← irvcassio.com
Workflow Edition

The AI Enthusiast

Overview on skills, automation, MCP servers, OpenClaw, and the Cincinnati AI community
Irv Cassio • February 2026
02

The 100 Bots Dream

As a person doing solo support for digital, drowning in repetitive tasks. The idea: what if I could create 100 bots to handle all my tasks — and free myself to focus on growth?

The dream was simple: Stop being caught in the operational details. Stop fighting fires. Build something that scales beyond one person's bandwidth — so I could finally focus on strategy, innovation, and the bigger picture.
You don't need 100 bots. You need one agent with the right skills.

Claude Code is an example of this 100 Bots dream coming to life. Skills handle the automation. Multi-session gives you parallelism. Sub-agents give you a team. And it all runs from your terminal.

03

Theme: Autonomous

Claude Code operates as a fully autonomous coding agent — reading files, running commands, managing git, and making decisions. Your CLAUDE.md file defines the rules of engagement.

📄
CLAUDE.md
Project context, conventions, and behavioral rules. Claude reads this at the start of every session. Hierarchy: global → project → subdirectory.
Plan vs. Act Mode
Shift+Tab toggles Plan Mode — Claude thinks before acting. Act Mode is full autonomy: read, write, execute.
Hooks
Event-driven automation: PreToolUse, PostToolUse, Notification, Stop. Rules that fire automatically — lint after edits, notify on push.
🔒
Permissions
Fine-grained allow/deny rules per tool. Our setup: full autonomy for file ops, git, builds. Only confirm before force-pushing main or deleting repos.

Deep Dive
What is Claude Code + CLAUDE.md
Terminal-native agent details, capabilities, and the CLAUDE.md hierarchy
What is Claude Code?

Terminal-Native AI Coding Agent

Claude Code is Anthropic's terminal-based AI coding agent. It reads your files, runs commands, edits code, manages git, and orchestrates multi-step workflows — all from your terminal. Think of it as a senior developer with full repo access that you pair-program with.

Terminal-Native
Runs in your terminal — no browser, no IDE plugin required. Works with any editor, any repo.
🛠
Full Agent
Reads, writes, edits files. Runs bash commands. Manages git. Creates PRs. Deploys code.
🤖
Sub-Agents
Spawns parallel agents for research, code review, testing — like a team of specialists.
🔎
Context-Aware
Understands your entire codebase, project structure, and conventions via CLAUDE.md files.

CLAUDE.md Deep Dive

CLAUDE.md — The Foundation

CLAUDE.md is the single most important file in your Claude Code setup. It's project instructions that Claude reads at the start of every session — your coding standards, project context, and behavioral rules.

Hierarchy
Location Scope Use For
~/CLAUDE.md Global (all projects) Personal preferences, behavioral rules, autonomy settings
./CLAUDE.md Project root Stack info, conventions, build commands, architecture decisions
./src/CLAUDE.md Subdirectory Module-specific instructions (e.g., API conventions, test patterns)
Demo tip: Our ~/CLAUDE.md sets Claude to "Fully Autonomous Agent" mode — no confirmation needed for file ops, git, builds, or package managers. Only confirms before force-pushing to main or deleting repos.

04

Theme: Multi-Session

Run multiple Claude Code instances simultaneously in separate terminal tabs. Different projects, different models, different roles — true parallelism on one machine.

Patterns
PatternExample
Different projectsTab 1: Canopy (SSH client) • Tab 2: Stoke (CRM) • Tab 3: ELSecurityAlert (iOS)
Different modelsclaude-irv-opus for architecture • claude-irv-haiku for quick lookups
Different rolesTab 1: Writing code • Tab 2: Email triage via MCP • Tab 3: Data imports
~/.claude
Default profile
~/.claude-irv
Personal
~/.claude-el
Work
Multi-profile setup: Shell aliases set CLAUDE_CONFIG_DIR per profile. Skills and plugins are shared (symlinked). Session state is isolated.
💡
Boris Cherny's Workflow CREATOR
Boris Cherny, creator of Claude Code at Anthropic, runs 5 local + 5–10 web instances simultaneously. His power-user patterns:
PracticeDetail
Tabs numbered 1–5Uses system notifications to know when Claude needs input
Always Opus + thinkingExtended thinking enabled for every session
Plan mode firstBegins most sessions in Plan mode before switching to Act
CLAUDE.md updated weeklyUpdated multiple times per week — compound institutional knowledge
Custom slash commandsFrequent workflows as commands (e.g., /commit-push-pr)
Read the full article on Push to Prod →

Deep Dive
🖥
Multi-Session & Multi-Profile Details
Profiles, aliases, symlinks, shared vs. separate config, screenshots
Multi-Session Workflow

Parallel Claude Code Sessions

Run multiple Claude Code sessions simultaneously using Warp (or custom-developed Canopy app) — Mac-native terminals with multi-tab support. Each tab runs an independent Claude Code instance with different profiles, models, or project contexts.

Multiple Claude Code sessions running in parallel in Canopy
Why Multiple Sessions?
PatternExample
Different projects Tab 1: cd ~/Canopy && claude-irv
Tab 2: cd ~/Stoke && claude-irv
Tab 3: cd ~/ELSecurityAlert && claude-el
Different models Tab 1: claude-irv-opus (complex architecture)
Tab 2: claude-irv-haiku (quick lookups)
Different roles Tab 1: Writing code (act mode)
Tab 2: Email triage via MCP (research)
Tab 3: Data imports (/import_all)
Parallel work Session A builds a feature while Session B runs imports — true multitasking with independent contexts

Multi-Profile Setup

Separate Identities, Shared Tools

Run separate Claude Code identities for different contexts — personal projects, work projects, different models — all from the same machine. Shared skills and plugins, isolated session state.

~/.claude
Default profile
Source of truth
~/.claude-irv
Personal profile
Symlinks → .claude
~/.claude-el
Work profile
Symlinks → .claude
How It Works
Shell aliases in ~/.zshrc set CLAUDE_CONFIG_DIR per profile:
# ~/.zshrc aliases
alias claude="claude"                          # default
alias claude-irv="CLAUDE_CONFIG_DIR=~/.claude-irv claude"   # personal
alias claude-el="CLAUDE_CONFIG_DIR=~/.claude-el claude"     # work

# Model variants (same config dir, different model)
alias claude-irv-sonnet="CLAUDE_CONFIG_DIR=~/.claude-irv claude --model sonnet"
alias claude-el-opus="CLAUDE_CONFIG_DIR=~/.claude-el claude --model opus"
alias claude-irv-haiku="CLAUDE_CONFIG_DIR=~/.claude-irv claude --model haiku"
Shared vs. Separate
Shared (symlinked)Separate per profile
skills/, commands/, plugins/
settings.json, settings.local.json
.mcp.json
.claude.json (session state)
projects/ (memory, per-project)
todos/, cache/, telemetry/

05

Theme: Multi-Agent & Workflow

Claude Code spawns specialized sub-agents that work in parallel — like delegating to a team. The /workflows pipeline takes a feature from idea to shipped code, launching multiple agents at each stage.

Brainstorm
Explore & clarify
Plan
Design approach
Work
Ship the feature
Review
Multi-agent audit
Compound
Document learnings
🚀
/lfg
"Let's go" — full autonomous mode. Plan, implement, review, ship. One command kicks off the entire pipeline with parallel agents.
🛠
/workflows:review
Spawns 6+ specialized agents in parallel: security-sentinel, performance-oracle, architecture-strategist, code-simplicity-reviewer, and more.
How it works: Each /workflows stage spawns sub-agents — Explore, Plan, General Purpose, Background — that run concurrently. The compound-engineering plugin adds 29 specialized review/research/design agents on top. One command, an army of agents.

Deep Dive
🤖
Workflows & Sub-Agents Detail
The brainstorm-plan-work-review-compound pipeline, sub-agent types, compound-engineering stats
Engineering Lifecycle

The /workflows Pipeline

The /workflows pipeline — from idea to institutional knowledge. Each stage is a separate command that can be used independently or chained together.

StageCommandWhat Happens
Ideation/workflows:brainstormExplore requirements and approaches through collaborative dialogue before planning
Planning/workflows:planTransform feature descriptions into well-structured markdown plans
Execution/workflows:workExecute a work plan efficiently — ship complete features systematically
Review/workflows:reviewExhaustive multi-agent code review with ultra-thinking and worktrees
Documentation/workflows:compoundDocument solved problems while context is fresh
The pipeline: /workflows:brainstorm/workflows:plan/workflows:work/workflows:review/workflows:compound
Each stage feeds into the next. Brainstorm outputs inform the plan. The plan feeds the work command. Review validates the work. Compound captures the learnings.

Sub-Agents

Sub-Agents & Task Tool

Claude Code can spawn specialized sub-agents that run in parallel — like delegating to a team. Each agent has its own context and toolset. They return results to the main session.

Explore Agent
Fast codebase search — find files, grep code, answer questions about structure. Used for research without polluting main context.
Plan Agent
Software architect — designs implementation plans, identifies critical files, considers trade-offs. Read-only.
General Purpose
Full agent with all tools. Used for complex multi-step tasks that need autonomy.
Background Agents
Run tasks in background while you continue working. Get notified on completion.
Parallel execution: Multiple agents can run simultaneously — e.g., one researching docs while another runs tests while a third reviews code. Massive time savings on complex tasks.

compound-engineering

The Compound-Engineering Plugin

"AI-powered development tools that get smarter with every use." — From the Every Inc marketplace.

29
Agents
22
Commands
21
Skills
1
MCP Server
Review Agents 15 AGENTS
Specialized code review from every angle. /workflows:review runs these in parallel against your PR or branch.
AgentFocus
security-sentinelOWASP vulnerabilities, auth flaws, hardcoded secrets
performance-oracleN+1 queries, memory leaks, caching, scalability
architecture-strategistDesign decisions, component boundaries, patterns
code-simplicity-reviewerYAGNI, unnecessary complexity, over-engineering
pattern-recognition-specialistAnti-patterns, naming, duplication
data-integrity-guardianMigrations, transactions, referential integrity
Research Agents 5 AGENTS
AgentFocus
best-practices-researcherExternal docs, community standards, industry patterns
framework-docs-researcherOfficial framework docs, version-specific constraints
git-history-analyzerCode evolution, contributor patterns, archaeological analysis
learnings-researcherInstitutional knowledge from docs/solutions/
repo-research-analystRepo structure, conventions, existing patterns

06

Theme: Automation (Playwright)

Headless browser automation embedded in Claude Code workflows. Playwright drives authenticated enterprise portals, scrapes data, and automates web interactions — all triggered by slash commands.

Automated Data Imports
CommandWhat It Automates
/import_storemastersharepointSharePoint download with auto-auth MFA recovery PLAYWRIGHT
/import_vodofoneVodafone M2M portal — SIM inventory scraping PLAYWRIGHT
/import_metarawGoogle Sheets → MongoDB (4,500+ device records) API
/import_storegpsUnified GPS table from Knox, Meraki, OneLRA API
🌐
Stoke Agent
LinkedIn engagement automation — Playwright + Claude Haiku for AI-generated comments on targeted posts.
🤖
agent-browser Skill
Vercel's agent-browser CLI — browse websites, fill forms, take screenshots, scrape data from Claude Code.

07

Skills Overview

Skills are folders with a SKILL.md descriptor plus optional scripts and resources. Unlike commands (user-only), skills can auto-activate when their description matches the task context.

9
Custom Skills
Built by us
22
Plugin Skills
compound-engineering + others
Skill Anatomy
~/.claude/skills/my-skill/
  SKILL.md          # YAML frontmatter + instructions
  scripts/          # Optional helper scripts
  templates/        # Optional templates/resources
Key YAML fields: name, description, argument-hint, user-invocable, allowed-tools, context (fork vs. inline)

Deep Dive
Skills — Full Custom & Plugin Lists
All 10 custom skills and 22 plugin skills with triggers and descriptions
9
Custom Skills
Built by us
Click to expand
22
Plugin Skills
compound-engineering + others
Click to expand

08

Skill: /import_all

One command to refresh all enterprise data sources. Five daily imports orchestrated in the correct dependency order with a combined summary table.

Import Pipeline
StoreMaster
SharePoint
Vodafone
M2M Portal
Store GPS
Knox + Meraki
Meta Raw
Google Sheets
7-Day
Compute history
Pattern: Each import is a standalone Python script wrapped in a slash command. /import_all chains them in sequence with error handling and a formatted summary table. Playwright handles authenticated portals, Google API handles sheets.

09

Skill: /screenshot

Load recent screenshots into the conversation. Claude sees and analyzes images — multimodal context injection from your screenshot folder.

📷
How It Works
Auto-activates when you mention screenshots. Reads the N most recent images from ~/My Drive/Screenshots/ using Claude's multimodal vision. Great for UI review, debugging visual issues, and design feedback.
Example: "Show me my last 3 screenshots" → Claude loads images, analyzes layouts, spots CSS issues, compares against design specs — all in one turn.

10

Skills: /mongo + /storelookup

Enterprise data skills — fetch MongoDB schemas and run full store deep-dives, all from the terminal.

🗃
/mongo
READ-ONLY AUTO-DETECT

Fetch sample documents from any MongoDB collection. Auto-detects personal vs. corporate instance. Injects real schemas into conversation for informed code generation.
🔍
/storelookup
DEEP DIVE MULTI-SOURCE

Full store deep-dive by number — identity, device counts across all systems, GPS sources, and recent tickets. Queries MongoDB + OneLRA API.
Safety first: Both skills are read-only. Pull data, inspect, analyze — never write. Connection strings stored in macOS Keychain.

11

Remote Control

Continue a local Claude Code session from your phone, tablet, or any browser. Start a task at your desk, pick it up from the couch. BUILT-IN FEATURE

Start Methods
MethodCommandNotes
New sessionclaude remote-controlShows URL + QR code. Press Space to toggle QR.
Existing session/remote-control or /rcCarries over current conversation.
All sessions (auto)/config → EnableEvery session becomes remotely accessible.
How it works: Claude Code runs locally. Remote Control creates a secure outbound tunnel via the Anthropic API — no inbound ports. Your phone is just a window into the local session. Available on Pro and Max plans (Research Preview).

12

Skill: /calendar

Query your calendar from the terminal using macOS EventKit. Works with Exchange/Outlook calendars — bypasses AppleScript's Exchange limitations via inline Swift compilation.

📅
Natural Language Queries
"What do I have this afternoon?" • "Show me tomorrow's meetings" • "Am I free at 2pm?"

Supports: today, tomorrow, morning, afternoon, week, and natural-language time questions.
Technical trick: Compiles and runs a Swift snippet at call time via swift -e. EventKit correctly reads all calendar accounts including Microsoft 365 — something AppleScript can't do reliably.

13

Skill: /usageleft

Check your Claude usage versus daily and weekly allotment with visual bar graphs. Know exactly how much capacity you have left without leaving the terminal.

📊
Usage Dashboard
5-HOUR ROLLING 7-DAY OVERALL PER-MODEL EXTRA USAGE

Unicode block characters with color-coded utilization levels. Fetches from the Anthropic OAuth API using stored Keychain tokens. Works across all Claude Code profiles via $CLAUDE_CONFIG_DIR detection.

14

Skill: /present

Generate a self-contained presentation HTML from a natural-language description. Each slide is a standalone HTML5 document. It's how I prototype decks in seconds.

🎬
Key Features
OFFLINE-READY SINGLE FILE KEYBOARD NAV OVERVIEW GRID

Uses iframe.srcdoc (not blob URLs) for file:// compatibility. Supports slide types: title, bullets, code, SVG charts, base64 images. Press Escape for thumbnail grid overview. Writes the .html and opens it immediately.

15

Skills: /topemail + /last30days

Two research skills — one for priority email from key contacts, one for real-time topic research across the web.

/topemail CUSTOM
Priority emails from key contacts via Exchange. Uses AppleScript with dedicated per-contact mailboxes (Exchange whose clause workaround). Urgent emails surfaced from _Urgent mailbox rule.
🔍
/last30days PUBLIC
Publicly installable skill — research any topic across Reddit, X/Twitter, and web search. Uses OpenAI + xAI APIs for engagement-weighted synthesis. Graceful degradation to web-only if API keys absent.

16

MCP Servers

The Model Context Protocol is Anthropic's open standard for connecting AI to external tools and data. MCP servers give Claude direct, structured access to APIs and services.

Apple Mail MCP INSTALLED
Search, read, send, reply, forward emails. Manage folders and attachments. Works with Exchange accounts on macOS. Custom-built Python MCP server.
search_messages get_message send_email reply_to_message flag_message
📝
Notion MCP INSTALLED
Search, create, update pages and databases. Manage comments and teams directly from Claude Code.
📖
Context7 MCP PLUGIN
Live documentation lookup for any library. Auto-configured via compound-engineering plugin.
Ecosystem
GitHub Slack PostgreSQL Playwright Linear Filesystem Jira — and hundreds more

Deep Dive
🌳
Canopy — Mac-Native SSH + Git + AI Client
Custom-built macOS app: terminal, SFTP sync, Git client, BYOK AI chat — developed entirely with Claude Code
The philosophy: Review and optimize everything. Build your own SSH client to have the exact workflow you want, with AI built in from day one. This is the "rethink everything to 10X ourselves" approach in action — and proof that Claude Code can ship real, production macOS apps.

A custom-built macOS app that combines SSH terminal, SFTP file sync, Git client, and BYOK AI chat in one window. Built with SwiftUI, zero external SDKs for networking or AI — everything is pure URLSession + Citadel SSH. Developed entirely with Claude Code.

By the Numbers
111
Swift Files
15.5K
Lines of Code
5
AI Providers
Swift 6
macOS 15+
Multi-Tab Terminal
SSH + local terminal tabs with split panes (Cmd+D). xterm-256color via SwiftTerm. Dynamic resize, copy/paste, per-tab session isolation.
🔒
SSH & Auth
Citadel SSH v2 — password auth, Ed25519 keys, RSA keys. macOS Keychain for credentials. Server profiles stored as JSON.
📁
SFTP File Sync
Side-by-side local/remote file comparison. Push & pull with progress tracking. Filter by status (modified, local-only, remote-only). Checksum or size-based diff.
👁
Full Git Client
Branch management (create, checkout, delete, push, pull). Commit history timeline. Unified & side-by-side diff views with syntax highlighting.
🤖
AI Chat — BYOK Multi-Provider
Built-in AI assistant with streaming responses, file attachments (drag-and-drop), @-mention snippets, and per-server conversation history. All via URLSession REST + SSE parsing — no vendor SDKs.
ProviderDefault ModelNotes
Claude (Anthropic)claude-sonnet-4-5Multimodal, SSE streaming, x-api-key auth
OpenAIgpt-4oMultimodal, Bearer token, function calling
OpenRouteranthropic/claude-sonnet-4-5200+ models, OpenAI-compatible API
Ollama (local)llama3.2Fully offline, configurable base URL
LM Studio (local)User-selectedOpenAI-compatible, localhost:1234
Productivity Panels
Dashboard
Server metrics — uptime, CPU load, memory, disk, IPs. Works for both SSH and local machine.
Prompt Snippets
Saved prompt templates with categories and tags. @-mention to insert into AI chat.
Goals & Tasks
Lightweight project tracking — goals (active/completed) and tasks (pending/done) with inline add/edit/delete.
Knowledge Vault
Personal docs library with full-text search. "Send to Chat" injects content into AI conversation for context.
Tech Stack
SwiftUI macOS 15+ SPM Citadel (SSH/SFTP) SwiftTerm HighlightSwift MarkdownUI
@Observable + @MainActor architecture. Actor-based services for thread safety. Window opacity control (30-100%). Theme system (Default + Canvas). Zero vendor AI SDKs — all providers via URLSession REST + SSE. Bundle ID: com.irvcassio.Canopy

17

OpenClaw

The open-source AI agent that lives in your messaging apps. 200,000+ GitHub stars. A phenomenon that went from weekend project to global movement — and a signal of what most people have been wanting out of AI.

The Timeline
ClawdBot
Nov 2025
MoltBot
Jan 27
OpenClaw
Jan 30
OpenAI hires Peter
Feb 15
Created by Peter Steinberger (PSPDFKit). Anthropic trademark cease-and-desist triggered three names in three days. Then OpenAI hired Steinberger to lead their personal agents effort — OpenClaw moves to a foundation and stays open-source.
Why this matters: The idea of a fully autonomous AI working for you — managing your todo list, self-learning, having memory and a personality — is transformative. OpenClaw is proof of the demand. Apple has seen a surge in Mac mini sales driven by people buying them specifically to host personal AI agents. This isn't a developer trend — it's mainstream.
Security Considerations
Security researchers have identified risks in third-party skills — including data exfiltration and prompt injection. With great autonomy comes real responsibility. Learn more
Global Unhackathon
On Feb 28, 2026 — 24+ AI Tinkerers cities worldwide. Heads-down build session: no pitches, no panels, just ship code. Cincinnati details

18

Community — Cincinnati AI

The Cincinnati AI scene is thriving. Multiple overlapping communities for builders, tinkerers, and professionals. Here's where to plug in.

📅
Cincy AI for Humans NEXT: MAR 3
Monthly meetup • 150+ avg attendees • UC Digital Futures Building, Room 140 • 3–5 PM
Eventbritecincyai.org
🛠
AI Tinkerers Cincinnati
Technical builders — live demos, walkthroughs, hackathons.
cincinnati.aitinkerers.org
👥
Cincinnati AI Builders
LinkedIn private group • Carl Fraik / Cincinnati AI Catalyst.
LinkedIn
🌟
CincyAI Week 2026
Annual AI week with talks, workshops, and networking. cincyaiweek.comPre-register

19

Getting Started

Everything you need to go from zero to autonomous agent. Install Claude Code, add skills, connect MCP servers, and start building.

Quick Install
WhatHow
Claude Codenpm install -g @anthropic-ai/claude-code
Custom SkillsCopy skill folder to ~/.claude/skills/
MCP ServersAdd to ~/.claude/.mcp.json
PluginsInstall from marketplace in Claude Code settings
Resources
ResourceLink
Official Docscode.claude.com/docs
Skills Documentationcode.claude.com/docs/en/skills
Anthropic Skills Repogithub.com/anthropics/skills
MCP Protocolmodelcontextprotocol.io
Awesome Claude Codegithub.com/hesreallyhim/awesome-claude-code
Claude Code Full Coursetheneuron.ai

Deep Dive
🚀
Installation Guide — Replicate Our Setup
Step-by-step: skills, MCP servers, plugins, and multi-profile configuration
Custom Skills

Install Skills

Skills are folders with a SKILL.md file. Copy a skill folder into ~/.claude/skills/ and it's immediately available. Our 9 custom skills are authored in-house — build your own following this pattern, or check the Anthropic Skills Repo for community examples.

Our Custom Skills
SkillWhat It DoesTrigger
/screenshotLoad recent screenshots into conversation (multimodal)Auto
/calendarmacOS EventKit calendar via inline Swift — works with ExchangeAuto
/usageleftClaude usage vs. allotment with visual bar graphsUser
/presentGenerate offline-ready HTML slide deck from descriptionUser
/mongoFetch MongoDB sample documents for schema reference (read-only)User
/storelookupFull store deep-dive — identity, devices, GPS, tickets (MongoDB + OneLRA)User
/topemailPriority emails from key contacts via ExchangeAuto
/projectsInteractive project directory pickerUser
/symlinksShow symlinks across multi-profile configUser
Skill File Structure
# Create a skill
mkdir -p ~/.claude/skills/my-skill

# Add the SKILL.md descriptor
cat > ~/.claude/skills/my-skill/SKILL.md << 'EOF'
---
name: my-skill
description: What this skill does (used for auto-activation matching)
user-invocable: true
argument-hint: "optional argument description"
---
# Instructions for Claude when this skill activates
Your prompt and instructions here...
EOF

MCP Servers

Connect MCP Servers

MCP servers are defined in ~/.claude/.mcp.json. Each server is a local process that Claude Code communicates with via the Model Context Protocol.

Apple Mail MCP
Read, search, send, and manage emails. Works with Exchange/Outlook on macOS.
# 1. Clone and set up
git clone https://github.com/your-repo/apple-mail-mcp ~/apple-mail-mcp
cd ~/apple-mail-mcp
python3 -m venv venv
./venv/bin/pip install -e .

# 2. Add to ~/.claude/.mcp.json
{
  "mcpServers": {
    "apple-mail": {
      "command": "~/apple-mail-mcp/venv/bin/python",
      "args": ["-m", "apple_mail_mcp.server"]
    }
  }
}
📝
Notion MCP
Installed via the Notion plugin marketplace. Search, create, and update Notion pages directly from Claude Code.
# Installed as a plugin marketplace
# In Claude Code settings, install from: notion-plugin-marketplace
# Requires a Notion integration token
📖
Context7 MCP
Auto-configured when you install the compound-engineering plugin. Provides live documentation lookup for any library — no additional setup needed.

Plugins

Install Plugins

Plugins extend Claude Code with curated skills, agents, and workflows. Install from marketplaces in Claude Code settings.

compound-engineering ESSENTIAL
The engine behind /workflows. Provides 29 agents, 22 commands, 21 skills, and the Context7 MCP server.
# Install from Every marketplace in Claude Code settings
# Marketplace: every-marketplace
# Package: compound-engineering
After install, you get: /workflows:brainstorm, /workflows:plan, /workflows:work, /workflows:review, /workflows:compound, /lfg, /slfg, review agents (security-sentinel, performance-oracle, etc.), and more.
🐦
swift-lsp OPTIONAL
Swift Language Server Protocol integration. Gives Claude Code type awareness, autocomplete, and symbol resolution for Swift/SwiftUI projects. Requires Xcode installed.
# Install from official marketplace in Claude Code settings
# Marketplace: claude-plugins-official
# Package: swift-lsp

Multi-Profile

Multi-Profile Setup (Optional)

Run separate Claude Code identities for personal and work contexts. Shared skills and plugins, isolated session state.

Shell Aliases
# Add to ~/.zshrc
alias claude="claude"
alias claude-irv="CLAUDE_CONFIG_DIR=~/.claude-irv claude"
alias claude-el="CLAUDE_CONFIG_DIR=~/.claude-el claude"

# Model variants (same config dir, different model)
alias claude-irv-sonnet="CLAUDE_CONFIG_DIR=~/.claude-irv claude --model sonnet"
alias claude-el-opus="CLAUDE_CONFIG_DIR=~/.claude-el claude --model opus"
alias claude-irv-haiku="CLAUDE_CONFIG_DIR=~/.claude-irv claude --model haiku"
Symlink Shared Config
# Create profile directories
mkdir -p ~/.claude-irv ~/.claude-el

# Symlink shared resources (skills, commands, plugins, settings, MCP)
for dir in ~/.claude-irv ~/.claude-el; do
  ln -sf ~/.claude/skills "$dir/skills"
  ln -sf ~/.claude/commands "$dir/commands"
  ln -sf ~/.claude/plugins "$dir/plugins"
  ln -sf ~/.claude/settings.json "$dir/settings.json"
  ln -sf ~/.claude/settings.local.json "$dir/settings.local.json"
  ln -sf ~/.claude/.mcp.json "$dir/.mcp.json"
done
Session state (.claude.json, projects/, todos/, cache/) stays separate per profile automatically.