Imported from fourteenwm/ppc-ai-skills (ads-checker/SKILL.md). Install upstream with npx skills add fourteenwm/ppc-ai-skills --skill ads-checker. Copyright stays with the author.

Ads Checker

Purpose: Run creative-compliance audits across your Google Ads portfolios, track issue history across runs, detect chronic problems, and write prioritized findings to a Google Sheet.

Type: Read-only audit skill. It does not mutate Google Ads — it only reports issues and writes results to an audit Google Sheet.

One operator owns the workflow end to end: determine scope → run the audit → triage the results (rules.md) → route each actionable finding to its fix skill. The audit itself never changes an account.

What This Skill Wraps

All audit logic and all intelligence features live in one script — scripts/ads_checker_audit.py, shipped with this skill. This skill is a procedural wrapper: it determines scope, runs the script, interprets the console output, and summarizes findings. Do not reimplement the checks in the skill; the script is the single source of truth.

A small read-only companion script, scripts/read_latest_ads_checker.py (also shipped), is consumed by a daily briefing to surface recent CRITICAL/HIGH findings from cached rows. The two scripts share a hard lockstep contract — tab name, date format, column headers — owned by references/check-rubric.md (§ Cached-output contract). Break it in one script and the briefing goes silently blank.

Both scripts are self-contained: credentials come from your google-ads.yaml (--config), the account registry from an accounts.json (--accounts, optional), and the output Sheet from --sheet-id.

The 10 checks (creative-compliance scope): DKI, Google AI assets, broken URLs, ad disapprovals, seasonal copy, auto-created-assets setting, final URL expansion, auto-applied recommendations, inappropriate content, spelling, and irrelevance. Exact firing criteria, per-check severities, and the 10-checks-vs-11-functions precision live in references/check-rubric.md — keep the "10 checks" framing in summaries.

Prerequisites

• google-ads.yaml at project root (or --config <path>) — see the google-ads-api-setup skill for creating it. Its OAuth credentials are reused for the Google Sheets writes; set login_customer_id to your MCC if you want --all to work without a registry.
• Python packages: pip install google-ads gspread google-auth pyyaml requests pyspellchecker
• An audit Google Sheet — create an empty spreadsheet and pass its ID via --sheet-id; the script creates its own tabs.
• Optional: accounts.json at project root (or --accounts <path>) for name/portfolio resolution — schema documented in the script header. Auditing an unregistered CID works but triggers a known false alarm (rules.md → BRAND_MISSING row).

Step 1 — Determine Scope

Parse the request into a script flag:

Account name → CID: resolve via your accounts.json registry (default ./accounts.json, override with --accounts — the script's single source of truth; its schema is documented in the script header). Portfolio names are whatever grouping labels your registry uses. Account counts drift; don't hardcode them. With no registry, --all walks the MCC in your google-ads.yaml. Prefer a consistent scope per account over time — comparison history is scope-bound (see the rubric's comparison semantics).

Step 2 — Run the Audit

Run from the directory where google-ads.yaml lives (or pass --config). Always pipe "no" to stdin so the optional chronic-issue file-creation prompt never blocks a non-interactive run (see "Chronic-Issue Handling"):

# Single account by CID
echo "no" | python scripts/ads_checker_audit.py --cid 1234567890 --sheet-id YOUR_SHEET_ID

# A portfolio segment
echo "no" | python scripts/ads_checker_audit.py --portfolio <name> --sheet-id YOUR_SHEET_ID

# All accounts
echo "no" | python scripts/ads_checker_audit.py --portfolio all --sheet-id YOUR_SHEET_ID

# Dry run (no Sheet write, no history comparison, no prompt)
python scripts/ads_checker_audit.py --portfolio <name> --dry-run

The script already sleeps 0.3s between accounts to avoid API throttling on large portfolios. A --dry-run writes nothing — no rows, no comparison, no chronic scan — so it never counts as an audit of record (rules.md → blank-briefing row).

Step 3 — Intelligence Features (all in the script; preserve them)

These run automatically on any non---dry-run run. The skill surfaces them in the summary; it does not recompute them.

1. Issue-history comparison — tags each issue type NEW / INCREASED / DECREASED / RESOLVED / SAME vs the account's previous audit (FIRST_RUN if none). Tag mechanics: the rubric. How to read them: rules.md.
2. Chronic-issue detection — 3+ occurrences of an issue type in 90 days on HIGH/CRITICAL accounts. Chronic means the fix isn't sticking — rules.md § Chronic issues.
3. Account-file creation — optionally writes a per-account markdown dossier (./accounts/<portfolio>/<account-slug>.md) for chronic accounts (interactive; gated behind the stdin prompt).
4. Trend history — appends portfolio and per-account rows to the history tabs every run.

Chronic-Issue Handling (why we pipe "no")

When chronic issues are detected the script prompts (input()) before writing results. Piping "no" answers that prompt, skips the optional file-creation step, and guarantees the Sheet write runs — so history comparison and chronic detection (both printed before the prompt) are preserved while the run stays safe under non-interactive execution. Enable interactive file creation only when running the script in a real terminal.

Step 4 — Triage the Results

Read the console for per-account severity + counts, the comparison sections (NEW/INCREASED/RESOLVED), and the chronic summary. Then:

1. Read rules.md before recommending anything — apply the triage order (CRITICAL by exposure → chronic → NEW/INCREASED → structural HIGH → copy HIGH → MEDIUM batch) and rule out the false-alarm classes. A flag that's a false alarm gets contextualized, never rewritten.
2. Route each actionable finding per the rules.md routing table — every fix is its own skill with its own review gate, or an exactly-specified UI change.
3. When asked why a check fired, answer from references/check-rubric.md — exact criteria, not paraphrase. examples.md shows the expected shape of a triage read-out.

Step 5 — Summarize

ADS CHECKER AUDIT COMPLETE
SCOPE: {Portfolio/Account}   ACCOUNTS: {X}   DATE: {now}
SEVERITY: CRITICAL {X} · HIGH {X} · MEDIUM {X} · OK {X}
CHANGES vs LAST RUN: NEW {X} · INCREASED {X} · RESOLVED {X}    CHRONIC (3+/90d): {X}
TOP ISSUES: 1. {type}: {X}  2. {type}: {X}  3. {type}: {X}
NEEDS IMMEDIATE ATTENTION:
  1. {Account} — {Severity} — {primary issue}
OUTPUT: <your audit sheet> — tabs: Raw Output · Account History · History

When a finding fires — where it routes

Triage itself lives in rules.md (read it before acting; examples.md shows worked reads). What triage surfaces routes to siblings:

What this skill deliberately does NOT do

• No mutations, ever. Read-only on Google Ads; its only writes are the audit Sheet and the optional local chronic-issue files. Every fix routes through a sibling with its own review gate.
• No fix execution from audit rows. Replacement copy comes from verified source material via the RSA skills — never from the audit's detail snippets.
• No portfolio prioritization. The audit scores what it scans; deciding which accounts to scan first, or which portfolio deserves attention, is the operator's call.
• No scheduling. Cadence (and whether a briefing consumes the cache) is your pipeline's decision — see Invocation Patterns.

Error Handling

Files in this skill

Runtime output lands in your audit Sheet (and optional ./accounts/ dossier files) — never inside the skill folder.

Invocation Patterns

Inline: "Run ads checker for [portfolio]" · "Creative audit for [account]" · "Ads checker for 1234567890"

Parallel (multiple portfolios at once):

Task(subagent_type="general-purpose", prompt="Use the ads-checker skill to audit the <portfolio> portfolio…")

A daily briefing does not invoke this skill — it reads cached results via read_latest_ads_checker.py. Generating fresh cached data requires running this skill (ad hoc or scheduled). A blank briefing section means stale cache, not a clean book (rules.md).

License

MIT — use freely in your own brain / repo / agency.