Your PPC Agency’s New Junior Auditor Runs in Claude Code

What it does Claude Ads is a Claude Code skill that audits paid-media accounts across Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat, and X. It scores each account on a 0-100 weighted scale using more than 250 checks, then emits a versioned JSON report with prioritized findings, quick wins, and a PDF summary for client delivery. The skill also gates account changes behind capability checks, so it audits first and acts only when explicitly allowed.

The interesting bit The project applies software-engineering rigor to marketing audits: a 41-test pytest harness enforces routing snapshots, bidirectional catalog coverage, and deterministic scoring math on every commit. It even imposes a shared “10-Principle Thinking Framework” on how the agent reasons about creative and budget decisions, which is more discipline than most PPC tools admit.

Key highlights

• Twelve platform-specific sub-skills, from Google’s AI Max era to Meta’s Andromeda creative-similarity detection.
• Deterministic, versioned JSON output with health scores, grades, and platform-by-platform breakdowns.
• Public MIT release plus a private community mirror for early access; the public repo is stable and requires no membership.
• Experimental install targets for Codex CLI, Cursor, Windsurf, Gemini CLI, and Goose, though only Claude Code is verified end-to-end.
• A 41-test evaluation harness running in CI to guard against regressions in scoring and routing.

Caveats

• Non-Claude hosts are explicitly experimental; the README warns that skill discovery and sub-skill routing may differ.
• Installation instructions in the README default to a private community mirror, so public-release users must swap the org and plugin slug manually.

Verdict PPC agencies and in-house marketers juggling multiple platforms should treat this as a repeatable second pair of eyes. If you do not manage paid media or live outside Claude Code, there is nothing here for you.