan, auto, dontAsk and bypassPermissions.
A setting that allows, asks about or denies a specific tool use
A settings entry that allows, asks about or denies a tool call based on the tool name and an argument pattern. Rules are checked in the order deny, then ask, then allow, and the first match wins. They are fine-grained controls on top of the broader permission mode.
A mode where Claude investigates and proposes a plan without editing
A permission mode where Claude researches and proposes changes without editing your source files. It can read, search and run exploration commands, then shows a plan for your approval before touching anything. Enter it with /plan or by pressing Shift+Tab.
A package of skills, hooks, subagents and MCP servers you install at once
A bundle of skills, hooks, subagents and MCP servers packaged as one installable unit. A plugin's skills are named plugin-name:skill-name, so many plugins can live side by side. Teams share plugins through a marketplace.
Approving a folder before Claude Code loads its settings
A dialog where you accept a folder before Claude Code loads its configuration. The answer is saved per project folder, except your home folder, where it lasts only for the current session. Until you trust a folder, Claude Code holds back some of what the repository supplies, such as allow rules in .claude/settings.json.
Hidden hostile instructions that try to hijack the AI
Hostile instructions hidden in a file, web page or tool result that try to steer Claude toward actions you never asked for. Claude Code defends with the permission system, command injection detection and trust checks. Auto mode adds a scan of tool results and a classifier that reviews actions without seeing the injected text.
Continuing a local session from your phone or browser
A way to continue a Claude Code session running on your computer from your phone or browser through claude.ai. The code runs and the files stay on your machine; only the screen is remote. It differs from a cloud session, which runs in a cloud sandbox.
Instruction files in .claude/rules/ that load with CLAUDE.md
Modular instruction files in .claude/rules/ that load together with CLAUDE.md. A rule can be limited to certain paths with paths: in its frontmatter, so it loads only when Claude reads or edits a matching file. This keeps the context small until the rule matters.
Fencing in the files and network that commands can reach
Operating-system-level isolation of files and network for the Bash tool. Commands run inside a boundary you set in advance, so Claude can work freely inside it without asking for approval on each command. Sandboxing is a separate layer from permission rules.
One conversation tied to a folder, with its own context
A conversation tied to your current folder, with its own separate context window. You can resume it with claude -c, fork it with --fork-session or run several in different terminals. /clear starts a new session, and the old one stays saved so /resume can open it.
The order in which Claude Code combines settings files
The order Claude Code reads settings in, from highest to lowest priority: managed policy, the --settings flag, .claude/settings.local.json, .claude/settings.json, then ~/.claude/settings.json. Lists are merged across layers, and single values in a higher layer win over lower ones.
A SKILL.md file of instructions Claude adds to its toolkit
A SKILL.md file with instructions, knowledge or a workflow that Claude adds to its toolkit. Claude loads a skill automatically when it fits the task, or you call it with /skill-name. Skills follow the open Agent Skills standard and replace the old custom commands, which still work.
A helper AI with its own context that does one delegated task
A specialized AI assistant that runs in its own context window with its own system prompt, tools and permissions. It works on a delegated task and returns a summary to the main conversation, so large searches don't fill your main context. Built-in subagents include Explore, Plan and general-purpose.
Any place you use Claude Code: CLI, IDE, Desktop or the web
Any place you access Claude Code: the CLI, VS Code, JetBrains, Desktop or claude.ai. All surfaces share the same engine. Sessions on your machine read your local CLAUDE.md, settings and skills, while cloud sessions start from a fresh copy of your repository.
The instructions Claude Code sends before your conversation
The instructions Claude Code sends ahead of your conversation on every request, covering how Claude uses tools, stays safe and formats answers. You can add to it with --append-system-prompt or replace it with --system-prompt. CLAUDE.md and output style instructions are not part of it; they arrive as system reminders.
Context Claude Code adds to the conversation for Claude
A message that Claude Code, as the harness, adds to the conversation to give Claude context. You don't send it yourself. CLAUDE.md, output style instructions, text returned by hooks, the list of skills and notes about changed files all reach Claude this way.
Pulling a cloud session into your local terminal with /teleport
The /teleport command pulls a cloud Claude Code session into your local terminal. Claude fetches the branch, loads the conversation history and continues from where the cloud session stopped. The opposite direction is --cloud, which sends a local task to run in the cloud.
An action Claude can take, such as reading a file or running a command
An action Claude can take: read a file, edit code, run a shell command, search the web or start a subagent. Tools are what make Claude Code agentic; without them, Claude can only answer with text. Each tool use returns a result that shapes Claude's next decision in the agentic loop.
The saved file record of a session
The saved record of a session. The conversation is what you and Claude exchange; the transcript is that conversation kept as a file, by default under ~/.claude/projects/. Claude Code reads the file back when you resume, which is how a conversation continues after the session ends.
One complete response from Claude, from your message to its finish
One complete response from Claude within a session. A turn begins when you send a message and ends when Claude finishes, with any number of tool calls in between. A session is made of many turns, and Stop hooks run at the end of each one.
Giving Claude a check to repeat until the work truly passes
How a session knows the work is really done instead of accepting a result that only looks right. You give Claude a check it can run, such as tests, a build or a screenshot comparison, and Claude keeps going until it passes. Without one, the only judge of "finished" is the agent itself.
Running Claude in its own git worktree so parallel work doesn't collide
An isolation mode that runs Claude in a separate git worktree under .claude/worktrees/, turned on with the -w flag or isolation: worktree in a subagent's settings. Changes stay on their own branch in their own folder, so agents working in parallel each edit their own copy of the files.
1 terms · Checked 2026-10-19
Describing software that uses built-in AI to think and decide
An adjective for software, tools and systems that have AI built in and let that AI think and make decisions as part of their work. So we say agentic software, an agentic system, an agentic tool or an agentic robot. An agent (the noun) finishes a job completely on its own, while something agentic uses AI to reason and decide but cannot reach a full result with AI alone. That is why Anthropic calls Claude Code "an agentic coding tool".
1 terms · Checked 2026-10-19
A tiny learning language model you build yourself in the browser
XSLM is a term SONUB coined itself, short for Extra Small Language Model. It is not a powerful large model but a model for learning: you type every line yourself, from the tokenizer, a tiny GPT, training and text generation to a Runner that serves the model and XChat, a chatbot that talks to it. Everything runs 100% inside the browser's Linux terminal, with no install, no pretrained weights and no external AI API.
No term matches your search.