Amazon Example PPC Skill

Purpose

Replace this with the operator job. A good purpose explains the decision the skill supports, the Amazon entities it reasons about, and the business outcome it protects or improves.

Examples:

• Find search terms ready for exact-match harvesting.
• Diagnose wasted spend without cutting strategic defense traffic.
• Build an inventory-aware PPC scaling plan.

Required Inputs

Gather, derive, or mark unavailable:

• Marketplace, profile/account, currency, timezone, ASIN and campaign scope.
• Exact analysis window and comparison period.
• Relevant Amazon Ads reports and metrics.
• Retail readiness: inventory, Featured Offer / Buy Box, price, reviews, rating, delivery promise, listing status, and conversion.
• Unit economics or target thresholds when profitability matters.
• BSR, organic rank, total sales, and competitor context when rank or incrementality matters.
• Known changes: bids, budgets, placements, negatives, launches, pauses, listings, pricing, promotions, and inventory events.

Data Schema

Document the minimum useful fields for this skill. Replace this placeholder with a concise schema, for example:

• Entity fields: ASIN, campaign ID, ad group ID, keyword ID, target ID, search term, match type, placement, state.
• Performance fields: spend, sales, orders, impressions, clicks, CTR, CPC, CVR, ACoS, ROAS.
• Retail fields: inventory, Featured Offer / Buy Box, price, reviews, rating, margin, BSR, total sales.
• Context fields: date window, comparison window, marketplace, currency, timezone, recent changes.

Safety Gates

Before recommending execution, classify the case as Actionable, Directional, or Non-actionable.

Do not recommend bid, budget, placement, negative, pause, relaunch, or campaign-creation actions unless:

• The data is fresh enough for the decision.
• The entity has enough volume for the claim.
• The action is specific and reversible or clearly risk-assessed.
• Missing margin, inventory, Featured Offer / Buy Box, retail readiness, total sales, or BSR data does not materially change the recommendation.
• The recommendation includes monitoring or rollback criteria.

If connected to Rocketcart MCP or any live Amazon Ads execution layer, require explicit approval, preflight against live state, and readback after execution.

Common Failure Modes

Replace or extend these with skill-specific blockers:

• Missing entity IDs but live write is requested.
• Same-day or stale data is treated as complete.
• Margin, inventory, Featured Offer / Buy Box, total sales, or BSR is missing but the output makes firm profitability or causality claims.
• A negative, pause, or budget cut would harm brand defense, own-ASIN defense, launch/rank support, or low-sample discovery.
• Action rows are vague and omit entity, evidence, current state, proposed state, risk, approval status, or monitoring.
• Rocketcart or other live-write tooling is available but no explicit approval, preflight, readback, or monitoring plan exists.

Workflow

1. Establish scope, data coverage, freshness, and attribution caveats.
2. Segment the data by the relevant Amazon entities.
3. Apply evidence thresholds and retail-readiness gates.
4. Separate confirmed facts from hypotheses and missing data.
5. Produce action rows with confidence, risk, and approval status.
6. Define monitoring windows and success/failure criteria.

Output Format

Return these sections unless the user asks for a shorter version:

1. Data Coverage And Trust: available data, missing data, date ranges, scope, attribution caveats, and confidence.
2. Executive Summary: the main decision, top opportunity or risk, and what to do first.
3. Findings: evidence grouped by ASIN, campaign, keyword, search term, target, placement, or other relevant entity.
4. Recommended Actions: action table with entity, action, reason, expected impact, risk, confidence, timing, and approval requirement.
5. Watchlist: items with promising or concerning signals that are not action-ready.
6. Monitoring Plan: metrics, windows, success criteria, failure criteria, and next response.

Optional References

Move detailed formulas, examples, schemas, and long workflow variants into references/ and link them here only when needed.

Minimum Example And Eval

When turning this template into a production skill, add:

• examples/<skill-name>/prompt.md: realistic synthetic or anonymized prompt.
• examples/<skill-name>/input-summary.md: available fields, missing fields, assumptions, and scope.
• examples/<skill-name>/expected-output-outline.md: required sections and safety behavior.
• At least one relevant evals/ or stress-tests/ addition when the skill introduces a new safety, causality, or actionability risk.