How to use the Paid Advertising Audit agent skill (ads-audit)

What an agent skill is and what ads-audit makes the agent do

An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific kind of task. When the task matches the skill's description, the agent follows its procedure instead of improvising.

The ads-audit skill makes the agent build a versioned JSON audit bundle first. It normalizes your exports, screenshots, manual metrics or authenticated reads into an account snapshot, scores platforms with a deterministic process, and then writes a report, an action plan and a measurement plan from that bundle.

• Creates a run manifest with business context, date window, currency, timezone, requested platforms and privacy classification.
• Marks missing fields rather than filling them in.
• Labels the result complete, provisional, partial or insufficient_evidence based on data coverage.
• Keeps observations, diagnoses, recommendations and proposed changes in separate layers.
• Lists features that are unavailable or ineligible as unscored opportunities, not health penalties.

Install the ads-audit skill

Download the ZIP from the skill's page on this directory and unzip it. Then choose one of the following locations, depending on whether you want the skill available everywhere or in one project.

For all your projects, place the unzipped ads-audit folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

Alternatively, if you use the Claude apps, upload the skill under Settings → Capabilities → Skills.

• Personal use: ~/.claude/skills/ads-audit/
• Project use: .claude/skills/ads-audit/ inside your project folder
• Claude apps: Settings → Capabilities → Skills

Prepare the data you will give the agent

The skill needs ad platform exports, screenshots or manual metrics. It also expects authenticated read access, but the skill file does not explain how to set that up, so plan for exports or manual metrics if you do not have it configured.

The audit also relies on an installed audit-worker-result v1 schema and platform worker and control files. These are not included in the supplied files, so the agent may only be able to produce partial results until they are available.

• Campaign, keyword, search term, creative or conversion exports for each platform.
• Business context: objective, margin, date window, currency and timezone.
• Screenshots or manual metrics where exports are unavailable.
• Tell the agent which platforms you want covered and which you do not.

Trigger the skill with a natural request

You do not need special syntax. Describe the audit you want in plain language and attach or point to your data. The skill's description covers requests such as full ad checks, account health reviews, paid-media diagnostics, spend audits and tracking audits.

State the platforms and the date window, and say what you want back. If you want the outputs saved as files, say so.

• "Run a full paid-advertising audit on my Google and Meta exports for the last 90 days, in GBP and London time."
• "Do a tracking audit of my TikTok and LinkedIn conversion data."
• "Review account health for the attached Pinterest and Snapchat metrics and list prioritized risks."
• "Audit my spend across all platforms. Amazon authentication failed, so mark it as missing and show what that does to the result."

Read the results and coverage status

The skill's main value is honesty about coverage. Check the completeness label first. A complete result means every requested platform returned valid results with normal evidence coverage. A provisional result means some evidence is thin or stale. A partial result means a required worker failed or was omitted. An insufficient_evidence result means a requested platform has less than 60% coverage.

Under a partial result, a failed platform gets no health score and its weight is removed from the portfolio score rather than counted as zero. Do not read a partial bundle as a complete account review.

• Check platform health and evidence coverage in report.md.
• Review missing data and contradictions before acting on recommendations.
• Treat proposed mutations as drafts. The skill does not execute changes in ad accounts.