#Two Ways to Use SkillDB: Cloud Brain vs. Local Arsenal

1:47 AM. My terminal has 47 tabs open.

I just finished a 12-hour session building a content creation studio from scratch. During that marathon, I pulled knowledge from 14 different SkillDB packs — web-polish-skills, marketing-skills, social-media-skills, design-systems-skills, copywriting-skills, data-visualization-skills — the works. I used that knowledge to build 4 tiers of features, from character counters to A/B testing systems to Dream 100 trackers.

And then someone asked me to purge the loaded skills.

$ skilldb purge
No .skilldb/ directory found. Run "skilldb init" first.

Wait. What?

I'd been using SkillDB for twelve straight hours and there was nothing on disk. No .skilldb/ directory. No cached files. No local installation at all. Every skill I read, every framework I applied, every anti-pattern I avoided — it all came through the API, lived in the cloud, and never touched my filesystem.

And that's when it hit me: there are two fundamentally different ways to use SkillDB, and most people don't realize they're choosing between them.

#Mode 1: Cloud Brain (API-Only)

This is what I did during the marathon session. The MCP server connects to SkillDB's API, searches the library, and streams full skill content directly into the conversation. The skills live in the cloud. Your agent reads them, applies the knowledge, and moves on. Nothing is cached. Nothing is installed. Nothing to clean up.

# This is all that happened — search, read, apply
mcp__skilldb__skilldb_search → "web polish" → 12 results mcp__skilldb__skilldb_get → "web-polish-skills/ui-audit.md" → 205 lines of expert knowledge
#Agent reads it, applies it, builds features. Done.

Think of it like Googling something vs. downloading a PDF. You get the information, you use it, and you don't accumulate files on your hard drive.

When to use Cloud Brain mode:

• Exploring skills you haven't used before
• One-off analysis or research sessions
• Cross-referencing multiple packs for strategic decisions
• When you need knowledge from 10+ packs but don't want to install them all
• When disk space matters (CI/CD pipelines, ephemeral environments)

What you need: An API key (free at skilldb.dev/api-access) and the MCP server configured.

The trade-off: Skills aren't persisted between sessions. Next time you start a conversation, the agent doesn't "remember" the web-polish-skills audit checklist. It has to fetch it again. For exploration, this is fine. For daily work on a specific project, it's wasteful.

#Mode 2: Local Arsenal (Installed)

This is the mode SkillDB was originally built for. You install skill packs into your project, they live in a .skilldb/ directory alongside your code, and your agent can read them anytime without an API call.

# Initialize SkillDB in your project
skilldb init
#Install the packs you need
skilldb add web-polish-skills skilldb add marketing-skills skilldb add social-media-skills
#Now they're on disk:
$ ls .skilldb/skills/web-polish-skills/ ui-audit.md color-system-repair.md component-unification.md form-elements.md navigation-patterns.md responsive-repair.md typography-cleanup.md modal-dialog-system.md design-token-extraction.md dark-mode-retrofit.md

The skills are now part of your project. Every Claude Code session that opens this repo has instant access to them — no API calls, no latency, no key required.

When to use Local Arsenal mode:

• You work on the same project daily and need consistent skill access
• You want skills available offline
• You're embedding skill knowledge into CLAUDE.md or agent configuration
• You want your team to share the same skill set via git
• You need deterministic behavior (no API dependency)

Managing disk space:

# See what's installed
skilldb list
#Remove inactive skills (keeps ones you're using)
skilldb purge --inactive
#Remove everything
skilldb purge --all
#Preview what would be removed
skilldb purge --dry-run

#The Decision Matrix

Scenario	Mode	Why
"I need to audit my UI real quick"	Cloud Brain	One-off, no need to persist
"I'm building a SaaS and need security skills daily"	Local Arsenal	Persistent, offline, team-shared
"Which marketing skills exist?"	Cloud Brain	Exploration, browsing
"My agent should always know React best practices"	Local Arsenal	Embed in project config
"I want to cross-reference 15 packs for a strategy doc"	Cloud Brain	Breadth over depth
"This is my agent's core personality"	Local Arsenal	Identity, not reference

#The Hybrid Approach (What Pros Do)

The real power move is using both. Install your core skills locally — the ones your agent needs every single day. Then use the API for exploration, research, and pulling in specialist knowledge when you hit an unfamiliar problem.

# Your project's core skills — always available
skilldb add software-skills skilldb add web-polish-skills
#When you hit a security audit and need specialist knowledge:
#Just search via API — no need to install
mcp__skilldb__skilldb_search → "penetration testing" mcp__skilldb__skilldb_get → "pentest-methodology-skills/web-app-pentest.md"
#Read it, apply it, move on. Don't install it.

Your .skilldb/ directory stays lean (just the essentials), while your agent still has access to all 5,000+ skills when it needs them.

#What I Learned in 12 Hours

During my marathon studio build, I pulled from these packs via API:

• web-polish-skills — UI audit, typography, forms, navigation, color systems
• design-systems-skills — Motion guidelines, component patterns
• marketing-skills — Social media strategy
• social-media-skills — Short-form video, paid social, engagement growth
• social-media-business-skills — Engagement routines, Dream 100
• copywriting-skills — Brand voice
• data-visualization-skills — Recharts patterns
• streaming-content-skills — Content repurposing
• autonomous-agent-skills — Accessibility, responsive design

I read over 3,000 lines of skill content. I applied patterns from every single one. And my disk usage? Zero bytes.

If I were going to keep working on this studio every day, I'd install the top 3-4 packs locally and keep the rest as API lookups. But for a single session of exploration and building? Cloud Brain was perfect.

#The Bottom Line

Cloud Brain = research mode. Broad, ephemeral, exploratory. Your agent is a consultant who reads the manual, applies the knowledge, and leaves.

Local Arsenal = work mode. Deep, persistent, reliable. Your agent is a specialist who has internalized the knowledge and carries it into every session.

Pick based on how often you'll need the skills, not how many you need right now.

Explore the full library at skilldb.dev/skills. Get your API key at skilldb.dev/api-access. And if you want to go deep, skilldb init and start building your local arsenal.

The skills are there. The question is whether you want a library card or a personal bookshelf.