Mods in Claude Code: draw panes, add commands, step into tool calls

Hooks, skills, status lines and MCP servers all extend Claude Code from the outside: they run a script, or give Claude more text or tools. But what if you want a pane beside the transcript charting how full your context is, or to hold an rm -rf with "proceed / cancel" buttons right in the interface? That's what a mod is for: a kind of plugin that runs inside Claude Code, written in JavaScript or TypeScript.

What mods can do that other tools can't

• Draw a usable interface: a pane beside the transcript or a band above the prompt, with tabs, buttons and text fields.
• Redraw built-in interface: change a tool call's row, the spinner, or the question dialog.
• Step into a tool call or request: hold a tool call to ask you something, answer it without running the tool, or send a request to a different model.
• Commands that run your code immediately: a /command that runs your function with no Claude turn, even while Claude is working.
• Share data between hooks: one hook counts, another displays.

What a mod looks like

// hooks/register.js: count tool calls and show them by the spinner
let calls = 0
export function register(on) {
  on('tool.call', async ($, e, next) => {
    calls += 1
    $.ui.invalidate('ui.render')
    return next(e)
  })
  on('ui.render', { component: 'Spinner' }, async ($, e, next) =>
    next({ ...e, props: { ...e.props, suffix: ' · tool calls: ' + calls + '…' } }))
}

A small mod has three files: .claude-plugin/plugin.json, hooks/hooks.json pointing to the code, and hooks/register.js. Each hook can observe an event, rewrite it, or answer it instead of the default behaviour. The fastest way to get a mod: describe what you want in a Claude Code session and let Claude write it.

Quick comparison

Installing and trying one

/plugin install token-chart@your-org     # install from a marketplace
claude plugin validate ./some-mod        # list its events and calls before installing
/plugin                                  # the "N mod active" line shows what loaded

Anthropic shares sample mods in the claude-code-playground repository, such as blast-radius, which holds risky shell commands and shows what they'd change before running. Some Claude Code features themselves, like /diff, are mods.

The risk: mods run with your permissions

• Mods are not sandboxed: they read and write files, start processes, make network requests, read environment variables and settings (including API keys), and see every prompt and tool call.
• A mod can rewrite prompts, submit prompts, and in some setups approve tool calls for you.
• Install only from authors and marketplaces you trust; run claude plugin validate to see its events and calls first.
• To turn them off: uninstall in /plugin, run --safe-mode for one session, or set "disableAllHooks": true.
• A mod can't change what a permission prompt shows you.

How to apply it

• Dev: Ask Claude to write a small mod that shows information you often need (tokens, branch, test status) beside the spinner.
• Tech Lead: Use a blast-radius-style mod to add a visual confirmation step for destructive commands across the team.
• PM/Admin: Set mod policy via managed settings; review mods like code before allowing them.