How to use the Attribution Audit agent skill (ads-attribution)

What is an agent skill, and what does this one do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. When the skill is installed and your request matches its description, the agent follows its steps.

The ads-attribution skill makes the agent review how conversions are counted across Meta, Google, GA4, mobile measurement partners such as AppsFlyer, Adjust, Branch or Singular, and server or offline sources. It checks whether the sources share the same conversion definitions, attribution windows, and deduplication rules, and explains the differences it finds.

• Declares the business conversion, value, data window, timezone, currency and decision the analysis must support
• Inventories each browser, server, platform, analytics, MMP, offline and app attribution source with its counting and deduplication rules
• Reconciles comparable events and explains gaps caused by eligibility, view-through rules, consent, modeled data, conversion lag, thresholds or scope
• Separates measurement quality from platform-reported performance
• Returns findings, contradictions, confidence levels, missing evidence and a measurement improvement plan

Install the ads-attribution skill

Download the ZIP for skills/ads-attribution from the directory page. Unzip it into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that project.

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

The skill's package contains a single file, SKILL.md. It is licensed under MIT.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check what you need before you start

The skill depends on the companion main 'ads' skill and its common JSON contract, which SKILL.md references. Without that contract the agent has no shared output format to follow, so make sure it is installed alongside this one.

You also need the data to analyse. The skill works from normalized account snapshots or exported platform reports. It does not fetch data itself: the package declares no scripts and no network use, so you supply the exports.

• The companion main 'ads' skill and its common JSON contract
• Normalized account snapshots or exported platform reports (for example, Meta, Google, GA4 or MMP exports)
• Enough context about the conversion, value, window and currency you care about

How to trigger the skill with natural requests

You do not need special syntax. Describe the problem in ordinary words that match the skill's scope. Requests about attribution audits, attribution models, conversion windows, adding or totalling Meta and Google conversions, GA4 attribution, MMP review, and cross-platform discrepancies are the ones it is designed for.

Name the platforms and attach your exports so the agent has evidence to compare.

• Why do my Meta and Google conversions for last month not match?
• Review our AppsFlyer export against the Meta app install numbers and explain the gap.
• Can you total Meta and Google conversions for this campaign?
• Run an attribution audit on these GA4 and Google Ads exports before we change budgets.

Understand the comparability rule

The skill will not add incompatible figures together. It rejects aggregation until the sources share, or are explicitly normalized to, the same conversion event and value definition, attribution window, click or view scope, counting method, deduplication identity, timezone, currency, attribution model and modeled-data treatment.

Until then, the agent reports the values side by side with their definitions. For example, Meta seven-day conversions and Google thirty-day conversions are incompatible, so the agent reconciles windows and definitions first and only totals a newly comparable dataset.

• Expect side-by-side tables when sources differ, not a single combined total
• A combined total appears only after the definitions are aligned

Read the output and act on it carefully

The output includes findings, contradictions, confidence levels, missing evidence and a measurement improvement plan. Pay attention to the missing-evidence section, since it tells you which exports to add for a stronger conclusion.

The skill does not choose an attribution model for you. It asks for your business decision context, and it assesses only the measurement evidence you provide. Treat its output as an input to your decision, not a verdict.

• Check confidence levels before changing budgets or settings
• Use the improvement plan as a checklist for tracking fixes
• Do not treat any one platform as ground truth; the skill does not assume it

Safety notes and tips

The skill runs no scripts and uses no network access, according to its package description. Your data stays in the conversation you choose to have with the agent, and you control which exports you share.

The skill does not replace legal or privacy review of consent and tracking setups. It only assesses the measurement evidence it is given.

• Share only the exports you are comfortable giving the agent; remove personal data where possible
• Include the timezone, currency and conversion windows in your exports so comparisons are possible
• Run the audit before any budget or attribution setting change, not after
• Ask for missing evidence explicitly if the first answer is inconclusive