Product Metrics agent skills
Skills teach a coding agent or AI assistant the real steps to set up, check and use Product Metrics.
A skill is a SKILL.md file of task instructions an agent loads when your request matches it. The Product Metrics skills tell the agent which productmetrics commands or MCP tools to call, in which order, what to hand to you in a browser, and how to check the result.
Skills are instructions, not code. They don’t run anything on their own and don’t hold credentials: the agent still signs in through the productmetrics CLI or the MCP server, and you approve browser steps yourself.
Available skills
Each skill works through the CLI or the MCP server. analyse-google-ads needs the MCP analysis tools for its reports.
From nothing to tracking active: workspace, store, Google Ads, Meta, consent, costs, first purchase.
Connect Google Ads and Merchant Center, or fix a broken connection.
Check every part of tracking and find a test order. Read only.
Set up and check POAS or ROAS bidding in Google Ads and Meta.
Answer “how are my campaigns doing” and “why did ROAS drop” with real figures. Read only.
Add them to Claude Code, Cursor or Codex with one command.
Install
From the project where your agent works (needs Node.js 22.20 or later):
npx skills add https://www.productmetrics.io/docsThe CLI lists the skills and the agents it found, and asks which to install. Flags for each agent are on Install skills.
Skills, CLI and MCP
| Path | Use it for |
|---|---|
| Skills | Teach the agent the real steps for a task |
CLI (productmetrics) | Setup and checks from a terminal or coding agent |
| MCP | Analysis and setup from Claude, ChatGPT or Cursor |
How agents find them
The skills are published with these docs. An agent (or the skills CLI) reads the discovery index:
curl -s \
https://www.productmetrics.io/docs/.well-known/agent-skills/index.jsonYou should see a skills list with one entry per skill. Each entry’s url serves the skill file.
Once installed, the agent sees each skill’s name and description, and loads the full instructions only when your request matches one: “set up Product Metrics for my Shopify store” loads set-up-tracking.
What every skill does the same way
- Discovers, never guesses. Ids come from earlier results; commands are checked with
productmetrics <command> --help --jsonor the MCPsearchtool first. - Hands browser steps to you. When a command exits with code 4 (
next_action.type: "browser"), the agent gives you the URL and waits. - Keeps secrets out of the chat. It never asks for passwords or tokens. The exceptions are credentials the API needs from you (a Lightspeed API key and secret, a Meta Conversions API token): it passes them on and never repeats them.
- Writes for store owners. Plain answers, with the exact app labels for the steps you take yourself.