Skip to content

commands_guide.md

Latest commit

 

History

History
60 lines (46 loc) · 2.1 KB

commands_guide.md

File metadata and controls

60 lines (46 loc) · 2.1 KB

⌨️ Analytical Commands Guide

This project includes specialized "Slash Commands" that turn your AI clients (Claude Desktop, OpenCode, Antigravity) into expert marketing analysts. These commands orchestrate multiple MCP tools to give you instant, governed answers.

🚀 How to Use Commands

1. Claude Desktop

Type the command directly into the chat box. Claude will see the definition in the cowork_plugin folder and execute the steps.

  • Example: "Run /marketing for last quarter."

2. OpenCode CLI

Type the command in the terminal chat. OpenCode uses these as "Expert Skills."

  • Example: > /attribution what is the ROAS difference between models?

3. Antigravity IDE

Ask your assistant to follow a specific command blueprint.

  • Example: "Use the /pipeline command to analyze our funnel bottlenecks."

📜 Command Reference

📈 /marketing

Goal: Blended performance analysis across Google Ads, Meta Ads, and GA4.

  • Tools used: google-ads, meta-ads, ga4, salesforce.
  • Primary Metric: Blended CAC (Cost / Conversions) and ROAS.
  • Best for: Weekly executive summaries and budget allocation reviews.

🔍 /attribution

Goal: Compare First-Touch, Last-Touch, and Linear models to find hidden gems.

  • Tools used: dbt-semantic-layer.
  • Primary Metric: Revenue credit by model.
  • Best for: Decisions on top-of-funnel (awareness) vs. bottom-of-funnel (direct) spend.

🧬 /pipeline

Goal: Full funnel visibility from CRM leads to deal revenue.

  • Tools used: hubspot, salesforce, dbt-semantic-layer.
  • Primary Metric: Win rate, Lead-to-Opp conversion, Pipeline Velocity.
  • Best for: Sales and Marketing alignment meetings and revenue forecasting.

🛠️ Adding New Commands

You can create your own commands by adding a new Markdown file to:

  • cowork_plugin/commands/ (for Claude Desktop)
  • .opencode/commands/ (for OpenCode CLI)

Template:

---
description: Brief description of the command.
---
# /yourcommand [query]
## Steps
1. Call [mcp-server].[tool-name]
2. ...
3. Provide summary.