Claude Code

A Claude Code launch option that loads a plugin straight from a folder on your computer for one session, so you can try it while you're still building it.

Typing @ followed by a file's name in a Claude Code prompt to point Claude at that exact file, so it doesn't have to search for it.

The repeating cycle Claude Code works in: look around the project, take one step such as editing a file or running a command, check what happened, and decide the next step until the task is done.

A restore point Claude Code saves before each edit Claude makes, so you can run /rewind (or press Esc twice) to go back to an earlier moment. It only tracks Claude's own file edits, so it doesn't replace Git.

A placeholder, written `${CLAUDE_PLUGIN_ROOT}`, that Claude Code swaps for the folder a plugin is installed in, so the plugin's file paths work on everyone's computer. That folder changes with each update, so anything the plugin needs to keep goes in `${CLAUDE_PLUGIN_DATA}` instead.

A notes file in your project that Claude Code reads at the start of every session, holding the rules and commands you'd otherwise repeat in every prompt (/init writes a starter version).

Shrinking a long Claude Code conversation into a short summary so the session can keep going when it nearly fills Claude's context window; it happens automatically, or you can run /compact yourself.

Announcing that a feature will be removed later while it still works, and pointing people to its replacement so they have time to switch.

A short block of settings at the very top of a file such as SKILL.md or a subagent file, written in YAML between two lines of three dashes, that tells Claude Code things like the file's name and when to use it.

Running Claude Code as a single command, `claude -p "..."`, so it does one task, prints the answer, and exits without opening a chat, which lets scripts and scheduled jobs use it.

A command you set up in Claude Code's settings that runs automatically every time a certain event happens, such as right after Claude edits a file, so a rule is followed without anyone having to remember it.

Pressing Esc while Claude Code is working to stop it mid-step so you can correct course, without losing the conversation so far (it doesn't undo edits already made; /rewind does that).

A connector that lets Claude Code use an outside service, such as GitHub, Slack, or a calendar, by giving Claude a set of tools for it through the Model Context Protocol (MCP).

Choosing which Claude model handles a task, such as a quick, low-cost one like Haiku for small edits or a more capable one like Opus for tricky work, which you switch in Claude Code with /model.

The Claude Code setting that decides how much Claude may do without asking you first, from asking before every edit, to editing freely, to only planning; press Shift+Tab to switch between them.

A line in Claude Code's settings that always allows, always blocks, or always asks before a specific action, such as running your tests, and Claude Code enforces it no matter what a prompt says.

A Claude Code mode where Claude reads your project and writes a step-by-step plan for you to approve, without changing any files until you say yes; press Shift+Tab (or type /plan) to turn it on.

A package that bundles several Claude Code add-ons, such as skills, subagents, and hooks, so the whole set can be installed, updated, or removed in one step.

The small `plugin.json` file inside a plugin's `.claude-plugin` folder that gives the plugin its name and, usually, a version number and a short description.

A catalog of Claude Code plugins, like an app store for your team, that you add once and then install plugins from by name.

An attack that hides instructions inside something Claude reads, such as a web page, an email, or a document, hoping Claude will obey them instead of you.

Reopening an earlier Claude Code conversation with its history intact, using `claude --continue` for the most recent one or /resume to pick from older ones.

A way of numbering versions as MAJOR.MINOR.PATCH, where the number you raise tells people what changed: PATCH for fixes, MINOR for new features, and MAJOR for changes that break how people already use it.

The file that defines a skill: a short frontmatter block naming the skill and saying when to use it, followed by the step-by-step instructions Claude follows.

Saved sets of instructions, each in its own folder, that teach Claude a specific workflow; Claude loads one on its own when your request matches it, or you run it by typing a slash and its name.

Short commands you type in Claude Code starting with a slash, such as /compact or /model, that run a built-in action or one of your project's skills.

Helper Claude sessions that the main conversation hands one focused job, such as searching the whole project, which work separately and report back only a short summary so your main conversation stays uncluttered.