C9PG
Claude Code Docs

Glossary

This glossary comes from the expanded vibe-coding guide and is intended to give teams a stable working vocabulary when they talk about Claude Code in design reviews, PRs, and onboarding.

TermDefinition
Plan ModeRead-only mode where Claude researches and analyzes without making edits. Toggle via Shift+Tab or /plan.
Auto-Accept ModeClaude proceeds without asking for edit or run permission. Use it only when you trust the current task scope.
Auto ModeAI classifier evaluates each tool call, blocking dangerous operations and allowing safe ones. Launched March 24, 2026.
RED TokenA state file at .claude/state/test-forward.json proving a failing verification existed before implementation.
RED → GREEN → REFACTORTDD cycle: write a failing test, implement the minimum to pass, then clean up.
PreToolUse HookFires before a tool call executes. The primary enforcement mechanism.
PostToolUse HookFires after a tool completes successfully. Used for formatting, tests, and feedback.
Stop HookFires when Claude finishes responding. Agent-type Stop hooks can inspect files and block completion.
PermissionDenied HookAdded in v2.1.89. Fires after Auto Mode classifier denials and can trigger retries.
Agent TeamsCoordinated multi-instance development with shared task lists and inter-agent messaging.
SubagentA focused worker spawned within a session with its own context window.
SkillsSKILL.md instruction files that teach Claude specific behaviors.
RulesMarkdown files in .claude/rules/ that scope instructions to specific file globs or directories.
HooksDeterministic commands, prompts, HTTP handlers, or agents triggered at lifecycle points.
Context WindowThe total conversation, files read, and command output Claude holds. The most important resource to manage.
CompactionCompressing the context window via /compact while retaining the important state.
WorktreeAn isolated git working copy used for parallel, conflict-free development.
Remote Control/rc. Synchronization between the local session and Claude mobile app or web. The session still runs locally.
Superpowersv5.0.6 plugin framework that layers brainstorming, planning, TDD, debugging, and code review skills on top of Claude Code.
1% RuleSuperpowers protocol in which Claude checks for relevant skills before any response.
opusplanModel alias using Opus for planning and Sonnet for execution.
deferv2.1.89 PreToolUse permission decision that pauses headless sessions for human review.
/batchParallel codebase-wide migration workflow that decomposes work into isolated units.
/simplifyBundled skill that spawns three review agents focused on reuse, quality, and efficiency.
/loopSession-level cron that fires prompts on a schedule while the session stays open.
MCPModel Context Protocol. Connects Claude Code to external tools such as GitHub, Playwright, Figma, or databases.
Elicitationv2.1.76 capability that allows MCP servers to display forms or open URLs to gather structured input.
Computer UseMarch 2026 feature allowing Claude to interact with the screen UI directly.
ChannelsResearch preview for forwarding chat streams such as Telegram, Discord, or iMessage into a session.

Why the glossary matters

The expanded guide treats vocabulary as an operational tool, not filler. Teams that can distinguish Plan Mode, Auto-Accept, Auto Mode, defer, Stop hooks, subagents, and Agent Teams tend to design cleaner workflows and make fewer bad assumptions about what Claude is actually allowed to do.