claude-ads for Claude Code and Codex: an ad account audit skill pack, installed and run

claude-ads is a skill pack for auditing, planning, and reporting on advertising accounts across 12 platforms. This post covers the problem it solves, who it suits, how to install it and run a first audit, and the dependencies that catch people before the first report renders.

What problem it solves

Taking over a new advertiser usually means opening Google Ads, Meta Ads, Amazon Ads, and whatever else the client runs, then judging each account by a different checklist. The result is hard to compare across platforms and hard to defend when someone asks what a grade was based on. claude-ads puts those platforms in one system. The README lists 12 of them: nine search, video, and social platforms and three commerce and retail media platforms.

What sets it apart is how it treats evidence. An audit marks each item as pass, fail, unknown, or not applicable. It gives a grade only when evidence coverage reaches 80%, a provisional grade between 60% and 79%, and an insufficient-evidence label below that. So a thin account produces an honest gap instead of a confident score.

Who it fits

The chapter rates four situations. Three are enough to decide.

Situation	Fit	Reason
Diagnosing a new account	Good fit	Evidence-based audits
Monthly performance reports	Good fit	Report output
Running small-budget ads	Poor fit	Substantial setup work

Writing creative briefs sits in the middle. It works, but the pack focuses on audits rather than creative work.

Install and first run

Install it as a Claude Code plugin, or clone the repository and install locally for Codex.

# Install the Claude Code plugin
/plugin marketplace add agricidaniel/claude-ads
/plugin install claude-ads@ai-marketing-hub-claude-ads
# Install locally for Codex
git clone https://github.com/AgriciDaniel/claude-ads.git
cd claude-ads
bash install.sh --target=codex --source=local

Start with /ads setup to set the client, accounts, KPIs, personal information handling, and a safeguard profile. Then call the audit and point it at the account you want reviewed.

/ads audit
[advertising account to review]

To check a single platform, add its name, as in /ads google or /ads amazon. In the Claude Code plugin, prefix commands with /claude-ads:. Results are saved as versioned JSON and rendered again as Markdown, HTML, or PDF.

What trips people up

The pack is not a single file you drop in. It needs Python 3.11 or 3.12, browser capture requires Playwright, and PDF report rendering also needs WeasyPrint and the Pango library. If the PDF step fails, check those last two before anything else. Claude Code is the reference environment; the README says other runtimes such as Codex, Gemini, Cursor, Windsurf, and Goose use the same skills within their supported capabilities. And because the pack talks to ad platforms, keep API credentials in environment variables or a keychain, never in repositories, profiles, reports, or logs.

Where this comes from

This post adapts part of chapter 02-03 of The Claude Code & Codex Skills Guidebook. The full chapter covers the six commands from setup through planning, asset creation, monitoring, and reporting, and how the pack rates source confidence in four categories, and the book has 52 more skills like this one. Read the full chapter in The Claude Code & Codex Skills Guidebook on Gumroad.