Best Claude Skills for Amazon Ads and Operations

Claude Skills for Amazon ads and operations exist as a public library: KuudoAI/skills on GitHub gives Claude, Codex and other Agent Skills clients 20 Amazon jobs, from Sponsored Products audits and budget pacing to Amazon Marketing Cloud analysis, listings, FBA stock, buyer messages, returns, profit and repricing. Eleven start without an account, and the license permits paid client work.

The questions reach me in the same three forms. Is there a Claude skill for our pay-per-click (PPC) work? Can we use it on client accounts? And does it need our Seller Central login before it does anything useful? Every answer is in the repository, in files anyone can read before installing: a catalog of 22 entries, a license notice, and one SKILL.md per job.

A Skill carries a method, and the budget pacing Skill also asks for your rules. It will not run until you confirm your monthly plan, one target return on ad spend (ROAS) rule per ad type, strategy and product line, and your product-line ceilings, because "a target rule from another product line is not a fallback." Connected through the Amazon Ads MCP, your live budgets meet your rules, and nothing changes until you approve the proposal it shows you.

Kuudo's Amazon Skills library covers 20 jobs across advertising, AMC, creative, listings, FBA stock, sales, margins and buyer messaging

As of 2026-09-29, the KuudoAI/skills catalog lists 22 Skills. Twenty do Amazon work. The other two set up the agent: using-kuudo is the router that loads with the suite and tells the agent to announce Using [skill] to [purpose]. before it starts, and connect-to-kuudo takes an agent through connecting to your Kuudo account, which you approve.

Of the 22, two carry the Certified label: budget pacing and reporting. The other 20, both setup Skills among them, are Reference. There is no Vendor Central Skill: Vendor Central work runs through the Amazon Vendor Central MCP.

So the best Claude Skills for Amazon depend on the job, and the table is the shortlist. For a PPC team the first three to install are the campaign auditor, the search-term explorer and budget pacing. For an operations team they are the Sales and Traffic analyzer, the returns monitor and the profitability calculator, and an FBA seller adds the stockout check. Each job Skill also has a page under Kuudo Skills with starter prompts and the systems it touches.

The Sponsored Products audit is read-only, and budget pacing writes only what you approve

The Sponsored Products campaign auditor is read-only by definition. Its SKILL.md says "Do not create, update, pause, archive, or otherwise mutate campaigns," and an audit counts as complete only when it "leaves the Amazon Ads account unchanged."

It resolves the advertiser and marketplace, creates a report through the Amazon Ads MCP, then measures spend concentration, budget use, delivery gaps and targeting mix. Zero-delivery campaigns come back as an investigation queue, because eligibility, targeting, bids, budgets, seasonality, launch timing and inventory can all affect delivery. The audit page lists its inputs, outputs and guardrails.

The budget pacing Skill does write, and its safety rules say how:

• A bundled allocator, scripts/allocate.py, owns the arithmetic, and the agent may not adjust its dollar outputs by hand.
• Approval binds to the proposal ID, the advertiser account, profile, marketplace, currency, run date and the displayed changes.
• Before a write, the Skill re-reads the live state and aborts on drift.
• It never represents a pause as a zero-dollar budget.
• A scheduler may deliver proposals, but it does not turn them into live writes unless you separately define and authorize an unattended execution policy.

The Skill covers daily Sponsored Products, Sponsored Brands and Sponsored Display budgets, not Amazon DSP (demand-side platform), lifetime or portfolio budgets. The pacing page lists what a proposal takes in and returns.

Amazon's own documentation, retrieved through Amazon Agent Atlas, explains the two checks the Skill makes before it proposes anything. The Understand budgets in Sponsored ads page defines a daily budget as the most you will spend "on average, each day over a calendar month," so a single day can run above it and the Skill reads the account's overdelivery setting.

The Sponsored ads budget rules overview says rules can raise budgets "for specific dates or based on performance." The allocator does not model those increases, so it stops when an active rule is present.

Eleven of the 20 Skills start from a report or your own figures before any account is connected

A Skill does not need a live connection when its SKILL.md accepts an export. The returns monitor "works from provided tabular exports without a live integration." Amazon's page on the Fulfillment by Amazon (FBA) returns report, FBA Customer Returns report (overview), says the report carries a customer comment that "explains their reasons for returning a product," which is the evidence the Skill reads.

The same page notes that refunds given without a return do not appear in that report, and the Skill's own limits match: it is not for "financial refund totals without transaction data." The returns page lists the files it takes.

The other exports work the same way:

• The Sales and Traffic analyzer reads the Sales and Traffic export. Amazon's Business Reports FAQ says the sessions and page views behind that report "include mobile app and browser traffic data" since March 29, 2022, so app and browser visits are counted together.
• The Sponsored Brands review takes "any ingest path: CSV upload, Ads API export, warehouse table."
• The per-ASIN Search Query Performance (SQP) Skill builds on per-ASIN SQP exports.
• Budget pacing proposes from bundled JSON inputs.
• Reporting generates report requests offline.

Some Skills fall back to a manual or partial mode

Two Skills degrade honestly rather than refuse. Without the Selling Partner MCP, the profitability calculator runs "in manual mode on seller-supplied fees and labels the output as such," and repricing drops to advisory mode, where floors depend "entirely on seller-supplied figures."

Creative is partial: ad specs and policy work standalone, while registering an asset or uploading a listing image needs a connection. The listing compliance gate is partial too: without a server it "can still classify and ask, but can't check gating," and it never writes a listing itself.

Which Skills need a live account

Nine need a live account. The search-term explorer is the clearest case. Amazon's Search term report for Sponsored Products shows "the search terms that customers use," but the Skill does not accept a hand-uploaded console CSV; it creates its own report through the Amazon Ads MCP so every column arrives with the field IDs its parser expects.

Account resolution, AMC and DSP launches follow, and four Skills on the Selling Partner MCP complete the nine:

• Listing edits
• The stockout check, which reads live FBA inventory and sales velocity and never changes a shipment
• FBA inbound shipments, which pauses at every fee-charging or irreversible step for your explicit choice
• Buyer messages on Seller orders, which sends only the message types an order allows, after you approve the exact preview

Separately, Amazon Agent Flow lands your Selling Partner and Amazon Ads data in a private lake on storage you own, so last month's export becomes a query.

The suite installs as a plugin in Claude Code or Codex, and the license allows paid client work

Installation follows the repository's plugin guide. Load the whole suite, which brings the router with it. The commands below use a local clone; both marketplaces also accept KuudoAI/skills straight from GitHub, and in Claude Code claude plugin install kuudo-skills@kuudo-skills-dev keeps the suite installed beyond one session:

git clone https://github.com/KuudoAI/skills.git ~/code/kuudo-skills

# Claude Code: load the suite for this session
claude --plugin-dir ~/code/kuudo-skills

# Codex: register the local marketplace, then add the plugin
codex plugin marketplace add ~/code/kuudo-skills
codex plugin add kuudo-skills@kuudo-skills-dev

For another Agent Skills client, install one portable Skill by name. A portable install brings that Skill without the suite's startup router:

npx skills add KuudoAI/skills --skill amazon-ads-budget-pacing

Choose one mode per client; installing both makes the same Skill appear twice. The packages follow the open Agent Skills specification, which is why one SKILL.md travels between clients.

The license is PolyForm Shield 1.0.0, and LICENSE.md adds a permission for customers: using the content "to operate, analyze, advertise, or manage Amazon selling or vendor accounts, whether the accounts are your own or belong to clients you serve as an agency, consultant, or service provider, is a permitted purpose." An agency can run the Skills on paid client work.

What the license forbids is repackaging the content as a competing skill library, agent plugin or software product. Anyone passing on a copy keeps each Skill's LICENSE.txt and the required notice.

A custom Skill is one SKILL.md, and the Certified label is earned by review and trigger tests

A Skill starts with a SKILL.md: a name, a description that says when to use it, then how to do the work, with the inputs it needs, the steps, and what a useful result looks like. References and scripts go alongside only where they help.

That is the minimum the Agent Skills specification requires, and it is how an agency's method becomes a file the agent can load: your team's judgment, written down. The budget pacing Skill is a good model, because its operator controls are your rules, and a missing one stops the run.

Inside the repository, npm run skill:new -- <skill-name> creates the package, a catalog entry and an evaluation file. CONTRIBUTING.md asks authors to follow the Agent Skills specification and its best-practices guide for authoring, then run npm run catalog:build and the aggregate npm test before a pull request. New work starts as Reference. If you would rather describe the job than write it, the README invites a skill request with an example of the result you want.

What earns the Certified label

Certification is a KuudoAI trust label, "not a format feature." It needs owner review, passing repository validation, and at least one positive and one negative trigger case in evals/<skill-name>/trigger-cases.json: proof the Skill activates when it should and stays quiet when it should not. Budget pacing and reporting carry it today. Third-party contributions are labelled Community and must not be represented as certified.

What happens next: install the suite, pace offline, then connect the Amazon Ads MCP for the read-only audit

Start with the jobs that cannot change anything, then build out in this order:

1. Install the suite in Claude Code or Codex, or one portable Skill in another Agent Skills client. AI clients lists where Kuudo connects.
2. Run budget pacing offline against your real plan in its JSON input, and read the proposal before any account is connected.
3. Connect the Amazon Ads MCP and run the read-only audit. "Audit my Sponsored Products campaigns" runs the campaign auditor, which states every threshold it used and leaves the account unchanged.
4. When the offline proposals match how your team paces, let budget pacing read live budgets and propose.
5. Write your first custom Skill from the correction your team makes most often.

The pattern holds across the library: the Skill carries the method, and your inputs carry the rules.

Next: how one AMC analysis becomes a Skill of its own, in Turn Amazon Marketing Cloud Workflows into Skills.