The audit skill reviews ad accounts on any of twelve platforms: Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat and X. It builds a run manifest and an account snapshot, dispatches a dedicated worker per platform, validates each result against a schema, then scores and synthesizes across measurement, budget, creative, landing pages and policy.
The bundle is labeled complete, provisional, partial or insufficient evidence, so you can see how far to trust it.
What You Get
- A run directory with manifest.json, account-snapshot.json, audit.json, action-plan.json and report.md, with optional HTML and PDF
- A completeness label of complete, provisional, partial or insufficient evidence
- Systemic findings across measurement, budget, creative, landing pages, experimentation and policy
- Recommendations with owner, priority, effort, expected effect and success measure
Common Use Cases
- A full health review of an agency client’s accounts
- An audit where one platform fails authentication
- Checking whether a beta feature should be scored or listed as an opportunity
- Producing an action plan with measurement steps
What Makes It Unique
It defines what happens when part of an audit fails. A missing platform becomes failed, its weight leaves the portfolio score, the remaining weights are renormalized and the bundle is called partial. Unavailable beta features go to an opportunity list with no health penalty.
Under The Hood
The skill is instructions with a worker contract and a JSON example. The agent creates run files on disk, reads supplied exports and screenshots, and can read ad platform accounts through connected APIs or MCP servers. It dispatches subagent workers, one per platform, then calls deterministic scoring code and renders the reports. Findings that fail schema validation are rejected, and credentials, raw customer lists and hidden instructions from external content stay out of the report. Account changes stay drafts.