Rocketcart Amazon Ads Live Review

Rocketcart Amazon Ads Live Review

Purpose

Initial review is read-only.

Review Amazon Sponsored Products accounts with a read-first, approval-gated operating model. Use static exports when Rocketcart MCP is unavailable. When Rocketcart MCP capabilities are available, use them as the Amazon Ads + product-intelligence connection: inspect live account state, campaign settings, product ads/ASIN mapping, product readiness, category rank/BSR movement, competitor signals, budget changes, live changes since optimization snapshots, and previous snapshots/changelogs before proposing actions.

MCP means Model Context Protocol: a structured way for an assistant to use approved external context and action capabilities. See the repo glossary for definitions of MCP, preflight, readback, action gates, and exact entity IDs.

This skill is the bridge between open-source Amazon PPC reasoning and Rocketcart's live product-aware account layer. It should produce sharper recommendations because it can compare proposed actions against current Ads state, product context, recent-change context, and prior snapshots, but it must not execute writes by default.

Connection Modes

Standalone Mode

Use this mode when the user provides pasted data, CSVs, exports, screenshots, or summaries.

• State that Rocketcart MCP was not used.
• Build the best possible read-only review from provided campaign, search-term, targeting, placement, budget, BSR, retail-readiness, and economics data.
• Lower confidence when live state, current entity IDs, snapshots, or recent changes are unavailable.
• Do not present write actions as executable without live preflight.

Rocketcart MCP Mode

Use this mode when Rocketcart MCP capabilities are available or the user asks for a Rocketcart live review.

• If profile is missing, use Rocketcart profile discovery first. If there is exactly one profile, use it; if there are multiple plausible profiles, ask the user which one to review.
• Inspect SP campaigns with the available live campaign-read capability, preferring enabled campaigns unless the user asks for paused/archived context.
• Inspect product ads and ASIN/SKU mapping with the available product-ad read capability when ASIN context matters.
• Inspect product intelligence, report freshness, and quality where available, including ASIN-level controls, category rank/BSR movement, product-vs-ads context, seasonality, and competitor signals.
• Detect recent budget changes and live changes since the latest optimization snapshot.
• Review targeting, negative, and entity changes when relevant.
• Review optimization snapshots, changelogs, and recent action history.
• When a write action would need live routing context, use that context only for the approval/preflight plan; do not use it as permission to execute.
• Read rocketcart-mcp-capability-map.md when you need Rocketcart capability categories, write surfaces, and preflight/readback expectations.

Review Modes

Choose one primary review mode. If the user's request is ambiguous, default to Live Optimization Review and note any secondary mode that would improve confidence.

Mode Guidance

• Use Live Optimization Review for the first Rocketcart pass unless the user explicitly asks for growth, preflight, or post-change monitoring.
• Use Product-Aware Growth Review when product context can change the PPC decision: inventory, availability, Featured Offer / Buy Box, price, rating/reviews, category rank/BSR movement, estimated demand, competitor signals, margin, or recent product events.
• Use Preflight / Approval Readiness Review only after candidate rows exist. It does not imply execution.
• Use Post-Change Readback / Monitoring Review only after the user identifies approved changes, expected outcomes, or an execution window.

Required Inputs

Gather, derive, or mark unavailable:

• Marketplace, profile, account, currency, timezone, and Sponsored Products scope.
• Business goal: profit, revenue growth, BSR/rank support, launch, defense, waste reduction, or balanced growth.
• Current and comparison windows, preferably T-1 anchored when same-day data may be incomplete.
• Campaign state, budget, bidding strategy, placement modifiers, targeting type, and recent changes.
• Performance metrics: spend, sales, orders, ACoS, ROAS, impressions, clicks, CTR, CPC, CVR, budget usage, and placement performance when available.
• Search-term, targeting, advertised-product, purchased-product, product-ad, and placement data when available.
• Product intelligence and retail readiness: ASIN/SKU mapping, margin or target ACoS/CPA, inventory or availability, Featured Offer / Buy Box, price, reviews, rating, delivery promise, listing status, category rank/BSR movement, estimated demand, BSR responsiveness, competitor signals, total sales, and TACoS when available.
• Previous optimization snapshots, changelogs, live drift, and user-approved changes.

Product-Aware Classifications

Use these labels for ASINs, campaigns, or action rows when product context matters:

• Grow: Ads performance and product readiness both support controlled scale. Inventory/availability, Featured Offer / Buy Box, price, rating/reviews, margin, category/BSR context, and competitor signals do not block the action.
• Fix Before Scaling: There is upside, but scale should wait for a product or retail-readiness fix such as weak reviews/rating, listing issue, price disadvantage, margin uncertainty, conversion issue, or incomplete ASIN-level control.
• Protect: The entity supports brand defense, own-ASIN defense, rank/launch support, revenue stability, or an active winner. Avoid reductions, pauses, or negatives unless waste is isolated and the strategic role is protected.
• Monitor: Evidence is promising or concerning but not action-ready. Use when sample size is thin, same-day data is incomplete, BSR movement has confounders, competitor/category movement may explain results, or recent changes need more time.
• Blocked: The action is unsafe or non-executable now because of low inventory, availability risk, unstable Featured Offer / Buy Box, missing exact IDs, stale current values, missing approval, product context unavailable for a product-level decision, or a failed preflight gate.

Review Workflow

1. Establish connection mode, review mode, and scope.

   • State whether the review is Standalone or Rocketcart MCP.
   • State the review mode: Live Optimization Review, Product-Aware Growth Review, Preflight / Approval Readiness Review, or Post-Change Readback / Monitoring Review.
   • Print profile, marketplace, ad type scope, date windows, freshness, attribution caveats, approval/execution boundary, and missing data.

2. Resolve profile and read current state.

   • In Rocketcart MCP mode, list or confirm the profile, inspect campaigns, product ads/ASIN context, product intelligence, report freshness/quality, budget changes, targeting or negative drift, live drift, and snapshots/changelogs.
   • In Standalone mode, map provided fields to the closest equivalent live-state concepts.

3. Build the product-aware map.

   • Join live campaigns to product ads and ASIN/SKU context where available.
   • Inspect category rank/BSR movement, inventory or availability, Featured Offer / Buy Box, price, reviews/rating, estimated demand, competitor signals, BSR responsiveness, margin or target economics, product readiness, and recent product/account changes.
   • If product context is unavailable and could change a bid, budget, launch, pause, negative, or rank-support decision, classify the action as Needs Data, Monitor, or Blocked.

4. Identify risks before upside.

   • Flag budget cuts, bid/placement drift, out-of-budget winners, paused or changed winner campaigns, waste concentration, poor query mix, retail-readiness blockers, inventory or availability risk, Featured Offer / Buy Box risk, weak product intelligence coverage, missing margin, weak reviews/rating, competitor price/deal pressure, BSR movement explained by category/competitor changes, and low-confidence data.
   • Preserve the guardrails from amazon-ads-performance-drop-diagnosis and amazon-growth-opportunity-finder when the review overlaps with drop diagnosis or growth discovery.
   • Classify each important ASIN/campaign/action as Grow, Fix Before Scaling, Protect, Monitor, or Blocked.

5. Build proposed action rows or review candidate rows.

   • Separate read-only findings from write candidates.
   • For every candidate, specify entity type, exact entity ID when available, current state, proposed state, reason, expected impact, risk, confidence, timing, approval status, preflight checks, readback checks, and monitoring window.
   • In Preflight / Approval Readiness Review, mark each row as Approval Ready, Needs IDs, Needs Current Value, Preflight Required, Needs Product Context, Stale Approval, Blocked, or Monitor Only.
   • In Post-Change Readback / Monitoring Review, compare expected versus current state and classify each prior action as Readback Confirmed, Partially Applied, Not Applied, Monitoring, Worked, Failed, or Needs More Data.

6. Keep execution separate.

   • Do not use write capabilities during the initial review.
   • If the user asks to execute, first restate the exact approved action rows and run live preflight.
   • Execute only the approved subset, then read back state and define monitoring.

Write-Action Safety Rules

Any bid, budget, placement, negative, pause, archive, relaunch, or campaign-creation action requires all of the following:

• Explicit human approval for the exact action row.
• Live preflight against current state immediately before execution.
• Exact entity IDs, not names alone.
• Current value and proposed value.
• Expected impact and primary risk.
• Product-readiness check when the action may affect demand, BSR, stock, launch, harvesting, or product-level scale.
• Rollback or next-response guidance.
• Readback after execution.
• Monitoring window with success and failure criteria.

If any requirement is missing, classify the action as Approval Required, Preflight Required, Needs IDs, Needs Data, or Monitor Only rather than executable.

Example Action Rows

Blocked write row:

Approval-ready but not executed row:

Example approval text for the second row:

Approve campaign budget change for entity ID 1234567890 from 40 EUR/day to 55 EUR/day after live preflight confirms the current value is still 40 EUR/day, campaign is enabled, inventory and Featured Offer are safe, and readback plus 3/7/14-day monitoring are reported.

Output Format

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

1. Mode, Scope, And Data Coverage: mode, profile/account, marketplace, windows, live-read status, snapshots reviewed, missing data, and confidence.
2. Executive Verdict: 3-6 bullets covering what to protect, what to scale or fix, what changed recently, and whether execution is safe.
3. Live State And Change Review: campaigns inspected, budget changes, live drift since snapshot, snapshot/changelog context, and any reconciliation caveats.
4. Product Intelligence And Readiness: ASIN/SKU mapping, inventory or availability, Featured Offer / Buy Box, category rank/BSR movement, price, reviews/rating, estimated demand, competitor signals, margin, and missing product context.
5. Product-Aware Classification: classify material ASINs/campaigns/actions as Grow, Fix Before Scaling, Protect, Monitor, or Blocked, with the product-context reason.
6. Read-Only Findings: risks, opportunities, anomalies, and missing data that do not require execution.
7. Proposed Action Rows: table with Entity Type | Entity ID | ASIN/SKU | Name | Current State | Proposed Action | Product Context | Classification | Reason | Expected Impact | Risk | Confidence | Preflight | Approval.
8. Execution Gate: which actions are blocked, which are approval-ready, and what exact approval text or decision is needed.
9. Readback And Monitoring Plan: readback checks after execution plus 3-day, 7-day, and 14-day monitoring rules.
10. Missing Data / Next Reads: data or Rocketcart reads that would improve confidence.

Default Stance

Prefer read, diagnose, propose, preflight, approve, execute, readback, monitor in that order. The review is successful even when no writes are recommended.