The commands I reach for most. Full reference below.
| Command | What it does |
|---|---|
cskip |
Launch Claude with permissions skipped — the daily driver |
cbrain |
Launch Claude inside your 2ndBrain vault (requires vault setup — see 2ndBrain-mogging) |
g2 |
Tile 2 Ghostty windows side by side (macOS) |
/fswarm <task> |
Launch a 15-agent FidgetFlo swarm — describe the task in plain English |
/fmini <task> |
Compact 5-agent FidgetFlo swarm for focused work |
/w4w |
Word-for-word, line-for-line. Max attention, zero skipping, no summarizing |
/concise |
Chat default — no fluff, no scaffolding, no headers on simple Qs. Suspends for copy/scripts/decks. |
/bullets |
Crush a paragraph into the shortest scannable bullets — core facts only. Always bulletizes (where /concise sometimes leaves prose). For handoffs |
/safetycheck |
Security audit — scans for exposed keys, injection vectors, supply-chain risks |
/gitfix |
Full repo sync — reads every file, fixes doc drift, makes reality match the README |
/recon <thing> |
Pre-build prior-art sweep — ranks competitors, finds the edge, GREEN/YELLOW/RED verdict before you build |
/verify-before-claim |
Check config + billing before answering — MCPs installed, what a plan costs, whether a model is real |
/hypersearch <question> |
Deep research — multi-step source sweep with adversarial critics (needs hyperresearch, Step 3) |
/osmani-build |
Phase-gated product-build lifecycle (define → plan → build → verify → review → ship) orchestrating the addyosmani/agent-skills plugin — offers itself when you start building a real product |
/save |
Capture a conversation into your 2ndBrain vault (requires 2ndBrain-mogging) |
The terminal works differently from a normal text editor or document. Here's what you need to know:
Typing and editing:
- You type commands at the blinking cursor and press Enter to run them
- You can't click to place your cursor in the middle of a line — the terminal isn't a text editor
- Highlight and delete doesn't work — forget what you know from Google Docs or Excel
- Backspace deletes one character at a time (that still works normally)
- Ctrl+U clears the entire line — this is your best friend when you mess up a command
Moving around:
- Left/Right arrows move your cursor within the line you're typing
- Up arrow brings back the last command you ran (press it again for the one before that)
- Down arrow goes forward through your command history
Important keys:
- Ctrl+C stops whatever is currently running (note: Ctrl, not Cmd on Mac)
- Enter confirms and runs a command
- Tab tries to auto-complete file names and commands
When the terminal asks you something:
[y/n]means type y for yes or n for no, then press EnterPress ENTER to continuemeans just hit Enter- If something looks stuck, try pressing Enter — it might be waiting for you
These aliases are added to your ~/.zshrc (or ~/.bashrc) and available in any terminal session.
| Command | What it does |
|---|---|
claude |
Normal mode — asks permission before each action |
cskip |
Launch Claude Code with all permissions skipped (claude --dangerously-skip-permissions) |
cc |
Short alias for claude |
ccr |
Resume last Claude conversation (claude --resume) |
ccc |
Continue last Claude conversation (claude --continue) |
ctg |
Skip-permissions + Telegram channel connected (any directory) |
cbrain |
Launch Claude Code in your 2ndBrain vault with skip-permissions (installed by 2ndBrain-mogging — not Step 1) |
cbraintg |
Same as cbrain but with Telegram channel connected (installed by 2ndBrain-mogging) |
g2 |
Tile 2 Ghostty windows side by side, filling your screen (requires Ghostty — Step 2, macOS only) |
g4 |
Tile 4 Ghostty windows in a 2x2 grid (requires Ghostty — Step 2, macOS only) |
Tip: After running any setup script, run
source ~/.zshrcto activate new commands. The scripts do this automatically, but just in case.Note: Until you set up Second Brain (2ndBrain-mogging), use
cskipinstead ofcbrain. Thecbraincommand requires an Obsidian vault to exist — if you haven't created one yet, it will error. Everything else works right away withcskip.
cskip runs claude --dangerously-skip-permissions. This tells Claude to execute commands, edit files, and make changes without asking you first. It's faster for setup scripts and guided sessions. Use normal mode (claude) when you want to review each action before it happens.
Tip: Press Shift+Tab inside a Claude session to toggle permissions on/off without restarting.
These work inside your terminal when a Claude session is active.
| Hotkey | What it does |
|---|---|
| Shift+Tab | Toggle auto-approve permissions on/off without restarting Claude |
| Ctrl+C | Stop whatever is running, exit Claude, or clear everything you've typed in the input box |
| Ctrl+U | Clear the entire line you are typing |
| Up Arrow | Recall previous command from history |
| Tab | Auto-complete file names and commands |
| Command | What it does |
|---|---|
/exit |
Quit Claude |
/help |
Show all available commands |
/permissions |
Check current permission settings |
/clear |
Clear the conversation |
/compact |
Summarize conversation to save context window |
/voice |
Enable voice input mode |
/resume |
Pick up right where you left off — reloads your last session's context |
!command |
Run a shell command without leaving Claude (e.g., !ls) |
These are custom skills installed by the setup scripts. Type them inside a Claude session.
| Command | Installed in | What it does |
|---|---|---|
/fswarm do the thing |
Step 4 | Launch a 15-agent FidgetFlo swarm — just describe what you want in plain English after /fswarm |
/fmini do the thing |
Step 4 | Launch a compact 5-agent FidgetFlo swarm — same power, tighter team |
/fswarm1 <task> |
Step 4 | 15-agent swarm with light extended thinking (~4k budget per agent) — Think. appended to every Agent prompt |
/fswarm2 <task> |
Step 4 | 15-agent swarm with hard/deep thinking (~10k budget per agent) — Think hard. appended |
/fswarm3 <task> |
Step 4 | 15-agent swarm with harder/deeper thinking (~31k budget per agent) — Think harder. appended |
/fswarmmax <task> |
Step 4 | 15-agent swarm at MAX thinking (~32k budget per agent) — Ultrathink. appended |
/fmini1 <task> |
Step 4 | 5-agent swarm with light extended thinking (~4k budget per agent) — Think. appended |
/fmini2 <task> |
Step 4 | 5-agent swarm with hard/deep thinking (~10k budget per agent) — Think hard. appended |
/fmini3 <task> |
Step 4 | 5-agent swarm with harder/deeper thinking (~31k budget per agent) — Think harder. appended |
/fminimax <task> |
Step 4 | 5-agent swarm at MAX thinking (~32k budget per agent) — Ultrathink. appended |
/fhive <goal> |
Step 4 | Launch a queen-led autonomous FidgetFlo hive-mind with raft consensus |
/w4w |
Step 4 | Maximum attention to detail — word for word, line for line. No skipping, no summarizing. Also works without the slash — just type w4w |
/concise |
Step 4 | Default chat shape — no fluff, no scaffolding, no sycophancy, no headers on simple questions. Suspends automatically for copywriting deliverables (tweets, scripts, decks, client docs). For always-on enforcement see the playbook in concise-skill/SKILL.md description. |
/bullets |
Step 4 | Crush a prose blob into the shortest possible bullets — one fact per bullet, ≤~12 words, zero fluff, signal preserved (names, numbers, dates, paths, decisions). Always outputs bullets, unlike /concise which decides shape per reply and sometimes leaves prose. Built for document handoffs where nobody should have to read paragraph blobs. Also triggers in plain English — "bullet this", "bulletize", "turn this into bullets" |
/gitfix |
Step 7 | Full repo sync — reads every install script, skill file, and doc in the repo, finds every inconsistency between the code and the documentation, and fixes all of it. Run this any time you've made changes to a repo and need the README, cheatsheet, and all other docs to reflect reality. Also responds to "fix the github", "sync the repo", or "update the readme" in plain English |
/recon |
Step 7 | Pre-build prior-art recon. Before you build a tool/app/CLI/MCP/library, sweeps GitHub (via gh) + the web for what already exists, ranks the top ~10 free and paid competitors in a comparison table, finds where an edge exists (or honestly calls it a red ocean), and ends with a GREEN/YELLOW/RED verdict. Output is a discussion — building starts only after. Auto-offers itself when you say "build / make / start a new X" or "does X exist?"; run it directly with /recon <thing> to skip the offer |
/osmani-build |
Step 7 | Phase-gated product-build lifecycle over the addyosmani/agent-skills Claude Code plugin (24 skills: define → plan → build → verify → review → ship). Auto-offers itself (alongside /recon) when you start building a real tool/app/webapp/CLI/MCP/API — not creative ideation. Greenfield builds enter at Define (requirements interview → idea refine → spec); builds already in motion never get retroactive planning — they enter at Verify → Review → Ship. Run /osmani-build <phase> to jump straight to a phase |
/safetycheck |
Step 8 | Security audit — scans any project for exposed keys, missing rate limiting, input sanitization gaps, dependency vulnerabilities, and insecure configurations. Also responds to "run a safety check" in plain English. Auto-activates 12 MCP-specific checks on MCP projects |
/tailssh |
manual (tailssh-skill/) |
Permanent passwordless ssh <name> into someone else's Mac over Tailscale. One-directional device share (they can't see your machines), ~/.ssh/config alias, ssh-copy-id public-key install, end-to-end verify — plus generates a plain-English instructions doc the machine owner can follow solo. Also responds to "set up ssh to 's laptop" / "onboard 's machine" |
2ndBrain-mogging skills (requires 2ndBrain-mogging installed)
| Command | What it does |
|---|---|
/save |
Capture a conversation, passage, or ADR into the vault — alias-driven classification, commits with [bot:save] |
/wiki |
Add / audit / heal / find across the vault — ingest sources, repair dead links, run graph audits |
/challenge |
Adversarial vault agent — argues against an idea using your own past notes and memory |
/emerge |
Surface rising topics, killed ideas, and new links across a time window |
/backfill |
Ingest historical Claude CLI transcripts into the vault with chunked summarization + dedupe |
/aliases |
Manage the alias classifier — add, edit, or audit entries in Claude-Memory/aliases.yaml |
/autoresearch |
Fetch a URL, summarize to 02-Sources/, spawn linked concept stubs in 03-Concepts/ |
/canvas |
Generate/maintain Obsidian Canvas files from vault queries |
/tether |
Repair orphaned notes, bidirectionally link projects and hubs |
/connect |
Bridge two notes — surfaces structural analogies, transfer opportunities, collision ideas |
These are explicit triggers — you type the command to activate the skill. This is different from the auto-triggered tools below, which respond to natural language. Exceptions:
/w4walso works without the slash (just typew4wanywhere in your message),/safetycheckresponds to "run a safety check",/gitfixresponds to "fix the github" / "sync the repo" / "update the readme",/reconand/osmani-buildauto-offer themselves (one combined question) whenever you say "build / make / start a new X" or "does X exist?", and/bulletsresponds to "bullet this" / "bulletize" / "turn this into bullets". All other slash commands require you to type the command.
These activate on their own when Claude detects a relevant task via natural language. You never type a command — just describe what you want and Claude picks up the right tool.
| Tool | Installed in | How it activates | Example prompt |
|---|---|---|---|
| UI/UX Pro Max | creativity-maxxing | Natural language — asks about UI, design, layouts, interfaces | "Build me a dashboard with a sidebar" |
| Taste Skill (8 variants) | creativity-maxxing | Natural language — anything frontend/design. Stops generic AI "slop" output. Name a variant to force it: "use minimalist-ui", "redesign this with high-end-visual-design" | "Build a premium landing page" · "Redesign this dashboard" |
| Remotion | creativity-maxxing | Natural language — video, animation, motion graphics | "Make a 30-second intro video" |
| No-flicker mode | Step 3 | Automatic — fullscreen rendering, no screen jumping while Claude works | (always on — set via environment variable) |
| Memory auto-save hook | Step 3 | Automatic on session end — saves context from the conversation to memory | (no prompt needed — runs automatically) |
| Notion | Step 5 | Natural language — pages, databases, knowledge management | "Search my Notion for the meeting notes" |
| Granola | Step 5 | Natural language — meeting transcripts and notes | "What did we cover in my last meeting?" |
| n8n (your own) | Step 5 | Natural language — trigger and inspect your own n8n workflows | "Run my lead-qualification workflow on this email" |
| Google Calendar | Step 5 | Natural language — direct Google Calendar access (secondary — use if Morgen not installed) | "What's on my Google calendar this week?" |
| Morgen (recommended) | Step 5 | Natural language — unified calendar + tasks across Google/Outlook/iCloud/native | "What's on my calendar this week?" · "Add a task called 'Review contracts' due Friday" |
| Motion Calendar | Step 5 | Natural language — Motion-specific features (teammate visibility, full event search) | "Who on my team has a conflict at 3pm?" |
| Playwright | Step 5 | Natural language — browser automation for web apps with no API (real Chromium, accessibility-tree snapshots — not a session hijack) | "Open Higgsfield and generate a video with prompt X" |
| SwiftKit | Step 5 | Natural language — hosted toolkit for iOS / macOS / Swift dev (100+ tools). Default for anything Apple-platform | "Build me a SwiftUI login screen" |
| Superhuman | Step 5 | Natural language — email triage + drafting via Superhuman's official remote MCP | "Triage my inbox" · "Draft a reply to the latest thread from Alex" |
| Google Drive | Step 5 | Natural language — browse, search, and read Docs / Sheets / PDFs / shared folders via Google's official hosted MCP | "Find the latest Q3 planning doc in my Drive" |
| Vercel | Step 5 | Natural language — deployments, build logs, runtime logs, domains, env vars via Vercel's official remote MCP | "List my recent deployments" · "Show build logs for the last failed deploy" |
| Telegram | Step 6 | Automatic when launched with ctg or cbraintg — reads and replies to Telegram messages |
(messages arrive automatically from connected chats) |
| GitHub | Step 7 | Natural language — repos, issues, PRs, code search, branches, commits | "List open PRs on cli-maxxing" · "Search my repos for any file that uses MORGEN_API_KEY" |
| Obsidian | 2ndBrain-mogging | Natural language — read/write/search a local Obsidian vault (set up via fidgetcoding/2ndBrain-mogging) | "Search my vault for notes about machine learning" |
| Canva | Add-on | Natural language — create or edit designs, social posts, presentations | "Design a social media post for our launch" |
Key distinction: Slash commands (
/fswarm,/fswarm1–/fswarmmax,/fmini,/fmini1–/fminimax,/fhive,/w4w,/concise,/bullets,/safetycheck,/gitfix,/recon,/osmani-build, plus the 2ndBrain-mogging/save,/wiki,/challenge,/emerge,/backfill,/aliases,/autoresearch,/canvas,/tether,/connect) require you to type the command. Everything in this table works by just talking to Claude naturally.Add-on tools (Canva) are not part of the step-by-step setup — they're optional MCP servers you can connect separately. Claude auto-detects them when they're installed. Figma, Excalidraw, and Gamma live in creativity-maxxing.
Installed by creativity-maxxing via npx skills add Leonxlnx/taste-skill. The pack ships 8 variants as separate skill folders under ~/.claude/skills/. They're passive skills — Claude activates the right one based on what you ask for. You can also name a variant explicitly in your prompt ("use minimalist-ui", "redesign this with high-end-visual-design") to force a specific style.
| Variant | When Claude uses it | Example prompt |
|---|---|---|
/design-taste-frontend |
Default premium frontend rules — layout, typography, color, spacing, motion (3 tunable knobs) | "Build a premium SaaS landing page" |
/high-end-visual-design |
Expensive, soft UI look — premium fonts, whitespace, depth, spring animations | "Make this feel expensive, soft, Apple-ish" |
/full-output-enforcement |
Anti-laziness — stops placeholder comments, skipped blocks, half-finished outputs | "Give me the full component, no TODOs" |
/redesign-existing-projects |
Upgrading an existing project — audits current design, then fixes problems | "Redesign this dashboard — it looks generic" |
/stitch-design-taste |
Google Stitch-compatible semantic design rules; exports a DESIGN.md |
"Set up stitch-compatible design rules for this project" |
/minimalist-ui |
Clean editorial style — monochrome, crisp borders, Notion/Linear inspired | "Build a minimalist admin panel" |
/industrial-brutalist-ui (beta) |
Raw mechanical interfaces — Swiss typographic print fused with CRT terminal aesthetics | "Make it brutalist, terminal-style" |
/gpt-taste |
GPT-leaning variant of the taste filter — useful when output needs to feel more GPT-aligned | "Build this with gpt-taste style" |
The main design-taste-frontend file has three numeric knobs at the top you can tune from 1-10. Edit ~/.claude/skills/design-taste-frontend/SKILL.md and change the numbers, or tell Claude "set DESIGN_VARIANCE to 9 for this build".
| Knob | 1-3 | 4-7 | 8-10 |
|---|---|---|---|
| DESIGN_VARIANCE | Clean, centered | Balanced | Asymmetric, modern |
| MOTION_INTENSITY | Simple hover | Smooth transitions | Magnetic, scroll-triggered |
| VISUAL_DENSITY | Spacious, luxury | Balanced | Dense dashboards |
No slash commands. Taste Skill activates automatically on design work, or when you reference a variant by name. There are no
/taste-skill-style commands to memorize. If you want Claude to NOT use it, say so — e.g. "skip taste-skill for this one, just give me raw HTML."
When these tools are active, you may see indicators in your Claude session:
| Indicator | Meaning |
|---|---|
| ⚡️ fidgetflo | FidgetFlo MCP server is connected |
| 🧠 Brain² | Working inside your Obsidian vault |
| 🐝 Swarm | Swarm is active — shows agent count (after /fswarm) |
| 🍯 Mini | Mini swarm is active — shows agent count (after /fmini) |
| 👑 Hive | Hive-mind is active (after /fhive) |
The status line also shows your current model, session duration, context-window usage, and Claude rate-limit bars for the 5h and 7-day windows (color-graded green → yellow → red as you approach each limit). Usage comes straight from Claude Code's own rate_limits data — no API token or network call.
In normal mode, Claude will ask before doing things like editing a file or running a command. You'll see a prompt — type y and hit Enter to approve, or n to deny. In cskip mode or with Shift+Tab toggled on, this is skipped entirely.
These are available in your terminal after Step 4 installs the FidgetFlo CLI.
| Command | What it does |
|---|---|
npx fidgetflo@latest doctor --fix |
Diagnose and fix FidgetFlo installation issues |
npx fidgetflo@latest daemon start |
Start the FidgetFlo background daemon |
npx fidgetflo@latest swarm status |
Check the status of any running swarm |
npx fidgetflo@latest memory search --query "..." |
Search your agent memory for past context |
npx fidgetflo@latest memory list |
List all stored memory entries |
| Command | What it does |
|---|---|
bash <(curl -fsSL https://raw.githubusercontent.com/fidgetcoding/cli-maxxing/main/update.sh) |
Re-run all steps, skip what is installed, pick up anything new |
source ~/.zshrc |
Reload shell config to activate new aliases |
claude update |
Update Claude Code itself to the latest version |
brew update && brew upgrade |
Update Homebrew and all installed packages (macOS) |
| Problem | Fix |
|---|---|
cskip not recognized |
Run source ~/.zshrc to reload your shell config |
| Claude asks for login | Run claude once normally and complete the browser login flow |
| Shift+Tab does nothing | Make sure you are inside an active Claude session (not at a normal shell prompt) |
| Swarm not responding | Run npx fidgetflo@latest doctor --fix to diagnose |
| MCP tools not connecting | Exit Claude, run claude mcp list to check connections, then relaunch |
cbrain not working |
Run cskip instead, then tell Claude: "cbrain isn't working — can you figure out why and fix it?" Claude will find the problem, fix it, and get it working for future sessions. |
| Obsidian vault not found | Vault setup lives in 2ndBrain-mogging. Once set up, tell Claude the full path to your vault (e.g., ~/MyVault). Avoid ~/Desktop/... — macOS TCC protection breaks CLI access to the Desktop. |
| Shift+Return acts like Enter | Try Option+Enter as an alternative for multi-line input |
If you ran the installer, opened a fresh terminal, and claude --version still says "command not found" — your ~/.zshrc is probably missing the lines that wire Homebrew, nvm, ~/.local/bin, and the cli-maxxing aliases together. Paste the block below verbatim into your terminal — it appends the four things that need to be on PATH (creating ~/.zshrc if it doesn't exist), then sources it and verifies claude works. Your existing ~/.zshrc content is untouched.
cat >> ~/.zshrc <<'EOF'
# --- cli-maxxing reset-path block ---
# Homebrew (Apple Silicon default; also works on Intel via brew --prefix)
if [ -x /opt/homebrew/bin/brew ]; then
eval "$(/opt/homebrew/bin/brew shellenv)"
elif [ -x /usr/local/bin/brew ]; then
eval "$(/usr/local/bin/brew shellenv)"
fi
# nvm (Node Version Manager)
export NVM_DIR="$HOME/.nvm"
[ -s "$NVM_DIR/nvm.sh" ] && . "$NVM_DIR/nvm.sh"
[ -s "$NVM_DIR/bash_completion" ] && . "$NVM_DIR/bash_completion"
# User-local binaries (cbrain, ctg, cbraintg land here)
export PATH="$HOME/.local/bin:$PATH"
# cli-maxxing aliases
alias cskip='claude --dangerously-skip-permissions'
alias cc='claude'
alias ccr='claude --resume'
alias ccc='claude --continue'
EOF
source ~/.zshrc
claude --versionWhat to expect: claude --version should print something like 2.1.112 (Claude Code). If it does, you're unstuck — close this terminal, open a fresh one, and continue from where you left off. If it still errors, the issue is upstream of PATH (Claude Code itself didn't install, or Node is broken) — paste the error into a cskip session and Claude can diagnose.
Using bash instead of zsh? (Your terminal saying "The default interactive shell is now zsh" means yes.) Swap
~/.zshrcfor~/.bashrcin the block above, then run this once — macOS opens bash as a login shell, which skips~/.bashrcunless~/.bash_profileloads it:printf '\n[ -f "$HOME/.bashrc" ] && . "$HOME/.bashrc"\n' >> ~/.bash_profileOr skip all of that and switch your account to zsh:
chsh -s /bin/zsh, then fully quit and reopen the terminal.
- Claude remembers context within a session. If it's getting confused, use
/compactto reset. - You can paste file paths, URLs, and error messages directly — Claude will read and understand them.
- Say "yes" when Claude asks for permission, or use the
!prefix to run commands directly without leaving the session. - Voice commands work in Claude Code — use
/voiceto enable microphone input. - If Claude seems confused or slow, run
/compactto summarize the conversation and free up context. - You can drag and drop files into your terminal to paste their full file path.
- Press Up Arrow to recall and re-run previous commands without retyping them.
- To start fresh, just exit and relaunch.