Agent skills

Two Agent Skills ship with SaaS Pro Max. A skill is a folder of markdown that teaches a coding agent a workflow: when to reach for it, what order to do things in, and what never to do. They load themselves when what you ask matches the skill's description — you do not invoke them by name.

Prerequisite#

A skill needs a way to reach the platform. Either:

1. The MCP server — every tool name in the skills is an MCP tool name:

   claude mcp add saaspro --env SPM_TOKEN=spm_pat_xxxxxxxx -- npx -y @saaspro/mcp
   

2. A personal access token exported as SPM_TOKEN, used against the Management API directly. Every tool has a matching endpoint and the workflow is identical.

Create a token at Account → Personal access tokens or Organization → Access tokens, scoped to the job: read for reporting and triage, write to create funnels, flags and alert rules or to resolve an issue, admin only when the agent will create a public or secret key during an integration.

Install#

Into one repository, so the whole team gets it:

mkdir -p .claude/skills
cp -R skills/saaspromax-integrate .claude/skills/
cp -R skills/saaspromax-analyze   .claude/skills/

Or into your own account, for every project:

mkdir -p ~/.claude/skills
cp -R skills/saaspromax-integrate ~/.claude/skills/
cp -R skills/saaspromax-analyze   ~/.claude/skills/

Restart the agent so it re-reads the directory. Other hosts that implement the Agent Skills format read the same folders — check where yours looks.

saaspromax-integrate#

The workflow, in order: reach the platform → detect the framework → get an application → get a key → install and wire → verify → report.

It carries a reference file per framework — Next.js App Router, Vite + React, Express, Hono, plain HTML, and the signed connector — each adapted from docs/sdk/recipes, so the tracking code matches the published SDK APIs rather than an approximation of them.

What it wires, in priority order: pageviews, identify on sign-in, two or three track calls on the actions that matter, error capture, one metric on a hot path, one feature flag gate. One real metric beats ten speculative ones, and the same is true of events.

It ends by loading a page and calling list_events. It will not report success from a clean build.

saaspromax-analyze#

Four recipes: the weekly report, funnel drop-off diagnosis, error triage, and alert rules with thresholds derived from the last 30 days rather than from intuition.

Its rules are what make the output trustworthy:

• Every number comes from a tool result, and the answer names the tool inline.
• "No data" and "zero" are different answers, and it says which.
• It does not assert a cause — it says what would explain the number and which call would confirm it.
• It says what it could not see: a module that is off, a permission missing, a connector not configured.
• Before any write — resolving an issue, creating an alert rule — it says what it is about to do.

Writing your own#

The format is a folder with SKILL.md and YAML frontmatter:

---
name: my-skill
description: What it does, and the phrases that should trigger it.
---

# My skill

…

description is the whole routing mechanism: it is what the agent matches against a request, so write the trigger phrases a person would actually use, not a summary. Put anything long — per-framework tracking code, output templates — in references/ and link to it, so the agent loads it only when it needs it.