AgriciDaniel/claude-ads

Claude Ads – AI‑powered ad‑account audit skill

What it is – An open‑source skill (plugin) for Claude Code (and compatible AI‑agent runtimes) that reads exported ad‑account data (CSV/JSON, screenshots, or summaries) and produces a structured audit. It works across twelve major ad platforms (Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat, X) and can also draft plans, creative briefs, monitoring checks, optimisation suggestions, and test designs.

How it works –

1. Setup – you tell the skill about your business model, goal, spend, platforms, and approval workflow (no API keys are ever pasted).
2. Evidence ingestion – you upload an account export or paste a summary; the skill stores the data locally (read‑only by default).
3. Audit – a set of platform‑specific workers analyse the data and return findings that are each marked pass, fail, unknown (data missing) or not_applicable. Every finding includes the evidence that led to it.
4. Outputs – you can render the audit as Markdown/HTML/PDF, view the most critical blocker (/ads next), or ask for downstream commands (/ads plan, /ads create, /ads monitor, /ads optimize --draft, /ads experiment).

Key commands

Command	Purpose
/ads setup	Record client, goal, spend, platform list, and approval guardrails
`/ads audit [platform	all]`
/ads plan	Generate channel, budget, competitor and measurement recommendations
/ads create	Produce copy, image, video or product‑photo briefs for new ads
/ads launch --draft	Draft a campaign‑change plan without touching the live account
/ads monitor	Ongoing checks for pacing, fatigue, policy compliance, etc.
/ads optimize --draft	Evidence‑backed improvement suggestions
/ads experiment	Design a controlled test or read out a completed test
/ads report	Export a saved run in Markdown, HTML or PDF
/ads next / /ads status	Show the current audit state and the highest‑priority blocker

Safety & trust – All adapters are read‑only unless a specific capability is enabled, the exact object IDs are supplied, a human‑readable diff is shown, and an owner‑approved ceiling is respected. No credentials are stored in the repo; they must be provided via environment variables or a secret manager.

Optional companion – Claude Ads Cockpit (a separate plugin) provides a local UI to explore audit runs, compare with previous runs, and view synthetic demo data.

Installation – Preferred via Claude Code’s marketplace:

/plugin marketplace add agricidaniel/claude-ads
/plugin install claude-ads@ai-marketing-hub-claude-ads

Alternatively, clone the repo and run bash install.sh --source=local (supports other hosts such as Codex, Gemini, Cursor, etc., but only Claude Code is officially verified).

Project status – development candidate: not release‑ready until all gates in control-plane/RELEASE_REQUIREMENTS.md pass. All twelve health‑scoring profiles are currently disabled, so audits only report findings and evidence coverage. The Cockpit companion is optional and not yet accepted as a native plugin.

Who it’s for – Marketing agencies, consultants, and in‑house teams that want an AI‑assisted, evidence‑backed review of their ad accounts without giving the AI direct write access.

License – MIT for the code and documentation; third‑party APIs retain their own terms.

---

All details above are taken directly from the repository’s README.