🎯 The Mental Model — In 3 Sentences
- Slash commands = personal shortcuts you type with
/. Recipes Claude follows. - Skills = capabilities Claude reaches for automatically when your request matches their description.
- MCPs = external services Claude can talk to (Drive, GitHub, Vercel, Supabase, etc.).
- Specialists (agents) = short-lived Claude sessions you call in for one specific job.
⚡ Slash Commands — Type / in Claude Code
Your daily drivers
| Type | What it does | When to use |
|---|---|---|
/scaffold app | Create a new project — PRD, spec, CLAUDE.md, SECURITY, package.json, .gitignore, .env.example all pre-filled | Starting any project. 8 templates: app, api, lib, site, nextjs-supabase, cli, mcp, docs |
/kickoff | Read PRD/spec, seed PROGRESS.md, propose next 3 actions | After /scaffold, OR starting a new feature |
/standup | Yesterday/Today/Blockers from git commits | First thing each morning. /standup · /standup yesterday · /standup week |
/review | Pre-commit code review — bugs, security, regressions, missing tests | Right before any /commit |
/commit | Secret scan first, then drafts a commit message for approval | Always use this instead of raw git commit — saves you from leaking API keys |
/deploy | Pre-deploy checks → confirms target → ships | Going live. Refuses dirty trees + failing tests |
/fix | Diagnose recent error → propose minimal fix | Something broke and you're not sure why |
/help | List available skills + recommend the right one | When stuck or unsure what to use |
/lead-magnet | Build complete lead magnet bundle: deliverable + opt-in + emails + GHL workflow + socials | New lead magnet — see /lead-magnet pdf or /lead-magnet vault |
TwisterHub-specific
| Type | What it does |
|---|---|
/new-client | Scaffold a client folder with locked subfolder tree (PPT, GHL Funnel pages, GPTs, Sessions, Emails, Automations, Assets, Resources, Deliverables, Working, Recordings) |
/heygen-skills | Create HeyGen avatar videos via the v3 Video Agent pipeline |
Skill suites — type / to see them grouped
brand-voice:
enforce-voicebrand-voice-enforcementdiscover-brandgenerate-guidelines
marketing:
email-sequencecampaign-plandraft-contentbrand-reviewseo-auditcompetitive-briefperformance-reportcontent-creation
product-management:
brainstormwrite-specroadmap-updatemetrics-reviewsynthesize-researchstakeholder-updatesprint-planning
engineering:
architecturecode-reviewdebugdeploy-checklistdocumentationincident-responsesystem-designtesting-strategytech-debt
productivity:
task-managementstartupdatememory-management
anthropic-skills:
pdfdocxpptxxlsxcanvas-designskill-creatorschedule
Built-in CLI commands
| Type | What it does |
|---|---|
/init | Initialize a CLAUDE.md from existing codebase |
/mcp | MCP panel — manage authentication for connected services |
/plugin | Browse + install plugins from the marketplace |
/clear | Clear conversation context |
🤖 Built-In Specialists (Agents) — Call by Name
Ship with Claude Code. Call them when a task is too big, too specialized, or would clutter your main conversation.
| Specialist | What they do | Best when |
|---|---|---|
Explore | Searches your project, reports back without polluting context | "Find every file that references X." Read-heavy lookups. |
Plan | Designs the step-by-step approach | Before any non-trivial feature. Architecture decisions. |
code-reviewer | Audits changes for bugs, security, quality | Right before committing anything touching auth, payments, secrets |
debugger | Traces errors, finds root cause, suggests fix | Test fails and you don't know why. New error. |
frontend-developer | Builds React/Vue components end-to-end | Multi-file UI work |
documentation-expert | Writes clear technical docs | After a feature ships, when README is stale |
general-purpose | Utility player — when nothing else fits | Default when in doubt |
How to call one in:
"Bring in the code-reviewer specialist and have them audit the changes I just made."
"Ask the Explore specialist to find everywhere we handle user login."
🔌 MCPs — Services Claude Can Talk To
13 connected. Each plugs in one specific external capability.
Documentation + research
| MCP | What it gives Claude |
|---|---|
context7 | Real, current library docs — no more making things up |
sequential-thinking | Structured step-by-step reasoning on complex problems |
Browser + memory
| MCP | What it gives Claude |
|---|---|
playwright | Open websites, click, fill forms, take screenshots |
memory | Remembers things across sessions — survives Claude Code restarts |
Cloud services (yours)
| MCP | What it gives Claude |
|---|---|
supabase (PAT) | Run SQL, manage tables, fetch keys — fallback method |
supabase-oauth | Same but via OAuth — preferred going forward |
plugin:vercel:vercel | Deploy, sync env vars, inspect deployments — also gives /vercel:* slash commands |
Productivity (claude.ai connectors)
| MCP | What it gives Claude |
|---|---|
| Google Drive | Read/search Drive files |
| Gmail | Read/search emails, draft replies |
| Google Calendar | List/create/update events |
| Airtable | Read/write Airtable bases |
| Make | Trigger Make.com scenarios |
| Canva | Read designs, generate from templates |
You can also call MCPs explicitly when accuracy matters:
"Use Context7 to look up the current Next.js routing API."
"Use the Memory MCP to remember that my main project uses Supabase."
🎬 Common Workflows — What to Type When
Starting a new project
mkdir my-project && cd my-project
claude
> /scaffold app
> /kickoff
Starting a new feature on an existing project
> /kickoff <feature description>
Daily standup
> /standup
Before committing
> /review
> /commit
Deploying
> /deploy
Stuck or broken
> /fix
> /help
Writing a marketing email
> Use marketing:email-sequence to draft a 5-email
welcome series for my AI Jumpstart Challenge audience
Brainstorming an idea
> Use product-management:brainstorm to think through
whether I should add a free 3-day challenge to my funnel
Building a lead magnet (TwisterHub)
> /lead-magnet pdf "Top 10 GHL Power Moves" --audience=ghl-users
> /lead-magnet vault "AI Prompts for Realtors" --audience=realtors
Setting up a new client folder
> /new-client Acme Roofing
Drafting on-brand copy
> Write a Facebook post about [topic] in our brand voice
(brand-voice:enforce-voice triggers automatically)
Recording a HeyGen video
> Make a video of me announcing the next Quponing campaign
(heygen-skills triggers automatically)
🛡️ Safety Hooks Running Automatically
| Hook | What it does |
|---|---|
| .env guardian | Blocks Claude from reading .env files — prevents API key leaks |
| Agent teams | CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 enabled — lets Claude spawn parallel teammates |
🚦 The 4-Color Framework — When to Step In
When Claude is working autonomously (long task, you've stepped away):
🟢 Green
What's happening: Claude debugging itself, repetitive execution working
Do this: Leave alone
🟡 Yellow
What's happening: 20+ min since you peeked at PROGRESS.md
Do this: Just peek, don't comment
🟠 Orange
What's happening: Stuck on same test 3x, scope creep, design choice you don't love
Do this: Redirect — "Try X instead."
🔴 Red
What's happening: Deploying to prod, DB migrations, deleting files, gut says off
Do this: Hard-stop — "Pause. Show me what you just did."
Rule: Claude's job is execution. Your job is judgment. Judgment wins.
📁 Where Things Live On Your Computer
~/.claude/ ← your global Claude Code home
├── settings.json ← env vars, hooks, permissions
├── CLAUDE.md ← your global handbook
├── commands/ ← your slash commands
│ ├── scaffold/SKILL.md
│ ├── kickoff/SKILL.md
│ ├── standup/SKILL.md
│ ├── review/SKILL.md
│ ├── commit/SKILL.md
│ ├── deploy/SKILL.md
│ ├── fix/SKILL.md
│ ├── help/SKILL.md
│ └── lead-magnet/SKILL.md
├── agents/ ← custom specialists you build
└── plugins/ ← marketplace plugins (Vercel, etc.)
Twister Hub AI/AI Training/Claude Code Workshop/
├── VCI-WORKSHOP-COMPLETION-NOTES.md ← deep reference for every SOP
├── CLAUDE-CODE-CHEAT-SHEET.md ← markdown of THIS page
├── BUILD_DAY_RESULTS.md ← live URLs of shipped apps
├── SOP-Markdown/ ← all SOPs as markdown for searching
└── Build Day Apps/ ← lead-magnet, roi-calculator, vault
🆘 When Claude Gets Stuck — Universal Fix
Describe what you saw, ask Claude to fix it. You don't need to know why it broke.
Examples that always work:
- "The /scaffold command is asking permission on every file. Can you add the safe tools to the pre-approved list?"
- "My .env hook is blocking .env.example. Can you fix the pattern?"
- "You committed code that fails tests. The pre-commit check should have stopped that. Can you check the hook?"
📚 Deeper Reference (on disk)
- VCI-WORKSHOP-COMPLETION-NOTES.md — full per-SOP summaries
- SOP-Markdown/G05-Prompts.md — official VCI prompt library
- SOP-Markdown/G06-Stuck.md — troubleshooting guide
- SOP-Markdown/G07-Glossary.md — every term defined
- Original SOP PDFs — in your Drive folder, permanent
YOU ARE READY
9 slash commands · 50+ skills · 13 MCPs · 7 specialists · brand voice baked in
This is more capability than 99% of Claude Code users have. Use it like a kitchen — pick the right tool for the right job, and lean on /help whenever you forget what's available.
With Your Success in Mind,
— Skeeter's Claude Code Setup