Welcome to The Agency.
 
This is a multi-agent orchestration system for Claude Code. Here's what
it gives you:
 
  Agents    225+ specialists across 19 departments (Engineering, Design,
            Marketing, Content, Sales, Testing, Game Dev, and more).
            Each agent has a role, a model assignment, and a protocol.
 
  Skills    270+ slash commands you can invoke right here. /save-state,
            /recall, /pd-resume, /delegate, /swarm, /graphify, and more.
 
  Memory    Persistent project state that survives across sessions.
            /save-state at the end, /recall at the start. Your agents
            remember where they left off.
 
  PDs       Project Directors — autonomous agents that decompose work,
            spawn specialists, run QA gates, and report back. You
            supervise, they execute.
 
  Model     Opus for planning, Sonnet for execution, Haiku for bulk
  routing   work. Every agent carries a model assignment.
 
Let's verify your setup and get you running.

# Check these directories
ls -d ~/.claude/skills/ 2>/dev/null    # Skills installed?
ls -d ~/.claude/agents/ 2>/dev/null    # Agents installed?
ls -d ~/.claude/core/ 2>/dev/null      # Core docs installed?
ls -d ~/.claude/projects/ 2>/dev/null  # Projects dir exists?
ls -d ~/.claude/memory/ 2>/dev/null    # Memory dir exists?

# Count skill directories (each has a SKILL.md)
find ~/.claude/skills -name 'SKILL.md' -maxdepth 2 | wc -l
 
# Count agent .md files
find ~/.claude/agents -name '*.md' | wc -l

which agency 2>/dev/null || echo "NOT ON PATH"

Some components are missing. Run this first:
  cd <repo-path> && ./install.sh
Or:
  cd <repo-path> && node cli/bin/agency.js init
Then re-run /onboard.

✓ Installation verified
  Skills: {n} installed
  Agents: {n} installed
  CLI: agency on PATH

Skills are slash commands. Type / in Claude Code and you'll see them.
The Agency installed {n} skills. Here are the essential ones:
 
  Session lifecycle:
    /save-state [slug]     Save everything before closing
    /recall [slug]         Load project briefing at session start
    /pd-resume [slug]      Resume a PD with full context
 
  Coordination:
    /swarm                 Status check across all projects
    /delegate              Hand off work to a specialist agent
 
  Content:
    /content-polish        Humanize + proofread in one pass
    /blog-pipeline         Research → write → polish blog post
 
  Quality:
    /qa                    Run QA gates on current work
    /review                Code review via subagent
 
  Planning:
    /autoplan              CEO + design + engineering review pipeline
 
Try one now — type /recall to see if any projects exist.

test -f ~/.claude/skills/save-state/SKILL.md && echo "save-state OK"
test -f ~/.claude/skills/recall/SKILL.md && echo "recall OK"
test -f ~/.claude/skills/delegate/SKILL.md && echo "delegate OK"

agency help 2>/dev/null

git --version

# Graphify — knowledge graph generation
which graphify 2>/dev/null && echo "graphify OK" || echo "graphify not installed (optional — for /graphify skill)"
 
# gws — Google Workspace CLI
which gws 2>/dev/null && echo "gws OK" || echo "gws not installed (optional — for Gmail/Drive/Calendar)"
 
# agent-browser — headless browser automation
which agent-browser 2>/dev/null && echo "agent-browser OK" || echo "agent-browser not installed (optional — for /browse skill)"

CLI tools:
  ✓ agency          Project management CLI
  ✓ git             Version control
  ○ graphify        Knowledge graphs (optional)
  ○ gws             Google Workspace (optional)
  ○ agent-browser   Browser automation (optional)
 
Optional tools can be installed later. The core system works without them.

MCP servers extend what Claude Code can do. You connect them in
Claude Code settings. Here's what's useful with The Agency:
 
  Already connected (detected in this session):
    {list any MCP servers visible in the current tool list}
 
  Recommended:
    Playwright     Browser automation for /browse and QA testing
    Pinecone       Long-term memory and semantic search
    Firebase       Deploy and manage Firebase projects
    Supabase       Database and auth
 
  Optional:
    Gmail/Calendar/Drive    via Google Workspace MCP
    Slack                   Team notifications
    Canva                   Design generation
    Zapier                  8000+ app automations
 
You can add these anytime in Claude Code settings → MCP Servers.

Let's create your first project.
 
What are you building? Give me a name and a one-line description.
Example: "my-saas-app" — "Task manager with React and Supabase"

agency new <slug> "<description>"

You already have projects:
  {list from agency status}
 
Want to create another, or start working with an existing one?

Your project "{slug}" is ready. Here's the daily workflow:
 
  START OF SESSION
    /recall {slug}
    → Loads where you left off. First time it'll be empty.
 
  DURING SESSION
    Tell the PD what to build. It decomposes, delegates, and reports.
    Example: "Build the auth system with email + Google OAuth"
 
  END OF SESSION
    /save-state {slug}
    → Saves everything. Tomorrow /recall picks up exactly here.
 
  UPDATING THE AGENCY
    agency upgrade
    → Pulls latest skills and agents from GitHub.
 
  IF THINGS BREAK
    bash rescue.sh
    → Fixes any git/CLI state.
 
Want to try /recall {slug} now to start your first session?

Onboarding complete. Here's your cheat sheet:
 
  /recall {slug}          Start a session
  /save-state {slug}      End a session
  /pd-resume {slug}       Resume with full PD context
  /swarm                  Check all projects at once
  /delegate               Hand off to a specialist
  agency upgrade          Get latest updates
  agency status           See all projects
 
  Docs:
    docs/ARCHITECTURE.md  How the system is structured
    docs/SKILLS.md        Full skill reference
    docs/SETUP.md         Setup troubleshooting
 
Welcome to The Agency.