Why this repository

Most AI advice for Amazon Ads sounds confident but cannot show its work. This collection is built for a different standard:

  • Operational, not generic — each Skill has a concrete workflow, input expectations, output contract, and decision gates.
  • Evidence-aware — account facts, current platform policy, training cases, and hypotheses stay explicitly separated.
  • Profit-aware — ACOS is connected to CPC, CVR, contribution margin, inventory, returns, and lifecycle goals.
  • Agent-ready — every top-level directory is a self-contained Skill with a standard SKILL.md entrypoint.
  • Human-controlled — diagnosis and drafts are the default; live changes require explicit, itemized approval.

One rule runs through every Skill: never turn a course example, benchmark, or AI guess into a live campaign setting without current account evidence.

🧭 Skill map

Skill What it helps an agent do Best for
ACOS Diagnostics Reconcile ACOS, ROAS, TACOS, CTR, CPC, CVR, AOV, placement, attribution, and profit signals; produce one attributable experiment High ACOS, wasted spend, weak conversion, unclear break-even point
Ad Architecture Work backward from sales and profit goals into campaign roles, keyword priorities, budgets, and stage plans Launch architecture, portfolio design, red-ocean categories, seasonal planning
US Apparel Lifecycle Ads Diagnose long-, short-, and seasonal-lifecycle apparel products with US-specific playbooks Fashion, underwear, swimwear, suits, accessories, lifecycle transitions
Canada Apparel Ads Combine lifecycle, seasonality, English/French discovery, margin, and inventory guardrails Amazon.ca apparel, coats, undergarments, bilingual demand
EU AMC Audiences Plan privacy-safe AMC analysis, journey measurement, audience creation, activation, and evaluation European AMC, path analysis, reach/frequency, new-to-brand, remarketing
Japan Apparel Ads Apply Japan-specific consumer, language, seasonality, Points, and lifecycle evidence Amazon.co.jp apparel, bags, underwear, swimwear, localized launches
Listing Optimizer Audit and draft titles, bullets, attributes, search terms, image briefs, A+, video, and controlled tests CTR/CVR gaps, indexing, return prevention, ad-to-listing relevance
Product Targeting Build ASIN/category targeting pools for competitor, substitute, complement, cross-sell, upsell, and defense use cases Product targeting, category targeting, detail-page traffic, ASIN defense
UK Apparel Ads Combine UK lifecycle, seasonality, sizing, returns, compliance gates, and advertising economics Amazon.co.uk apparel, Black Friday, Boxing Day, swimwear, outerwear

⚡ Quick start

Clone the collection:

git clone https://github.com/xjli360/sealeap-amazon-ad-skills.git
cd sealeap-amazon-ad-skills

Choose the Skill that matches the job, then point your agent to its SKILL.md file or copy that directory into the Skills directory supported by your agent runtime.

Example prompt:

Use sealeap-amazon-acos-diagnostics in DIAGNOSE mode.

Marketplace: US
Date range: last 30 days
Goal: determine whether the ACOS problem is driven by CPC, CVR,
traffic mix, attribution, or unit economics.

Do not change campaigns. Show missing evidence and propose exactly
one single-variable experiment for human review.

Each Skill tells the agent which reference files to load, what data is still missing, which claims are safe to make, and where human approval is mandatory.

🧠 How the Skills think

flowchart LR
    A[Business question] --> B[Lock marketplace, account, ASIN and time range]
    B --> C[Separate evidence types]
    C --> D[Reconcile metrics and economics]
    D --> E[Diagnose the bottleneck]
    E --> F[Draft one-variable experiment]
    F --> G{Human approval?}
    G -- No --> H[Read-only report]
    G -- Yes --> I[Scoped action + verification + rollback]

The shared evidence language keeps an agent honest:

Label Meaning
ACCOUNT_FACT / ACCOUNT_ACTUAL Verified data from the current authorized account and scope
CURRENT_POLICY A current platform rule checked against an authoritative source
TRAINING_CASE / COURSE_BASELINE A teaching example or historical benchmark, never a live setting by itself
HYPOTHESIS / SELLER_HYPOTHESIS A testable explanation that still needs evidence
NEEDS_DATA / HOLD A hard stop: the agent must not invent the missing fact

🛡️ Built-in operating guardrails

  • Read-only diagnosis is the default mode.
  • Marketplace, profile, seller, store, ASIN, SKU, currency, attribution window, and date range must stay explicit.
  • Campaign writes require current scope verification and itemized human approval.
  • Every change plan includes evidence, expected effect, stop condition, and rollback value.
  • One experiment changes one primary variable so the result remains attributable.
  • Listing claims must be backed by verified product facts; competitor copy and invented claims are out of bounds.
  • AMC workflows stay aggregated and privacy-safe; no user-level export or re-identification.

📦 Repository structure

sealeap-amazon-ad-skills/
├── sealeap-amazon-acos-diagnostics/
├── sealeap-amazon-ad-architecture/
├── sealeap-amazon-apparel-lifecycle-ads/
├── sealeap-amazon-ca-apparel-ads/
├── sealeap-amazon-eu-amc-audience/
├── sealeap-amazon-jp-apparel-ads/
├── sealeap-amazon-listing-optimizer/
├── sealeap-amazon-product-targeting/
└── sealeap-amazon-uk-apparel-ads/

A Skill may include:

SKILL.md              # Agent entrypoint and operating workflow
agents/openai.yaml    # Optional agent-facing metadata
references/           # Evidence model, playbooks, examples, output contracts
scripts/              # Deterministic checks and analysis helpers
transcripts/          # Source-linked learning material where included

🌊 Built by SeaLeap

SeaLeap turns e-commerce operating knowledge into reusable, auditable Agent Skills. The goal is simple: help agents and operators move faster without losing evidence, accountability, or control.

🌐 Website: sealeap.cn

If you are building an Amazon Ads agent, an internal operating copilot, or a repeatable advertising workflow, use these Skills as composable building blocks—not as a substitute for current account data or professional judgment.

⭐ Star & contribute

If this repository saves one wasted budget cycle, one unsupported claim, or one irreversible campaign change, please give it a star. It helps more operators and Agent builders discover the project.

Useful contributions include:

  • marketplace-specific policy refreshes with authoritative citations;
  • anonymized test cases and reproducible metric checks;
  • safer output contracts, approval gates, and rollback patterns;
  • new marketplace or category Skills that preserve the same evidence standard;
  • fixes for broken links, ambiguous terms, or stale platform assumptions.

Open an issue before proposing any workflow that writes to a live advertising account.

Trademark, source, and affiliation notice

This is a SeaLeap-maintained repository for education, research, and agent workflow design. The Amazon Advertising logo is used with permission and displayed in its approved, unmodified form. That logo-use approval does not by itself state that Amazon authored, sponsored, jointly published, or maintains this repository unless a separate written authorization expressly says so. Amazon, Amazon Ads, Amazon Advertising, and related marks are trademarks of Amazon.com, Inc. or its affiliates; their use remains subject to the current Amazon Advertising marketing guidelines.

References to Amazon products and services are descriptive. Platform capabilities, eligibility, attribution, interfaces, and policies can change; verify them against current official documentation before acting. Source transcripts and course-derived notes, where present, remain subject to the rights of their respective owners and are included for traceability rather than as a transfer of ownership.