How to use the Paid Media Optimization agent skill (ads-optimize)

What is an agent skill and what does ads-optimize make the agent do?

An agent skill is a folder containing a SKILL.md file, which gives an AI agent a documented procedure to follow for a certain kind of task. When the skill is installed and your request matches its description, the agent uses that procedure.

The ads-optimize skill makes the agent load the latest normalized campaign snapshot, prior decisions, monitoring results, experiment state, conversion lag, economics, and the platform capability manifest. It then identifies the decision and the causal evidence, compares no change, an experiment, and a change to the account, and produces ranked recommendations with a confidence level and success measures for each.

• Default behaviour is draft mode (--draft): the agent proposes, it does not change the account.
• It does not optimise a single metric in isolation, and it will not treat a fixed CPA multiple, budget ratio, benchmark, or novelty claim as sufficient reason to change an account.
• Approved recommendations can only be turned into mutation plans through the main mutation gate.

Install the ads-optimize skill

Download the ZIP for the skill from this page and unzip it. You can install it in one of two ways, depending on where you use your agent.

For Claude Code, unzip the skill folder into ~/.claude/skills/ to make it available to you across projects, or into a project's .claude/skills/ folder to make it available only in that project. The folder should contain SKILL.md.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need a special command. The skill is meant to be triggered by describing the paid advertising task in plain language. The skill description covers campaign optimization, budget reallocation, bid changes, pausing or archiving ads, requests to delete campaigns, search-term or negative-keyword actions, creative rotation, and improving CPA or ROAS.

Give the agent your data and your goal together. The more it knows about the decision, the constraints, and the recent history, the better the recommendation.

• Should I move budget from Campaign A to Campaign B this month, given our target CPA?
• Review the search terms report and tell me which terms are worth reviewing as negatives, with the evidence for each.
• Rank the bid changes for our top ad groups by confidence, and tell me how we would measure success.
• Which ads should be paused or archived, and what should we check before rotating creatives?

Examples of what to ask and what you will get back

A typical answer is a ranked list of options, each with its evidence, the expected effect, a confidence level, and the measure you would use to decide whether it worked. The agent also notes learning, tracking, policy, inventory, and opportunity-cost effects where relevant.

If you ask for a search-term or negative-keyword action without a search terms report and conversion data, the agent will ask for that evidence instead of guessing. It will not invent negative keywords to illustrate the point.

If you ask the agent to permanently delete campaigns, ad groups, ads, audiences, or conversions, it will refuse. It will offer reversible options such as leaving objects paused, archiving where supported, applying labels, exporting a backup, and setting a retention or later-review date.

What you need before you start

The skill depends on inputs you provide or that your setup supplies. SKILL.md refers to a normalized campaign data snapshot and a platform capability manifest but does not name specific platform APIs or ad accounts, so check how your own data is prepared.

For search-term actions you need a search terms report, business-relevance evidence, and conversion data. Without these, the skill asks for them rather than proceeding.

• A normalized campaign data snapshot and a platform capability manifest, as referenced by the skill
• A search terms report and conversion data for any search-term actions

Safety notes: scripts, network and live account changes

The skill as documented does not run scripts and does not use the network. Its only file is SKILL.md.

Changes to a live account are the sensitive part. The skill is designed to apply a change only when that exact operation is explicitly enabled and the remote account still matches the draft precondition. SKILL.md does not describe how apply mode is enabled, so do not assume that account changes will happen automatically. Review every draft yourself before acting on it, and keep the rollback and audit trail the skill describes.

• No scripts are run and no network use is documented.
• Draft mode is the default; apply mode is not explained in the file.
• Permanent deletion is refused and is not something the skill can make safe with confirmation.

Tips for getting useful results

Give the agent a clear decision to make, a target such as a CPA or ROAS goal, and any financial constraints. Mention recent experiments and changes so it does not repeat or undo them. Conversion lag matters, so say if recent conversions may still be arriving.

Ask for confidence levels and success measures explicitly if you want them in a particular format.

• State the decision, the goal, and the budget or financial constraint up front.
• Include the latest experiment status and recent changes.
• Provide the search terms report and conversion data together when working on negatives.