Claude Code commands and slash commands, for teams.
Claude Code commands start with a slash: built-in ones like /init, /model and /permissions, and custom ones you write as Markdown files, which are now skills. Commit them to .claude/skills and everyone on the repo gets them (Anthropic's docs, October 2026).
Type / in a session to see them all. Anthropic's commands reference lists more than a hundred, including aliases and bundled skills, and not every one appears for every user. The ones a team leans on:
/init writes a starting CLAUDE.md for the project, and /memory edits CLAUDE.md files and auto memory.
/model switches the model and saves it as your default for new sessions.
/permissions manages the allow, ask and deny rules for tools (more below).
/usage shows the session's cost, your plan's usage limits and activity; /cost and /stats are its aliases.
/context draws context use as a grid, /compact frees context by summarizing the conversation, and /clear starts over with empty context.
/code-review (alias /review) checks the current diff, a branch or a pull request for correctness bugs, and /security-review looks for vulnerabilities in the branch.
/plan enters plan mode, /resume reopens an earlier session and /rewind rolls the conversation or the code back.
/mcp, /plugin, /hooks and /skills manage MCP servers and plugins and list the hooks and skills in effect.
Custom slash commands vs skills
Anthropic has merged custom commands into skills. A file at .claude/commands/deploy.md and a skill at .claude/skills/deploy/SKILL.md both create /deploy and work the same way. Command files are the older format and keep working, but the docs suggest a skill for new work, and when both share a name, the skill wins. More on how Claude picks one in Claude Code skills.
Project: .claude/skills/<name>/SKILL.md, or .claude/commands/<name>.md, committed with the repository. A subfolder adds a prefix: .claude/commands/frontend/component.md is /frontend:component.
Personal: ~/.claude/skills/<name>/SKILL.md, in every project on your machine.
Arguments: $ARGUMENTS is everything typed after the command. $0 is the first argument and $1 the second, since they count from zero.
Frontmatter: description (the one field the docs recommend), argument-hint for autocomplete, and model. disable-model-invocation: true keeps Claude from running it on its own, so it runs only when someone types it.
How do I allow all commands in Claude Code?
Start narrow. Permission rules in settings.json allow the commands you trust and nothing else, for example Bash(npm run *) and Bash(git commit *) under permissions.allow. A bare Bash rule allows every shell command. Claude Code checks deny rules first, then ask, then allow, and a deny rule blocks in every mode. /permissions edits the rules, and --allowedTools sets them for one run.
Permission modes go further. Shift+Tab cycles them in a session:
default asks before anything beyond reading files; acceptEdits also lets Claude edit files without asking; plan only plans.
auto runs everything with background safety checks and is now the starting mode in the terminal and VS Code. Anthropic says it reduces prompts but does not guarantee safety.
dontAsk denies whatever would prompt, for locked-down CI and scripts.
bypassPermissions runs everything without asking. --dangerously-skip-permissions turns it on.
Is --dangerously-skip-permissions safe?
Only in a sandbox. Anthropic's docs say to use it only in isolated environments, such as containers, VMs or dev containers without internet access, where Claude Code can't damage your host, and note that it offers no protection against prompt injection. It won't run as root.
A team can switch it off. Set permissions.disableBypassPermissionsMode to disable in managed settings and Claude Code rejects the flag; disableAutoMode does the same for auto mode. The bypass and auto defaults also can't come from a repository's settings: they take effect only from your own ~/.claude/settings.json.
Sharing commands across a team
Commit .claude/skills/ (and any .claude/commands/) so everyone who clones the repository gets the same commands. The committed .claude/settings.json carries the team's permission rules, hooks and plugins with them.
For more than one repository, ship the commands in a plugin. Its skills run as /plugin-name:skill-name, and a marketplace listed under extraKnownMarketplaces in the project's settings is offered to everyone who opens the repo. Org-wide skills and rules can also go out through managed settings. MCP servers add commands too: each MCP prompt shows up as /server:prompt.
Which commands your team actually runs
A command in the repository isn't one anybody uses. Zest's Claude Code plugin records every custom command and skill a session runs, by name, whether someone typed /name or Claude picked it. Built-in commands such as /clear and /compact, and Zest's own /zest: commands, are left out.
The AI Stack view lists each engineer's commands and skills next to their models and MCP tools, so a lead sees which ones spread across the team and which stay with their author. What else the plugin records, and how, is on Claude Code usage and Claude Code hooks.
TRY IT ON YOUR TEAM
See what your team ships with AI.
Starter is free (3 seats, no credit card). Pro is $49 a month with 5 seats and Max is $149 a month with 10 seats — priced per workspace, not per engineer, with extra seats at $19 a month.
Yes. List them in the allowed-tools field of its frontmatter, such as Bash(git add *): Claude can use those tools without asking during the turn that runs the command. Your deny rules still apply.
Do Claude Code custom commands work in Codex?+
A skill does: Codex reads the same SKILL.md format from .agents/skills, and you run it as $skill-name. Files in .claude/commands are Claude Code's own format.
