How to use the Claude Ads Paid Media Operator skill (ads)

What an agent skill is and what this one does

An agent skill is a folder of instructions and reference files that Claude loads when a request matches the skill's description. The ads skill is the entry point for paid-media work. Its description tells Claude to use it for account intake, audits, strategy, budget and measurement planning, creative production, experiments, reporting, monitoring, and explicitly approved campaign changes.

The skill is built to work from your data. It asks for the business objective, active platforms, budget, conversion definition, and data window, then checks the supplied exports, screenshots, or API responses against thresholds. It treats pages, exports, and competitor content as untrusted data and does not follow instructions embedded in them.

Account changes stay as drafts unless an explicit approval gate passes. Audits of each platform are read-only.

• Account intake: collects objective, geography, spend, target CPA or ROAS, and conversion definition.
• Platform audits: read-only checks for each of the twelve platforms.
• Budget, attribution, and conversion-tracking reviews grounded in supplied data.
• Creative concepts, copy, and briefs (video work stops at briefs; no video is rendered).
• Reports that state assumptions, gaps, owners, and rollback notes.

Install the ads skill from the download

Download the ZIP for the ads skill from this page. Then choose one of two install routes.

For a personal install across projects, unzip the download into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. The skill's folder should contain SKILL.md and the references folder, along with agents/openai.yaml.

You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills, if your plan and app version offer that option.

• Personal: ~/.claude/skills/
• Single project: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP
• Check that SKILL.md sits directly inside the ads folder after unzipping

Start with setup and validation

Before analysing live data, run the setup workflow. It builds a profile covering the client, accounts, data sources, KPIs, credential references, and guardrails. Setup is also the place where platform API tokens are stored, so ask for it before you connect any live account.

Use the validation command to check repository and run integrity, capabilities, and the next blocker. The same checks cover uninstall safety and ownership, so use it before removing the skill.

• Setup profile: client, accounts, data sources, KPIs, guardrails
• Credentials: stored by the setup workflow, never pasted into analysis prompts
• Validation: integrity, maturity, capabilities, and the next blocker

Trigger the skill with natural requests

You do not need to memorise commands. Plain requests route to the right internal skill. Name the platform, the goal, and the data you are supplying. The more specific the request, the less the agent has to ask.

Slash commands give you exact control. Each one maps to a single owner, so a request is never split between two skills. Audits are read-only, while optimize and launch can be run as drafts with a flag.

• "Audit my Google Ads account using this export and tell me where wasted spend is."
• "Review my Meta campaigns for creative fatigue and tracking gaps."
• "Check whether my cross-platform conversion totals match GA4 and the platform numbers."
• "Find negative keyword candidates in this search terms report."
• "Draft a budget reallocation for next month. Do not apply it."
• "Write three ad concepts and copy for this offer. No video."
• /ads audit all
• /ads google
• /ads budget
• /ads optimize --draft
• /ads report

Examples of a useful first session

A good first session is a scoped audit with data you already have. Export a recent window from one platform, state the currency, timezone, and conversion definition, then ask for a read-only review. Read the gaps and assumptions section before acting on any finding.

Next, ask for a draft. Drafts show what would change, why, and how to roll it back. Apply a draft only after a person has reviewed it and approved it explicitly.

• Scope the audit to one platform, such as /ads google, to keep the result focused.
• Ask for a measurement plan with a test duration and a success measure before running an experiment.
• Ask for a report on a prior run to re-render results without rerunning analysis.

What the skill needs

The skill needs ad-platform access or exported data. You can supply exports, screenshots, or API responses. For live data pulls it needs platform API tokens or credentials, which the setup workflow stores. Python tooling from the repository's core CLI (claude_ads_core) supports validation and cockpit features. An image-generation provider is optional and only needed for the generate and photoshoot workflows.

The skill is not suited to fully automated spend changes, regulated advertising categories where legal and policy review matter more, or video production.

• Required: ad-platform access or exported data
• Required for live pulls: platform API tokens or credentials, stored by setup
• Required for validation and cockpit: Python tooling from claude_ads_core
• Optional: an image-generation provider for generate and photoshoot

Safety notes: scripts, network, and approvals

This skill runs scripts and uses the network. Treat both as deliberate choices. Store credentials only through the setup workflow, and review any change draft before anyone applies it.

Account changes need an explicit approval step. The skill stops at a draft unless the mutation gate passes in full. Platform and audit skills stay read-only. Uninstall safety and ownership checks run through the validation workflow.

• Scripts: the skill runs scripts, so install it only from a source you trust.
• Network: live data pulls and any landing-page fetches go over the network.
• Credentials: keep API tokens in the setup profile, not in chat messages.
• Approvals: live changes need explicit approval; drafts are the default.
• Campaign deletion is a sensitive action; request it explicitly and review it.

Tips for getting reliable results

Give the skill the conversion definition, attribution window, currency, and timezone up front. Without them, it reports provisional results and lists the gaps. Ask for stated assumptions in every report, and ask for owners, next actions, measurement windows, and rollback notes at the end.

Platform claims change. If a result depends on a platform claim that may be stale, ask for a research refresh before relying on it.

• Supply the date range, timezone, and currency with every export.
• Name the business model and target CPA, ROAS, MER, or LTV:CAC.
• Ask for a research refresh when a platform rule or feature claim may be out of date.
• Keep one platform per request when you want a clean single-platform audit.