The Claude Code slash commands a website team actually uses

Claude Code has dozens of built-in commands, but a website team leans on a short list of them week to week. You type a command at the start of a message, beginning with a slash: /init to write a starter CLAUDE.md, /plan before a big change, /diff and /code-review before anything ships, /compact or /clear when a session gets long, and /rewind when something goes wrong. The rest are worth knowing exist, not memorizing.

Anthropic now calls these "commands" rather than "slash commands" (its glossary notes "'Slash' dropped from product copy"), though everyone still types the slash. This guide groups the commands by the point in a website job where you need them, as of October 2026. Claude Code ships changes weekly, so type / in your own session to see the current list, or check Anthropic's commands reference.

How commands work

Type / to open the command menu, then keep typing to filter it. A few rules from the docs are worth knowing:

• A command is recognized only at the start of your message. Text after the name is passed to it.
• If you send a command while Claude is still responding, Claude Code queues it until the turn finishes. A few, such as /status and /usage, run straight away.
• Commands you can't use on your plan or platform are left out of the menu.
• Commands change. /ultraplan, /pr-comments and /vim have been removed (use plan mode instead of /ultraplan), and /agents now prints a reminder to ask Claude or edit the agents folder rather than opening a manager. Older tutorials may still show them.
• Starting a message with ! instead runs a shell command directly, such as ! git status or ! npm run build, and adds its output to the conversation without Claude needing to interpret or approve it.

The commands come in two kinds. Most are built in, with fixed behavior. Others are bundled skills: prompts Claude follows, which means their results vary like any other Claude output and need the same review. Your team can add its own commands, such as a /publish-post checklist, by writing a skill; see Claude Code skills explained. If a term here is new, our Claude Code glossary defines it in plain English.

Setting up a website project

These run once when you start using Claude Code on a site, then occasionally after that:

/init gives you a draft, not a finished file. What a website CLAUDE.md should contain is covered in how to write a CLAUDE.md for a website, and copy-ready settings files are in our Claude Code templates.

Planning and steering a task

• /plan switches into plan mode, where Claude researches and proposes changes without editing your source files. Add a description to start straight away: /plan add FAQ schema to every service page. Use it before any change that touches more than a page or two. Our guide to plan mode covers how to review a plan.
• /model switches the model and saves it as your default for new sessions. In the picker, press s on a row to switch for this session only.
• /effort sets how much reasoning Claude applies, from low to xhigh, max, or auto. Anthropic's glossary puts the trade-off simply: higher effort means deeper reasoning, and lower effort is faster and cheaper.
• /btw asks a side question about the session without adding it to the conversation, which is handy for "what does this function do?" in the middle of a long job.

Press Esc at any point to stop Claude mid-step. Esc is a key, not a command, but it's the one you'll use most.

Keeping long sessions on track

Every session has a context window, the working memory Claude reads on each turn. Website jobs that read many pages fill it fast. Four commands manage it:

• /context shows what's filling the window as a colored grid, with suggestions for what to trim.
• /compact summarizes the conversation so far to free space. You can tell it what to keep: /compact keep the list of pages still to fix. Instructions given only in conversation can be lost in a summary, so put lasting rules in CLAUDE.md.
• /clear starts a new conversation with empty context while keeping project memory. Aliases are /reset and /new.
• /resume returns to an earlier conversation, and /rename gives the current one a name you'll recognize in the picker.

Our house rule on eseospace.com is one task per session, cleared between unrelated jobs. A session that rewrote meta descriptions shouldn't carry that context into fixing a contact form. How compaction and memory files work in detail is in context, compaction and memory.

Reviewing before anything ships

These are the commands that protect your live site. Use them on every change, in this order:

1. /diff shows the changes in your working tree, including Claude's edits and anything else you haven't committed. Read it all. If it shows changes you didn't ask for, find out where they came from before going further. When we rebuilt eseospace.com, two sessions working on the same site at once meant one regenerated pages the other had just edited, and those edits disappeared. Checking for changes you didn't make before a deploy is now a standing rule.
2. /code-review reviews the current diff for correctness bugs. Add --fix to have Claude apply the findings, or pass a pull request number to review that instead. /review is an alias.
3. /security-review analyzes the changes on your branch for security vulnerabilities such as injection, authentication issues and data exposure. Run it on anything that touches forms, logins, payments or user input. It needs an origin remote, because it compares your branch with origin's default branch.
4. /verify, a bundled skill, confirms a change works by building your project's app, running it and observing the result, rather than relying on tests alone.

A clean review from Claude is a second opinion, not an approval. A person still reads the diff, looks at the rendered page and decides what ships. Anthropic's Code Review page covers the separate, automatic pull request review for Team and Enterprise, which is in research preview and billed through usage credits.

Undoing a change

/rewind opens the rewind menu, where you can restore the code, the conversation or both to an earlier point. Pressing Esc twice on an empty prompt does the same. Aliases are /checkpoint and /undo.

Know the limits before you rely on it. Checkpoints track Claude's file edits, not changes made through shell commands, and actions on remote systems, such as a deploy, a database update or a CMS API call, can't be rewound. Git is still your real undo. The details are in checkpoints and rewind.

Watching usage and cost

• /usage shows session cost, plan usage limits and activity stats. /cost and /stats are aliases. On a Pro, Max, Team or Enterprise plan it breaks down what counts against your plan limits; the session dollar figure is intended for API users.
• /usage-credits opens your usage-credits billing settings when you hit a limit. Team and Enterprise members without billing access send a request to their admin instead.
• /status shows your version, model, account and connectivity, which is useful when two teammates see different behavior.

Bundled skills worth knowing

A few bundled skills fit website work well, with care:

• /batch handles large changes across a codebase. It researches, splits the work into 5 to 30 independent units, presents a plan, and once you approve, runs each unit in its own git worktree with a background subagent. That suits a sitewide change such as updating a component on every template, but it produces many changes at once, so plan the review before you start.
• /loop runs a prompt repeatedly while the session stays open, for example /loop 5m check if the deploy finished.
• /debug turns on debug logging and reads the session log to troubleshoot Claude Code itself.
• /fewer-permission-prompts scans your transcripts for common read-only commands and adds an allowlist to the project's .claude/settings.json. Read what it adds before you commit it.

A cheat sheet for website jobs

Conclusion

You don't need the whole command list. A website team that uses /plan before big changes, reads /diff and runs /code-review before every pull request, clears context between jobs and knows what /rewind can't undo is using Claude Code well. Pair those habits with git, a staging copy and a person approving each deploy. If you want your team trained on a working routine like this, our Claude Code consulting includes setup and training. More guides are in our Claude Code Guide.

Frequently asked questions