Mode Switcher

Mode Switcher

One command to reconfigure Claude Code for a specific workflow. Loads skills, sets behavioral rules, adjusts permissions.

What Modes Do

A mode is a preset configuration that:

1. Loads specific skills — CCC domains and individual skills relevant to the workflow
2. Sets behavioral instructions — how Claude should think, respond, and prioritize
3. Configures permissions — acceptEdits (confirm changes) vs autoAccept (fully autonomous)
4. Emphasizes hooks — which automated checks matter most for this workflow
5. Sets context strategy — when to compact, what to prioritize in context window

Modes are not exclusive — you can switch at any time. Switching modes replaces the previous mode's instructions.

Usage

/cc mode <name>        # Switch to a mode
/cc mode               # Show current mode + list all
/cc mode normal        # Reset to default

Or reference naturally in conversation: "switch to design mode", "go yolo", "enter night mode".

Available Modes

Mode Details

normal (default)

The baseline. No special skills loaded, standard confirmation flow. Use this when switching away from a specialized mode or when doing general-purpose work that doesn't fit a category.

When to use: General coding, maintenance, debugging, mixed tasks.

design

Activates the full ccc-design ecosystem: animations, visual effects, Impeccable polish suite, landing page builder, component design. Claude prioritizes visual output, suggests screenshots/previews, and thinks component-first.

When to use: UI work, landing pages, design systems, animations, visual polish.

saas

Loads ccc-saas (auth, billing, DB, API, frontend, growth) plus ccc-devops (CI/CD, infrastructure, monitoring). Claude thinks in production terms — error handling, scaling, security, billing edge cases.

When to use: Building or extending SaaS products. Full-stack feature work.

marketing

Combines ccc-marketing (content, email, ads, social) with ccc-seo (technical SEO, content SEO, analytics). Claude writes like a marketer — compelling copy, conversion-focused, data-driven recommendations.

When to use: Content creation, SEO optimization, email campaigns, landing page copy, CRO.

research

Activates ccc-research for deep analysis workflows. Claude uses extended thinking, spawns parallel research agents, cites sources, and produces structured findings.

When to use: Technical research, competitive analysis, architecture evaluation, deep dives.

writing

No special skills — just behavioral tuning. Claude minimizes technical jargon, focuses on prose quality, grammar, flow, and structure. Ideal for long-form content, documentation, blog posts.

When to use: Blog posts, documentation, copywriting, any prose-heavy work.

night (autonomous overnight)

The overnight workhorse. All relevant CCC domains loaded, full auto-accept permissions. Claude auto-checkpoints every 5 edits, self-verifies before marking tasks done, saves session before compacting, and writes detailed execution logs.

When to use: Large batch tasks you want done by morning. Set task list, enter night mode, walk away.

yolo (speed mode)

Full auto-accept with whatever skills you choose. Claude skips explanations, minimizes confirmation overhead, and just builds. Hooks remain active as the safety net.

When to use: When you know exactly what you want and speed matters more than discussion.

unhinged (max creativity + autonomy)

Everything YOLO offers plus maximum creative freedom. All CCC domains loaded. Claude makes bold architectural choices, aggressively refactors, and pushes boundaries — but backs it up with comprehensive testing.

When to use: Greenfield projects, major refactors, creative exploration where you trust Claude to make big moves.

caveman (token-saving communication)

Ultra-compressed output mode. Strips articles, filler, pleasantries, and hedging from all prose responses. Technical content — code blocks, error messages, exact terms — is untouched. Three intensity levels: lite (no fluff, grammar intact), full (fragments, no articles), ultra (telegraphic, abbreviations, arrow notation). Stacks with any other mode.

When to use: Long debugging sessions, architecture discussions, any work where explanation volume is high and you want to cut output costs by ~65-75%. Deactivate with /caveman off.

adhd (answer-first output)

Reorders every response so the fix leads and the reasoning follows: answer/command/file:line first, context last. This is a priority reorder, not a compression mode — full accuracy stays, only the order changes. Based on ayghri/i-have-adhd (MIT). Stacks with caveman: caveman governs length, adhd governs order — together the fix comes first, in the fewest words. Full contract: commander/cowork-plugin/skills/ccc-adhd/SKILL.md. Toggle with /ccc-adhd on|off|status, or say "stop adhd mode".

When to use: You want the fix first, every time, without asking — beginners especially benefit since the answer is always the first line. Debugging loops where "what do I run right now" matters more than narrative.

Switching Protocol

When a mode is activated:

1. Acknowledge — Confirm the mode switch with a brief status line
2. Load — Reference the mode file from modes/{name}.md
3. Pre-flight — Run the mode's pre-flight checklist
4. Context check — Verify context window has room for the mode's skills
5. Activate — Apply behavioral instructions for the remainder of the session

When switching between modes, the previous mode's instructions are fully replaced (not layered).

Mode Files

Each mode is defined in modes/{name}.md with:

• Overview of what the mode optimizes for
• Skills loaded (CCC domains + individual skills to suggest)
• Behavioral instructions (how Claude should operate)
• Hook emphasis (which hooks matter most)
• Context strategy (pre-flight context check, compaction thresholds)
• Pre-flight checklist (what to verify before entering)