Setup Agent

MCP connectors

Connector	Purpose
Slack	Validate the Chief of Staff delivery channel

Core principle: derive from ROLE.md, confirm with the CAO

Each agent already has a ROLE.md and PROMPT.md in agents/<name>/. Read them first. Derive scope, tasks, and cadence from those files. Only ask the CAO to confirm.

When to use

Run /sylph-setup-agent <name> after completing /sylph-setup (global setup) and before scheduling any agent. Run it once per agent you want to activate.

Usage

/sylph-setup-agent chief-of-staff
/sylph-setup-agent cmo
/sylph-setup-agent all

Available agents

Agent	Directory	Default cadence
chief-of-staff	agents/chief-of-staff/	Daily 08:00
cmo	agents/cmo/	Daily (content), Weekly (planning)
product-manager	agents/product-manager/	Daily
customer-success	agents/customer-success/	Weekly (Monday)
head-of-data	agents/head-of-data/	Daily
head-of-sales	agents/head-of-sales/	Daily
executive-assistant	agents/executive-assistant/	Daily
brand-designer	agents/brand-designer/	On-demand

---

How it works

Step 1: Read the agent definition

1. Read agents/<name>/ROLE.md for identity, responsibilities, decision boundaries, and tone
2. Read agents/<name>/PROMPT.md for the execution routine
3. Read CONTEXT.md for company context

From these files, derive:

• Scope: what this agent owns (list of responsibilities from ROLE.md)
• Tasks: the concrete steps she runs (from PROMPT.md routine)
• Default cadence: how often she runs (from ROLE.md "Working rhythm" or similar)
• Outputs: what she produces and where it goes (from ROLE.md "Output rules")
• Dependencies: which other agents or tools she needs

Step 2: Confirm scope and schedule

Present the derived scope to the CAO:

Here's what [agent name] will do:

Scope:

• [responsibility 1 from ROLE.md]
• [responsibility 2]
• [responsibility 3]

She produces:

• [output 1 and where it goes]
• [output 2]

She escalates to you:

• [escalation 1 from decision boundaries]
• [escalation 2]

Proposed schedule: [cadence from ROLE.md, e.g. "Daily at 08:00"]

Anything to add, remove, or change?

Incorporate the CAO's corrections. If the CAO adjusts the schedule, note it.

Step 3: Chief of Staff only - validate Slack delivery

The Chief of Staff is the only agent that writes to Slack. She delivers her daily briefing as a Slack DM to the CAO.

Only for chief-of-staff:

1. Connect to Slack - if not already connected, ask the CAO to connect the Slack MCP
2. Ask the CAO:

Where should the Chief of Staff deliver her daily briefing? (Slack DM to you, or a specific channel?)

3. Validate the channel or DM exists via Slack MCP
4. Save the Slack user ID or channel ID for delivery

For all other agents, skip this step. Other agents write to the repo (_drafts/, _logs/, _plans/) and deliver in chat. They don't post to Slack.

Step 4: Write agent config

Create or update agents/<name>/CONFIG.md with the confirmed settings:

---
agent: [name]
status: active
---

# [Agent Name] - Configuration

## Schedule
- Cadence: [daily / weekly / on-demand]
- Time: [HH:MM local time, if scheduled]
- Special: [e.g. "Monday: extended run with weekly planning"]

## Scope
- [responsibility 1]
- [responsibility 2]
- [responsibility 3]

## Escalation rules
- [what she escalates to the CAO]

## Dependencies
- Agents: [other agents she delegates to or receives from]
- Tools: [MCP connectors she needs]

For the Chief of Staff only, add the delivery section:

## Delivery
- Channel: [Slack DM or channel name]
- Channel ID: [Slack channel/user ID]
- Fallback: chat message (if Slack fails)

Step 5: Schedule the Chief of Staff as a cloud agent

The Chief of Staff should run automatically every morning. After writing her CONFIG.md:

1. Create a scheduled remote agent using the scheduling system (Claude Code's /schedule or equivalent)
2. Configure the schedule:
   • Trigger: cron, daily at the CAO's preferred time (default: 08:00 local)
   • Prompt: "Run the chief-of-staff routine: load ROLE.md and PROMPT.md from agents/chief-of-staff/, execute the full daily routine, commit the briefing, and deliver via Slack DM."
   • Monday variant: the routine automatically branches for weekly planning (this is handled in PROMPT.md, no separate schedule needed)
3. Verify the schedule is active and show the CAO a confirmation:

Chief of Staff is now scheduled to run every day at [HH:MM]. She'll deliver her briefing to [Slack channel/DM]. You can also invoke her on demand with /chief-of-staff.

For other agents with daily cadence, inform the CAO they can be scheduled the same way but don't auto-schedule them. Ask:

Want me to schedule any other agents to run automatically? (e.g., CMO daily, Head of Data daily)

Only schedule additional agents if the CAO explicitly asks.

Step 6: Test the agent

1. Run a dry test of the agent's routine
2. Verify she can access all required MCP connectors
3. For the Chief of Staff: verify she can post to the configured Slack channel
4. Show the CAO a sample output
5. Capture any feedback in the agent's CONFIG.md

---

Batch mode: /sylph-setup-agent all

When the CAO runs /sylph-setup-agent all:

1. List all agents in agents/ directory
2. For each agent, run Steps 1-4 in sequence
3. For the Chief of Staff, also run Step 5 (schedule)
4. Present a summary table at the end:

| Agent | Schedule | Status |
|-------|----------|--------|
| chief-of-staff | Daily 08:00 (scheduled, Slack DM) | Active |
| cmo | Daily (on-demand) | Active |
| product-manager | Daily (on-demand) | Active |
| customer-success | Weekly Monday (on-demand) | Active |
| head-of-data | Daily (on-demand) | Active |
| head-of-sales | Daily (on-demand) | Active |
| executive-assistant | Daily (on-demand) | Active |
| brand-designer | On-demand | Active |

5. Ask: "Chief of Staff is scheduled. Want me to schedule any other agents too?"

---

Adding a new agent

If the CAO asks to set up an agent that doesn't exist yet:

1. Create the directory: agents/<name>/
2. Ask the CAO to describe what this agent should do (2-3 sentences)
3. Generate ROLE.md following the pattern of existing agents:
   • Identity section
   • Responsibilities (numbered)
   • Decision boundaries (you decide / you escalate)
   • Tone
   • Output rules
4. Generate PROMPT.md with the execution routine
5. Create the matching skill in .claude/skills/<name>/SKILL.md
6. Add to the skill index in AGENTS.md
7. Then proceed with the normal setup flow (Steps 2-6)