Performance Marketing skills

01 · What it does

The problem it solves

A skill library for AI coding agents: 25 workflows backed by 88 reference documents covering Google Ads (Search, Performance Max, Shopping, Display/YouTube, measurement), Meta Ads (structure, audiences, creative, optimisation, Conversions API), plus Microsoft, LinkedIn, TikTok, programmatic and Amazon. Each skill states what inputs to gather, a step-by-step workflow and a quality checklist, so an agent asked to audit an account follows a documented method rather than improvising. Four standard-library scripts handle the arithmetic and the safety gates.

02 · Setup

Before you start

• ✓An agent that reads AGENTS.md or supports the Agent Skills spec
• ✓Node.js with npx, if you install through the CLI
• ✓Python 3 to run the scripts — all standard library, no pip install
• ✓Separate credentials and a platform MCP server if you want live account data; this repo provides none

03 · Step by step

Getting it running

1. 01

   Install it into your project

   The CLI path pulls all 25 skills; --skill takes a named subset and --list prints what is available. Claude Code users can add the marketplace instead, which wires up skills, agents and commands together.

   $npx skills add ca-who-codes/Ultimate-Performance-Marketing-Google-Ads-Meta-ads-etc.-
2. 02

   Point your agent at the manual

   AGENTS.md is the portable operating manual — role, navigation order, operating loop and non-negotiables. CLAUDE.md is the short version that defers to it for detail.

3. 03

   Start from the foundations skill

   It holds the funnel and KPI glossary, the channel-selection matrix and full-funnel budget allocation. Every other skill assumes it has been read.

4. 04

   Ask in plain language and let it route

   An audit request goes to the audit command or the auditor subagent, a launch brief to the launch planner, a CPA spike to the diagnose command.

   $/ads-audit google
5. 05

   Run the math directly when you want a number

   Eight subcommands — roas, breakeven-roas, cpa, breakeven-cpa, ltv-to-cac, sample-size-per-variant, test-duration-days and budget-pacing. The example prints "ROAS: 4".

   $python3 scripts/ads_math.py roas --spend 2000 --revenue 8000
6. 06

   Hash identifiers before any Conversions API call

   Normalises and SHA-256 hashes email, phone and name fields the way Meta CAPI and Google Enhanced Conversions expect, returning JSON.

   $python3 scripts/capi_hash.py --email user@example.com
7. 07

   Gate anything that writes to a live account

   The guardrails script is meant to be sourced, not executed. It defines the approval gate so any write-shaped call passes through it before reaching a platform API.

   $source scripts/safety_guardrails.sh

04 · What you get

Capabilities

05 · Before you rely on it

Limitations, honestly

Every tool has edges. These are the ones that will cost you time if nobody tells you first.

• !Benchmark figures are directional starting points — verify CPC, CTR and ROAS bands against the account’s own data.
• !Ad platforms change bidding, targeting and policy often, so platform-specific detail can go stale between updates.
• !No MCP server is included. Live account access means attaching one separately and applying the guardrails yourself.
• !The guardrails script only catches action names matching its mutation regex, so a new write action with an unmatched name passes silently — its own comment flags this.
• !The hashing script echoes raw input at the CLI and its phone normaliser is regex-based; it is a local testing tool, not production code.
• !The full library runs to roughly 150,000 words, so an agent loading everything at once will exhaust its context — the progressive-disclosure order exists for that reason.
• !The compliance skill is orientation on platform policy, explicitly not legal advice. No tests and no CI.