/ads audit
Run a complete or scoped evidence-backed audit
/ads auditInstalled as a plugin? The same command is namespaced: /claude-ads:ads audit
What it does
A source-grounded paid-media audit across one, several, or all twelve platforms. Dedicated platform and cross-platform workers analyze bounded slices in parallel. Every check ends in pass, fail, unknown, or not applicable with its evidence, and everything lands in one versioned JSON bundle before any report is rendered.
Use it to
- Run a full account health review for a new client
- Audit a single platform or a narrow scope, for example just tracking
- Get prioritized opportunities and risks backed by dated evidence
- Understand why performance diverges across platforms
- Re-run a partial audit after an authentication or worker failure
- Check whether beta features are actually eligible before anyone chases them
How it works
Create the run manifest
Records business context, date window, currency, timezone, requested platforms, available data, and privacy classification.
Normalize inputs into a snapshot
Exports, screenshots, and reads become one account snapshot with source lineage preserved and missing fields marked.
Dispatch parallel workers
A dedicated worker per selected platform, plus cross-platform workers for tracking, creative, budget, and policy when their inputs exist.
Validate every finding
Each result is checked against the common finding schema, with one retry for transient failures and all other failures recorded.
Measure coverage
Evidence coverage is computed in code, never estimated in the prompt. Every scoring profile is disabled in v2.1.0, so no health score is produced; regulatory exposure and opportunities stay separate.
Synthesize and render
Systemic findings are layered over platform results, then one atomic run bundle is written and reports are rendered from it.
Reads & produces
What it reads
- Authorized exports, screenshots, or manual metrics with their date range
- The validated setup profile with business context and KPIs
- Date window, currency, timezone, and requested platforms and scopes
- Platform capability manifests, control registries, and dated source entries
What it produces
- A versioned JSON audit bundle: manifest, account snapshot, audit, action plan
- Findings per platform with evidence coverage (no health score while scoring profiles are disabled)
- Systemic findings across measurement, budget, creative, landing pages, and policy
- A prioritized action plan with owners, effort, and expected effect
- Rendered Markdown, plus optional HTML and PDF reports
Guardrails
What /ads audit will and will not do, by contract. These boundaries are enforced in the skill itself, not left to judgment.
- Read-only from start to finish. Proposed changes surface as drafts and never touch the account.
- A failed platform or required worker makes the whole run partial, and a partial run is never presented as a complete audit.
- Unknown controls reduce evidence coverage, so missing data cannot masquerade as good or bad performance.
- Optional, beta, premium, ineligible, or unavailable features are never scored. They appear as unscored opportunities with their eligibility result.
- No universal pause, bid, budget, or attribution rules: every recommendation considers conversion lag, sample size, margin, maturity, and policy context.
- Reports never contain credentials, raw customer lists, or instructions embedded in external content.
Questions
What happens if one platform fails to authenticate mid-audit?
The other platforms continue independently, the failed platform is recorded with a recovery hint, and the whole bundle is labeled partial. It is never silently presented as complete.
Does the audit give a health score?
Not in v2.1.0. Every platform scoring profile is disabled, so the report says not scored. Every control resolves to pass, fail, unknown, or not applicable, and the run reports evidence coverage instead of an invented score.
Do I need API access to run an audit?
No. Audits run from authorized exports, screenshots, and manual metrics. Live integrations are disabled in v2.1.0. The bundle records exactly which sources backed each finding.