Claude Code Glossary
The Claude Code Glossary is a collection of terms used in 📝Claude Code, Anthropic's CLI for Claude, spanning built-in commands, configuration concepts, and the Brian-extended ecosystem layered on top.
Glossary
A
- 📝Claude Code Agent View —
claude agentsscreen for dispatching background sessions on your machine and seeing which are working, done, or waiting on you. - 📝AGENTS.md (Claude Code) — alternative instruction file read by Codex and other agents that don't load CLAUDE.md; commonly co-located for cross-tool compatibility.
- 📝Auto Mode — opt-in mode where Claude Code makes routine decisions autonomously and only surfaces taste-level calls for explicit approval.
B
- Background task — work the harness runs asynchronously via
run_in_background=true, notifying the agent on completion instead of blocking the conversation.
C
- 📝CLAUDE.md as Infrastructure — instruction file Claude Code auto-loads at session start; layered at project (
./CLAUDE.md), user (~/.claude/CLAUDE.md), and system levels. - Claude Agent SDK — SDK for building custom Claude-powered agents with their own tool sets, permissions, and orchestration logic.
- Claude Code — the subject of this glossary; available as terminal CLI, desktop app, web app, and IDE extensions for VS Code and JetBrains.
- 📝Codex Integration for Claude Code — skill that wraps OpenAI's Codex CLI as a second-opinion reviewer inside Claude Code.
- 📝Compaction — automatic summarization the harness performs when conversation tokens approach the model's context limit.
- 📝context window — total tokens the model can hold in a session; Opus 4.7 supports up to 1M with the
[1m]model variant. - 📝Claude Coordinator — the agent that scopes work, delegates it to workers, tracks their state, and assembles results; the project conversation in Claude Code Projects.
- 📝Claude Code Cross-Session Messaging — lets one session send plain text to another of your sessions with
ListAgentsandSendMessage, locally or via Remote Control (v2.1.224+). - 📝How to Message Between Claude Code Sessions — getting-started walkthrough.
D
--dangerously-skip-permissions— CLI flag that pre-approves every tool call for the session, used for fully autonomous runs.- 📝/dream — Brian's slash command for capturing aspirational or speculative ideas directly into the library.
- 📝Claude Code Dynamic Workflows — scripts Claude writes to orchestrate many subagents in the background; triggered with "use a workflow" or
ultracode, watched with/workflows.
E
- 📝Everything Claude Code (ECC) — Brian's curated index of every Claude Code resource worth knowing.
F
- Fast mode — toggled by
/fast; runs Opus with faster output, available on Opus 4.6 and 4.7.
G
- 📝/goal — Claude Code 2.1.139 slash command that locks the session on a completion condition across turns until a separate evaluator confirms the goal is met.
H
- 📝Headless Mode — non-interactive Claude Code invoked via
claude -p "<prompt>", used for scripting, CI, and automation. - 📝Hook — shell command the harness executes in response to lifecycle events like
SessionStartorUserPromptSubmit.
I
- IDE integration — Claude Code extensions for VS Code, JetBrains, and Cursor that surface the CLI inside the editor.
/init— built-in slash command that generates a starter CLAUDE.md for a new project.
K
- Keybindings — keyboard shortcuts configured in
~/.claude/keybindings.json, including chord bindings for multi-key shortcuts.
L
/loop— built-in skill that runs a prompt or slash command on a recurring interval, e.g./loop 5m /foo.
M
- 📝Model Context Protocol (MCP) — the standard for exposing tools, resources, and prompts to LLM agents.
- 📝MCP Server — process implementing MCP that Claude Code connects to via stdio or HTTP to add domain-specific tools.
- 📝Claude Code Memory — file-based persistence at
~/.claude/projects/<project>/memory/, indexed byMEMORY.md, that survives across conversations. - 📝Model — the specific Claude model in use, switched via
/model; current family is Opus 4.7, Sonnet 4.6, and Haiku 4.5. - 📝Claude Code Monitor Tool — built-in
Monitortool that streams output from a background task line by line. - 📝Claude Code Multi-Agent Orchestration — hub comparing every way Claude Code runs agents in parallel: subagents, workflows, agent view, agent teams, Projects, and cross-session messaging.
- 📝Claude Code Multi-Agent — Cheat Sheet — commands, settings, and gotchas on one page.
- 📝MythOS Claude Code Skill — Brian's skill connecting MythOS as an external memory and reference system for Claude Code.
O
- Output style — alternative system prompt (e.g., "Explanatory") that changes how Claude responds, selected via settings.
P
- 📝Permissions — granular allow/ask/deny rules in
settings.jsoncontrolling which tools, Bash commands, and MCP calls run without prompting. - 📝Plan Mode — read-only mode entered via
EnterPlanModewhere Claude proposes an approach before any writes, with user approval viaExitPlanMode. - 📝Claude Code Plugin — bundle of skills, slash commands, hooks, and subagents distributed as a single installable unit.
- 📝Claude Code Projects — one ongoing conversation at claude.ai/code where a coordinator runs related tasks as parallel threads with shared instructions and memory (public beta, Pro and Max).
- 📝How to Start a Project in Claude Code — getting-started walkthrough.
- 📝Prompt Caching — Anthropic API feature that caches a stable prompt prefix across requests with a ~5-minute TTL (one hour via
ENABLE_PROMPT_CACHING_1H).
R
- 📝Ralph Loop — worker-judge-loop execution pattern implemented by
/goalthat locks the agent on a target across turns until the goal is met. /remember— built-in slash command that prompts Claude to save a piece of information to memory.- 📝Claude Code Routines — recurring remote agents that execute on a cron schedule.
S
/schedule— built-in skill for creating, updating, listing, or running scheduled remote agents.- 📝/scheduled — slash command surfacing scheduled jobs registered with the harness.
- 📝Claude Code Scheduled Triggers — pattern for triggering Claude Code runs from external events.
SendMessage/ListAgents— built-in tools for messaging a subagent, teammate, or another of your sessions, and for finding which ones are reachable;/list-agents(alias/peers) shows the same list.- Session — a single conversation instance with its own context, transcript, and memory access.
- 📝Claude Code Sessions from Dispatch — pattern for spawning Claude Code sessions from a central dispatcher.
- 📝settings.json — configuration file at
~/.claude/settings.json(or.claude/settings.local.jsonper-project) controlling permissions, hooks, env vars, and MCP servers. - 📝Skill (Claude Code) — file-based capability with YAML frontmatter, invoked via the
Skilltool when its description matches the task. - 📝Slash Command — user-typed
/foothat invokes a built-in (like/helpor/clear) or a skill registered via plugin. - 📝Claude Code Subagents — delegated workers inside one session that run a side task in their own context window and return a summary; background by default.
- 📝Claude Code Forked Sub-Agents — subagents that inherit the full conversation and prompt cache; started with
/subtask, on by default since v2.1.232. - 📝How to Dispatch Subagents in Claude Code — getting-started walkthrough.
- 📝Claude Code Forked Sub-Agents — subagents that inherit the full conversation and prompt cache; started with
- 📝Claude Code Swarm Mode — Anthropic's agent teams: a lead session spawns teammates that share a task list and message each other; experimental, enabled with
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.
T
- TaskCreate / TodoWrite — built-in tools for breaking work into trackable tasks with pending, in-progress, and completed states.
- 📝Tool — a capability Claude can invoke, whether built-in (Read, Write, Edit, Bash), MCP-provided, or skill-provided.
U
/ultrareview— user-triggered multi-agent cloud review of the current branch or a specified PR.- 📝Ultrathink — keyword in a user message that requests high effort (deeper reasoning) for that turn; reintroduced in v2.1.68.
W
- WebFetch / WebSearch — built-in tools for retrieving URL content and running web searches.
- 📝Worktree (Claude Code) — isolated git worktree created for agent work, spawned via
isolation: "worktree"in theAgenttool.
