Skip to main content
ALL GUIDES
6 min readUpdated

Claude Code Genius Mode: 40 Commands Most People Never Type

The 8 from the video plus 32 more, grouped by what they're for: context and tokens, session control, code quality, background work, git, config and debugging. All from the official CLI reference.

The 8 from the video, plus 32 more. Every command here is from the official Claude Code CLI reference. Nothing invented.


The 8 from the video

/advisor [model|off] Turns on the advisor tool, which consults a second, stronger model when a call is hard enough to be worth it. Reach for it when a decision feels risky and you want a sanity check before Claude commits to an approach.

/plugin Opens the plugin marketplace so you install ready-made sets of skills instead of writing your own. Reach for it before you build a workflow from scratch. Someone probably already packaged it.

/schedule Creates a repeating task that runs in the cloud, so your laptop doesn't have to stay open. Reach for it for daily summaries, recurring reports, anything on a cron.

/btw [question] Asks a side question mid-task. Claude answers without dropping the work it's already doing. Reach for it when you need a quick clarification and don't want to interrupt a long run.

/remote-control Runs your session from your phone, approvals included. Reach for it when a long task needs one yes to keep going and you're away from your desk.

/insights Generates an HTML report on your recent sessions: projects, usage patterns, errors, features you've never touched. Reach for it monthly. It's the one that actually changes how you work.

/clear [name] Starts a fresh conversation with empty context and labels the old one. Project memory stays. Reach for it the moment a task is finished, so you stop paying to carry it.

/resume [name] Returns to an earlier session by name. Reach for it instead of re-explaining yesterday's work.


Context and tokens

/context [all] — Shows current context usage as a colored grid with suggestions on what to cut. Use it before you hit the wall, not after.

/compact [instructions] — Summarizes the conversation to free up context while keeping key decisions. Pass instructions to control what survives.

/autocompact [auto|<tokens>] — Sets the threshold for automatic compaction, so long sessions manage themselves.

/focus — Toggles a stripped view showing only the last prompt, a tool-call summary, and the final response.

/goal [condition] — Sets a success condition Claude keeps working toward without you re-prompting.


Session control

/cd <path> — Moves the session to a different directory without losing conversation or prompt cache.

/add-dir <path> — Adds a second working directory so you can reach files across folders in one session.

/branch [name] — Branches the conversation at the current point so you can try another approach and keep the original.

/teleport — Pulls a web session from claude.ai/code down into your local terminal.

/export [filename] — Exports the whole conversation to a file or clipboard.

/copy [N] — Copies the last response, or the Nth latest, to your clipboard.


Code quality

/code-review [level] [--fix] [--comment] [target] — Reviews your diff, a PR, a branch, or a path. Levels run low through max, and ultra hands it to a cloud review agent.

/simplify [--fix] — Finds simplifications in working-but-verbose code. --fix applies them.

/verify — Runs tests, type checks, and lints to confirm a change actually holds.

/trace — Runs code with print statements inserted so you can follow execution without a debugger.

/diff — Interactive diff viewer for uncommitted changes, turn by turn.


Background work

/background [prompt] (or /bg) — Detaches the session to run as a background agent and gives your terminal back.

/subtask <instruction> — Hands a side task to a subagent that reports back into the conversation.

/loop [interval] [prompt] — Runs a prompt on a repeating interval, e.g. /loop 5m check deploy status.

/tasks — Lists everything running or finished in this session, including subagents.

/autofix-pr [prompt] — Spawns a cloud session that watches your PR and pushes fixes when CI fails or a reviewer comments.


Git

/switch-branch <branch> — Stashes, switches branch, and merges your work onto the new one.

/rewind [turn|time] — Rolls code and conversation back to an earlier checkpoint.


Config and setup

/doctor — Setup checkup: install health, unused tools, slow hooks, version checks, CLAUDE.md deduplication.

/permissions — Manages allow, ask, and deny rules for tools and directories.

/fewer-permission-prompts — Scans your transcripts for repeated read-only calls and builds an allowlist from them.

/trust-workspace [path] — Trusts a workspace so routine operations stop prompting.

/memory — Edits CLAUDE.md, toggles auto memory, and shows what Claude has learned about your codebase.

/hooks — Shows what's configured to run on each tool event.

/mcp [reconnect <server>|enable|disable] — Manages MCP server connections and re-auth.

/effort [level|auto] — Sets model effort from low through max. Drop it for fast iteration, raise it when accuracy matters more than speed.


Debugging and odds and ends

/debug [description] — Turns on verbose logging and reads the session debug log back to you.

/plan [description] — Enters plan mode straight from the prompt.

/recap — One-line summary of the session, on demand.

/when — Shows when you last ran each command in this project.


Source: the official Claude Code CLI command reference. Two names get misheard often, so for the record it's /remote-control (not /rec) and /insights (plural).


Using Claude Code to build things rather than just drive it? Website to video with Claude Code + Remotion is the end-to-end version.

@gracefrank.ai