--- name: run-a-daily-account-health-check description: >- A strict 10-15 minute read-only morning triage of a Google Ads account: verifies conversion actions are still recording (gads_list_conversion_actions), compares yesterday's spend and conversions to normal (gads_get_account_performance, gads_run_gaql_query), sweeps for new disapprovals (gads_get_disapproved_ads), scans change history for auto-applied or unknown edits, and checks bid-strategy status, flagging issues rather than fixing them. Run it every business day and daily for the first 1-2 weeks after any launch or major change. Do not use it to diagnose a flagged anomaly (investigate-performance-anomalies), to clear disapprovals (resolve-ad-disapprovals), or as a substitute for the weekly performance review, this skill detects damage, it does not optimize anything. --- # Daily Damage-Detection Triage ## Purpose The daily check exists to answer one question: did anything break since yesterday? Broken tracking, a spend blowout, a disapproved top ad, or a stealth auto-applied change each get more expensive by the hour, and none of them announce themselves. This skill is a fast, fixed-order, read-only sweep that separates "act now" from "note for the weekly review", and deliberately refuses to do analysis, because diagnosis during triage is how a 13-minute scan becomes a lost morning. ## When to run - Every business day, ideally at a consistent time each morning. - Daily for 7-14 days after launching campaigns or making structural changes (tracking, bidding, budgets). - On demand after any event that could have destabilized the account overnight. ## When NOT to run - Something is already flagged and you need the cause, that is investigate-performance-anomalies. - Disapprovals were found and need clearing, that is resolve-ad-disapprovals. - You want week-over-week trends and optimization decisions, that is the weekly review (run-a-weekly-performance-review), not this scan. - You are trying to build the automated detection layer itself, configure-account-alerts. ## Prerequisites - Read access to the account via VigilDog. - Reference values for "normal": expected daily spend per campaign, typical daily conversion volume (derivable once from 30 days of history and reused). - The flag log location from previous days, so today's entry appends to a running record. ## Procedure Hold the whole pass to roughly 15 minutes. This skill makes no Google Ads writes; anything requiring a change is flagged and routed. 1. Tracking pulse (most important two minutes). Call `gads_list_conversion_actions` and confirm every primary action's status is active and none report a recording problem. Then pull yesterday's account-level conversions via `gads_get_account_performance` and compare to the typical day: - In normal range → continue. - Low but nonzero → flag for the weekly review. - Zero or near-zero in an account that converts daily → stop the triage; this is a tracking emergency until proven otherwise. Route immediately to investigate-performance-anomalies / the conversion-tracking skill, and inform the user now. 2. Spend sanity. Query yesterday's cost per campaign with `gads_run_gaql_query`, sorted descending, against expected levels: - Within about ±20% of expected → normal. - Above ~120% of expected → check whether a budget change explains it (step 5 will show it); flag if not. - Zero spend on an ENABLED campaign → flag urgent; check status, bid strategy, and ad/keyword eligibility during follow-up, not now. Also glance at month-to-date pacing; note if the account is running >10% hot against the monthly plan. 3. Budget ceilings. Same GAQL pass: read budget-lost impression share and flag campaigns that are newly budget-capped, a priority campaign hitting its ceiling is a quiet volume leak. 4. Policy sweep. Run `gads_get_disapproved_ads`. New disapprovals since yesterday: - On high-impression ads → same-day action; route to resolve-ad-disapprovals. - On minor ads → flag for the weekly batch. Include asset-level status in the sweep (sitelinks/images), which is where quiet losses hide. 5. Change audit. Pull yesterday's entries from `gads_get_change_history`: - Only known, intended changes → fine. - Auto-applied recommendation changes → flag; review whether they conflict with strategy and revert through the owning skill if so. - Changes from an unrecognized user or integration → flag urgent and tell the user today. 6. Delivery anomaly scan. Compare yesterday's impressions and clicks per campaign against the prior day and the same weekday last week via `gads_run_gaql_query`; deviations beyond ~40% get flagged (not investigated). 7. Bidding status. Call `gads_list_bidding_strategies` and confirm each strategy reports a healthy/serving state or an expected learning phase. Anything limited or misconfigured is a same-day flag. 8. Opportunistic extras (only if under time): skim `gads_list_recommendations` for repair-type items that indicate breakage (not for optimization ideas, those belong to manage-google-recommendations). 9. HUMAN STEP (outside VigilDog): click through 2-3 top ads to confirm landing pages load and match the ad's message. VigilDog cannot render the pages; if the account has many URLs, the user should run an automated URL monitor instead of manual spot checks. 10. Log and route. Append a dated entry: tracking OK/broken, spend OK/flags, disapprovals count, anomalies flagged, changes found, actions taken. Same-day items: broken tracking, high-volume disapproval, unexplained spend spike, misconfigured bid strategy. Everything else rides to the weekly review. ## Decision rules - Hard time cap ~15 minutes. Consistently exceeding it means detection should move into configure-account-alerts, not that triage should get longer. - Zero conversions overnight in a converting account is always a stop-everything event. - Spend tolerance band ±20%; breach without a matching change-history entry = flag. - Delivery deviation threshold ~40% day-over-day or vs. same weekday last week. - Flag versus fix: only the four same-day categories in step 10 get acted on today; everything else is written down and left alone. - Compare weekday to same weekday, not to yesterday alone, weekly seasonality fakes anomalies. - After a launch or structural change, keep the daily cadence for 7-14 days even if the first days look clean; instability typically shows once learning phases settle. ## Common failure modes - Triage becoming analysis: one interesting flag swallows the morning. Log it, route it, move on. - Skipping days because "nothing ever happens", the cost of the miss is precisely that the rare breakage runs for days. Consistency is the entire value. - Never opening change history, and therefore never noticing auto-applied recommendations quietly reshaping the account. - Watching only ad-level policy status while an asset disapproval bleeds reach. - Treating flags as fixed because they were written down; the weekly review must consume the flag log, or the log is theater. ## Related skills - configure-account-alerts, build the automated layer that shortens this scan. - investigate-performance-anomalies, where flagged anomalies get diagnosed. - resolve-ad-disapprovals, where flagged disapprovals get cleared. - run-a-weekly-performance-review, consumes this skill's flag log for real analysis. - manage-google-recommendations, the monthly, strategic pass over recommendations.