Paid Ads Account Auditor
Decide whether a paid-media account is safe to launch or scale using its own evidence.
The skill explicitly limits itself to normalized user-owned evidence, writes only a permissioned v3 audit artifact, and forbids enabling campaigns, changing bids, uploading audiences, or scaling budgets without separate approval. It discloses data sources, unknown handling, persistence authorization, and fail-closed behavior. Deductions apply because ad exports may contain sensitive business data, while concrete redaction, field-minimization, and retention rules are not stated in this skill; WebFetch data-flow boundaries largely depend on external runbook material. Publisher identity is unverified, so no extra attribution credit is given.
Required context, 20-item states, unknown handling, NOT_SCORED fallback, immutable standalone runtime, and validation checkpoints are specified with generally consistent instructions. Key paths were not executed during this static review; the skill depends on several root references, scorer/validator components, and catalogs not included in the supplied material. Reproduction and abnormal-input feedback are therefore incomplete, warranting a score below the static ceiling.
Triggers, required inputs, output contract, non-fit cases, and next-skill handoffs are clear, with Chinese labeling and a user-export-based path that does not require paid APIs. Deductions apply because platform export schemas, complex portfolio boundaries, localized output expectations, and complete alternatives when WebFetch or referenced runtime files are unavailable are not defined.
The skill has structured front matter, versioning, Apache-2.0 licensing, compatibility metadata, quick start, data-source table, validation checkpoints, persistence rules, references, and repository-level security and maintenance signals. Deductions apply because installation and runtime assumptions are distributed across external references; the skill lacks standalone parameter definitions, example outputs, FAQs, a changelog, and a clearly named maintenance owner. Version governance is mostly asserted by repository context.
The paid-account audit objective is concrete, with a 20-item ROAS profile, business constraints, attribution normalization, veto gates, and directly usable SHIP/FIX/BLOCK/UNDECIDED outcomes. Deductions apply because no real audit artifact or executed result is shown, and the workflow requires substantial normalized account, outcome, attribution, and incrementality data followed by business review.
The supplied immutable runtime snapshot includes versioning and a source digest, while repository tests cover architecture, artifact validation, and fail-closed rules. Deductions apply because the material contains no real execution of this skill, independent third-party corroboration, or independently reproduced end-to-end audit; the tests primarily cover shared infrastructure rather than the skill's complete logic. Static calibration therefore limits this to 5.
- This is a source-only review; the skill, scripts, WebFetch, and tests were not executed. Do not treat the score as a validated audit result.
- Before use, provide and verify normalized first-party data, deduplicated order/lead IDs, currency, attribution window, conversion lag, business constraint, goal, and observation date; missing required evidence should remain NEEDS_INPUT/UNDECIDED.
- Root references, scorer, validator, and catalogs cited by the skill were not included in the supplied material; standalone installs must not calculate scores, emit gate verdicts, or persist artifacts.
- Ad exports may contain sensitive business information; confirm redaction, access scope, and storage boundaries before exposing them to an agent.
What does this skill do, and when should you use it?
This skill audits one paid-media account or portfolio for incremental contribution, wasted spend, and measurement integrity. It requires normalized account evidence, attribution windows, conversion lag, currency, and business constraints, then scores 20 ROAS criteria. Verified vetoes drive a SHIP, FIX, BLOCK, or UNDECIDED gate. The skill is judgment-only: it does not enable campaigns, change bids, upload audiences, or scale budgets.
Reads campaign, query, placement, audience, and change-history exports; deduplicated order or lead IDs; economics; attribution timestamps; and safety or claims evidence. It normalizes currency, windows, IDs, lag, and portfolio scope, then scores R1 through S5 with source, date, type, and confidence. It separates platform self-report from the account's own outcome truth, labels incrementality evidence as causal, proxy, or estimated as appropriate, verifies ROAS-R1/R2, ROAS-O1/O2, and ROAS-A1 vetoes, runs the typed scorer, and reports the verdict, score or coverage interval, confidence, unknown evidence, reconciliation table, prioritized fixes, owners, and rerun conditions.
- A marketing lead is preparing to launch a paid account and needs a go/no-go decision from account exports.
- A growth team wants to increase spend materially and needs to test whether reported ROAS reflects branded-demand capture or attribution inflation.
- An analytics or media team suspects duplicate conversion counting and wants to reconcile platform results against order or lead IDs.
- A brand-safety or compliance owner needs to review placements, claim approval, disclosure state, or restricted-category exposure.
- An operator has incomplete placement, attribution, or business-context data and needs a precise list of what prevents a decision.
What are this skill's strengths and limitations?
- Uses the account's own order or lead evidence as the outcome truth instead of treating platform ROAS as sufficient.
- Combines a typed 20-item profile with verified vetoes and structured SHIP/FIX/BLOCK/UNDECIDED outcomes.
- Treats missing evidence as Unknown rather than automatically converting it into failure.
- Has an explicit safety boundary against unauthorized account or budget mutations.
- Requires normalized first-party evidence, attribution windows, currency, conversion lag, and business constraints.
- Missing placement, approved-claims, or incrementality evidence can leave key controls Unknown and prevent an overall score.
- It does not design campaign structure or produce ad creative; those are assigned to campaign-architect and ad-creative-builder.
- The scorer and referenced runbook, benchmark, and catalog files must be available in the runtime; their contents are not supplied here.
How do you install this skill?
Install the repository as a collection. In Claude Code, run /plugin marketplace add aaron-he-zhu/aaron-marketing-skills, then /plugin install aaron-marketing@aaron. On a compatible Agent Skills host, run npx skills add aaron-he-zhu/aaron-marketing-skills, or clone it with git clone https://github.com/aaron-he-zhu/aaron-marketing-skills. The skill is located at ad/activate/ad-account-auditor/SKILL.md. The supplied source does not document a dedicated single-skill installation command for this auditor.
How do you use this skill?
Provide normalized exports for one paid account or portfolio and declare the target, profile, currency, attribution window, conversion lag, business constraint, goal, and observation date. Example: Audit this USD account for direct response using 7-day click, 3-day lag, and $120 CAC ceiling. Run the incremental-profit profile against the holdout and order-ID exports. If required context is missing, it should return NEEDS_INPUT/UNDECIDED; if the scorer, validator, or typed catalogs are unavailable, it should return NOT_SCORED without a gate verdict.
How does this skill compare with similar options?
Unlike campaign-architect, this skill audits account readiness and issues a gate rather than designing account or campaign structure. Unlike ad-creative-builder, it does not produce ad assets. conversion-signal-qa, attribution-reconciler, and budget-pacing-monitor cover adjacent tracking, attribution, and pacing work.