product-os: Command for Claude Code

Set Up Persistent Agent

Set up a new persistent agent for a PM. This command creates the full workspace, personalises the agent definition, and bootstraps context from every available source — so the PM gets a working agent they can start testing immediately.

Arguments: {initials}, {name}, {agent} where agent is one of: cos, analyst, uxr, coach, manager, strategist, engineer

Example: /setup-agent SJ, Steve, cos

Instructions

Phase 1: Validate inputs and check tools

1. Parse the arguments. If any are missing, ask for them:

   • Initials (e.g. SJ)
   • Full first name (e.g. Steve)
   • Agent type — must be one of: cos, analyst, uxr, coach, manager, strategist, engineer

2. Check which MCP integrations are available by testing each one silently (don't show errors to the user, just note which work):

   • Calendar (e.g. Google Calendar) — try to list today's events
   • Meeting notes (e.g. Granola, Otter) — try to list recent meetings
   • Messaging (e.g. Slack) — try to read a channel
   • Data warehouse (e.g. BigQuery, Snowflake) — try a simple query (only if agent is analyst)

3. Tell the user which tools are connected and which aren't. Then decide what to do based on coverage:

   If the critical MCPs for this agent are connected (see table below), report the status briefly and continue to Phase 2:

   Tools connected: Calendar, Meeting notes, Messaging Not available: Data warehouse (not needed for COS)

   If critical MCPs are missing, pause and explain why they matter before continuing. Use this framing:

   Tools connected: Messaging Not connected: Calendar, Meeting notes

   I can set up your agent now, but I'd recommend connecting Calendar and Meeting notes first. Here's why:

   • Calendar — without it, I can't see who you meet, how often, or when. I'll have to create blank meeting context files that you fill in manually instead of pre-populated ones with real people and topics.
   • Meeting notes — without it, I can't read your meeting transcripts to find action items, decisions, or standing topics. Your todo list and state file will be based on team priorities only, not your actual work.

   The difference is significant: with these tools, your first test output will reference your real meetings, real people, and real priorities. Without them, it'll be generic and you'll spend much longer correcting it.

   To connect them: open Claude Code settings (type /mcp in Claude Code) and enable the integrations you have available. Once they're connected, run /setup-agent again.

   Or if you'd rather continue now with what's available, just say "continue" and I'll do the best I can with local files.