Fires when one or more campaigns have a cost-per-conversion more than 25% above the account target CPA over the trailing 7 days, while at least one other campaign in the same account is converting at or below target. Both conditions must be true; the skill will not draft a shift if there is nowhere better to move the budget.

Calls get_account_snapshot for pacing, then get_performance with a 7-day window to pull cost, conversions, CPA, and ROAS per campaign and ad set. Read-only — mcp.read scope only at this stage.

Calls propose_change with type budget_reallocation. The payload contains: source campaign ID + current daily budget, destination campaign ID + current daily budget, proposed delta in account currency, and a rationale string quoting the CPA gap. The approval queue shows a two-column before → after diff.

The draft amount is sized so the net daily spend change fits inside the account's envelope ceiling. If the optimal shift would exceed the ceiling, the skill drafts the largest amount that fits and notes the cap in the rationale. A second shift cannot be drafted until 24 hours after the first approved change.

Fires when a keyword or audience bid is more than 30% above the implied max CPC derived from target CPA and measured CVR, or more than 30% below it and the entity is converting. At least three conversions in the trailing 14 days are required before the skill will touch a bid — below that volume it has no signal.

Calls query_entities for keyword and audience bids, then get_performance with a 14-day window. Derives implied max CPC as target_cpa × cvr per entity.

One propose_change call per entity, type bid_adjustment. Payload: entity ID, entity type (keyword / audience), current bid, proposed bid, implied max CPC, and conversion count that justified the signal. The queue groups these into a single reviewable set if multiple entities fire in the same cycle.

Bid changes do not directly move spend, so they are not counted against the spend delta ceiling. However, the skill is still rate-limited: it produces at most one batch of bid drafts per 24-hour window per account, regardless of how many entities qualify.

Fires when a campaign's spend-to-time ratio at the current hour implies it will over-deliver by more than 15% or under-deliver by more than 20% by midnight in the account timezone. Evaluated once per loop cycle; the skill skips campaigns with shared budgets — those require a manual review.

Calls get_account_snapshot for today's spend-so-far, then reads the campaign daily budget and the account timezone from query_entities. Computes spend rate as spend_so_far / (current_hour / 24).

Type budget_pacing_fix. Payload: campaign ID, current daily budget, proposed daily budget (the corrected value), projected end-of-day spend at current rate, projected end-of-day spend at proposed rate, and the account-local hour the draft was generated. The queue shows both projections side by side.

The corrected budget is capped so the implied remaining spend for today does not exceed the envelope ceiling. If correcting the pacing would require exceeding the ceiling, the skill drafts the largest correction that fits and adds a warning to the rationale. A correction drafted in the morning may become stale by afternoon — the queue shows the generation timestamp so you can re-evaluate before approving.

Any of the following trips the skill: spend in the last hour is more than 3× the hourly average for the trailing 7 days; CTR drops more than 40% day-over-day for a campaign with at least 500 impressions; one or more ads enter a disapproved or limited status; daily budget is more than 90% exhausted before 18:00 account-local time.

Calls get_account_snapshot for current status flags, then get_performance with a 1-day and 7-day window to compute the anomaly ratios. Ad status is read via query_entities with a status filter.

Always emits a notification event first. If a corrective draft is possible (e.g., pausing a runaway ad set), calls propose_change with type anomaly_response. Payload: anomaly type, affected entity ID, observed metric value, threshold that was crossed, and the proposed corrective field change. The queue renders the alert banner above the diff so you have context before acting.

Anomaly-response drafts count against the envelope like any other spend-affecting change. If the envelope is exhausted, the skill still raises the alert but notes it cannot produce a corrective draft until the window resets. You can approve the situation manually — only autonomous drafts are gated.

Fires when a creative has served at least 5,000 impressions in the trailing 14 days and its CTR has declined more than 30% compared to its own 7-day-prior baseline. A replacement creative must exist in the same ad set with a higher or flat CTR trend — the skill will not draft a pause if there is nothing to rotate toward.

Calls get_performance at the creative level with a 14-day window, then computes week-over-week CTR delta per creative within each ad set. Reads creative status via query_entities to confirm the replacement is active and not disapproved.

Type creative_rotation. Payload: fatigued creative ID + current status + CTR trend, replacement creative ID + CTR trend, ad set ID, and the impression and CTR values that justified the signal. The queue shows the creative names (or IDs if names are unavailable), the CTR delta, and the proposed status change side by side.

Creative pauses do not directly change spend, so they are not counted against the spend delta ceiling. Like bid changes, the skill produces at most one creative-rotation batch per 24-hour window per account. If multiple ad sets qualify, the skill prioritizes the one with the largest absolute CTR decline and defers the rest to the next cycle.