Ready-to-use AI agent skills for marketing, PPC, sales, finance, ecommerce, and business reporting. Coupler.io skills give AI assistants repeatable analytics workflows backed by live data from your Coupler.io workspace through the Coupler.io MCP server. Use them to analyze campaign performance, sales pipelines, financial metrics, ecommerce data, generate reports, investigate anomalies, and turn business data into actionable insights.
AI agent skills for marketing, sales, finance, ecommerce, PPC, and reporting analysis with live Coupler.io data.
Install only what you need. Every skill's description stays in context for the whole session, so a plugin costs tokens in proportion to how many skills it carries. See Installation for the full plugin list, team setup, and manual install.
You also need a Coupler.io account with the MCP server connected, so the skills have live data to work against.
Skills library
Skills are organized by business function - Finance, Sales, Ecommerce, and Marketing & Ads - with additional shared capabilities and utilities.
Paid-ads (PPC) performance — per-platform funnel traces, efficiency-first comparison (ranked on CPA/ROAS), budget pacing, ad-fatigue checks, and honest attribution (never blends conversions across platforms; offers an independent-source cross-check and blended CAC). Facebook + Google Ads first-class.
Google Ads deep dives
Eight Google-Ads-only skills in marketing-and-ads/google-ads/. Each one owns a single question and hands off to the others rather than duplicating them — ppc-analytics stays the cross-platform paid view. When a request needs a field no packaged report type carries, they route to capability/google-ads-custom-gaql.
The baseline read — what changed in the account and why. Period comparison, cost-per-lead moves, impression-share loss. Run this before deciding anything else.
Whether the conversion numbers can be trusted at all — double-counting, Analytics disagreement, untracked spend. Run before any cost or return conclusion.
Account configuration priced in the spend flowing through it — search partners, Display expansion, location intent, auto-apply, ad rotation, conversion windows, bid strategies.
The monthly client write-up — what to say, how to present a missed target. Built only when asked.
TikTok Ads deep dives
Eleven TikTok-only skills in marketing-and-ads/tiktok-ads/. Same shape as the Google pack — each owns a single question and routes to its siblings rather than duplicating them, with ppc-analytics still the cross-platform paid view.
The monthly client write-up — the kind that survives being questioned. Built only when asked.
Meta / Facebook Ads deep dives
Eleven Meta-only skills in marketing-and-ads/facebook-ads/. Same shape as the Google pack — each owns a single question and routes to its siblings rather than duplicating them, with ppc-analytics still the cross-platform paid view.
The baseline read — what moved this period and which campaigns moved it. Cost decomposed into CPM, CTR and conversion rate. Run this before deciding anything else.
How long the winners have left — CTR decay against first-week baseline, CPM drift, a refresh queue ranked by spend at risk, and monthly creative volume needed.
Whether the conversion numbers can be trusted — pixel and CAPI coverage, deduplication, view-through share, the platform-vs-store gap. Run before any cost or return conclusion.
Configuration priced in the spend flowing through it — optimisation goals, bid strategy, Advantage+ and expansion toggles, attribution setting. Most of these default in Meta's favour.
How the account is organised and what that costs in learning — ad sets stuck below event volume, fragmentation, audience overlap, consolidation with the spend affected.
The monthly client write-up — KPIs against goal, creative winners, the attribution caveat stated once and plainly. Built only when asked.
Microsoft Ads deep dives
Eleven Microsoft Ads (Bing) skills in marketing-and-ads/microsoft-ads/. Same shape as the Google pack, with triggers anchored to Microsoft, Bing and UET so the two packs don't collide. Adds what Microsoft carries and Google doesn't: share of voice with click share, LinkedIn profile targeting (company, industry, job function), and the negative keyword conflict report.
The monthly Microsoft Ads report a client can read — scored against agreed targets, misses explained honestly, next month's plan attached.
LinkedIn Ads deep dives
Ten LinkedIn-only skills in marketing-and-ads/linkedin-ads/. Member demographics and country/region come from single-dimension ad analytics (one member dimension per source, not crossable with campaign), and there's no hourly split, so there's no separate audience or dayparting skill. Adds a native lead gen form skill instead.
The monthly LinkedIn Ads report a client can read — scored against agreed targets, misses explained honestly, next month's plan attached.
Amazon Ads deep dives
Ten Amazon-only skills in marketing-and-ads/amazon-ads/, built on the standard Amazon Ads connector, with extra coverage from Amazon Ads (Unified) where that connector is enabled on the account. Every run declares one sales basis (ACOS or ROAS, one attribution window) and never totals marketplaces in different currencies.
The monthly Amazon Ads report a client can read — scored against agreed targets, misses explained honestly, next month's plan attached.
SEO deep dives
Seven organic-search skills in marketing-and-ads/seo/, built on Google Search Console and GA4. Same shape as the ad packs — each owns a single question and routes to its siblings rather than duplicating them.
The earning view — queries ranking just off page one, pages seen but not clicked, and pages cannibalising each other for the same query, each with the action.
Traffic from ChatGPT, Perplexity, Gemini and Claude against organic search — sessions, engagement, conversions, revenue. Measures referrals, not citations. GA4 only.
GEO deep dives
Generative engine optimization skills in marketing-and-ads/geo/, built on AI-visibility exports (Peec AI and similar) joined to GA4, Search Console and your business data.
Follows the pages ChatGPT, Perplexity and Gemini cite to AI-referred, organic, referral and direct sessions, Search Console clicks, GA4 key events and the orders, deals or payments behind them — and names the stage where the chain breaks.
Financial performance — P&L review, MRR/ARR bridge, cash runway, cost-center investigation. Works with QuickBooks, Xero, NetSuite, Stripe, Sage, and more.
E-commerce performance — funnel conversion, AOV, cohort retention, repeat purchase, anomaly detection. Works with Shopify, WooCommerce, GA4, Klaviyo, Stripe, and more.
Shopify deep dives
Four Shopify-only skills in ecommerce/shopify/. Same shape as the ads packs — each owns a single question and routes to its siblings rather than duplicating them, with ecom-analytics still the cross-platform store view. They share their definitions deliberately: store performance is the baseline the others read against, and sell-through, inventory value and dead-stock cash live in one skill only.
The baseline read — the full sales ladder, orders, AOV, refunds vs returns, new vs returning, and which of the three drivers moved revenue. Run this before deciding anything else.
Which products and variants earn their place — ranked by money and by margin on the right cost basis, with stock-out distortion flagged instead of buried.
Which campaign and which first offer bought customers who came back — cohorts cut by UTM and discount status, with open cohorts excluded from every comparison.
Capabilities
Shared building blocks. Compose these with any domain skill above.
Onboard a brand-new or empty workspace from zero to the user's own data in the chat — anchor on stated intent, route template-or-scratch, script the guided credential round-trip, build + run the first dataflow, then offer a refresh schedule.
Configure a Coupler.io dataflow end to end — pick an integration, attach a credential, wire source → destination, test it with a preview run, then offer the full run.
Build a Custom GAQL source when no packaged Google Ads report type carries the field — and name the packaged type when one does. Covers resource choice, date macros, and the raw column names and micros that come back.
Sharpen a vague or underspecified analytics request into a detailed, actionable prompt — filling in time period, metrics, data sources, and output format — before analysis.
Industry-agnostic report formatter and validator. Turns analysis output into a structured TL;DR → Metrics → Context → Recommendations → Next Questions report, then runs a Phase 2 validation pass (arithmetic, units, claim-vs-data, logical gates). Compose with a domain skill for domain-flavored reports.
Builds a live Cowork artifact — a persistent, re-openable HTML widget backed by a Coupler.io dataflow that auto-refreshes (live dashboards, daily-check pages, data explorers).
Example questions
Marketing & PPC
Why did Google Ads ROAS decline last week?
Which campaigns are wasting budget?
Which Meta Ads creatives are showing signs of fatigue?
Are we on track to spend our advertising budget this month?
Sales
What is our current pipeline coverage?
Which stage has the biggest conversion drop?
Which reps have the shortest sales cycles?
Finance
What changed in our P&L this month?
How many months of cash runway do we have?
Explain the movement in MRR versus last month.
Ecommerce
Which products contributed most to revenue growth?
How has repeat purchase rate changed?
Which customer cohorts have the strongest retention?
How Coupler.io skills work
A skill is a reusable set of instructions that tells an AI agent how to perform a specific analytics task.
The skill uses Coupler.io MCP tools to discover datasets, inspect schemas, query live business data, and return a structured analysis.
Skills define how to analyze your data.
Coupler.io MCP provides access to the data and tools needed to perform the analysis.
Skills work in all popular AI tools, such as Claude, ChatGPT, and Cursor. Any AI assistant connected to the Coupler.io MCP server can load them, even without installing the plugins.
Installation
From the marketplace — add this repo as a plugin marketplace, then install the per-cluster plugin(s) you want:
Each business function is its own plugin, so you install only what you need. Available plugins:
Plugin
Skills it installs
coupler-finance
finance-analytics
coupler-sales
sales-analytics
coupler-ecommerce
ecom-analytics
coupler-marketing-ads
marketing-analytics, ppc-analytics, + the 8 Google Ads, 11 Meta / Facebook Ads, 11 TikTok Ads, 11 Microsoft Ads, 10 LinkedIn Ads and 10 Amazon Ads deep dives, + the 7 SEO skills and 1 GEO skill
Plugin skills are namespaced by their plugin, so they can be invoked explicitly as /coupler-finance:finance-analytics. Most of them are model-invoked too — Claude reaches for them when the request matches their description. If the install summary says Run /reload-plugins to activate., run that.
Install only the plugin you need. Every skill's description stays in context for the whole session, so a plugin's cost scales with how many skills it carries. coupler-marketing-ads bundles 71 and adds roughly 13k tokens to every session; the single-skill plugins add a few hundred. Check any plugin before installing it:
claude plugin details coupler-marketing-ads@coupler-io-skills
For a whole team — register the marketplace in the repository's .claude/settings.json so collaborators get it once they trust the folder:
This adds the marketplace, not the plugins. Each person still runs /plugin install <plugin>@coupler-io-skills to pick what they want. To pre-enable plugins as well, see enabledPlugins.
Manual — copy any skill folder (each contains a SKILL.md) into your project's .claude/skills/ directory. Skills installed this way are not namespaced, and Claude activates them when the request matches their description.
Coupler.io MCP tools
These are the tools the Coupler.io MCP server exposes. Skills use them automatically — you only need to know them if you are writing a new skill or calling tools directly.
Skills
Tool
Description
list-skills
Names + one-line descriptions of every available skill. Always the first call.
get-skill
Full procedure for a named skill.
Discovery
Tool
Description
list-templates
Pre-built recipes filtered by source, metric, or category. Fastest path to skip manual setup.
list-integrations
Every integration the platform supports (400+).
list-credentials
What the user is already authorized for.
list-datasets
Every dataset across every dataflow. Use for full-picture browsing.
search-datasets
Narrows datasets by name, source, or free-text. Prefer over list-datasets when you have a keyword.
Inspection
Tool
Description
get-integration
Parameters and auth shape for one integration. Required before configuring anything.
get-integration-field-options
Resolves dynamic dropdown values (e.g. Salesforce objects, Google Ads accounts). Call when get-integration flags resolve_options_with_tool: true.
get-dataflow
Current state of one dataflow — sources, destinations, schedule, last run.
get-schema
Column types and AI context for one dataset. Required before querying.
get-data
SQL against the dataset's data table. Start with SELECT * FROM data LIMIT 5 to confirm shape.
Configuration
Tool
Description
create-dataflow
Create a new empty dataflow.
create-dataflow-from-template
Create a fully configured dataflow from a template. Use when list-templates returns a match — skips manual assembly.
create-dataflow-source
Attach a source (Stripe, Postgres, …) to a dataflow.
create-dataflow-destination
Attach a destination (Sheets, BigQuery, Looker Studio, …) to a dataflow.
update-dataflow-source
Change source parameters; merges into existing config.
update-dataflow-destination
Change destination parameters.
Run & persist
Tool
Description
run-dataflow
Run a dataflow: preview samples the sources to check the setup and resolve schemas; full refreshes all data.
update-dataset
Save a markdown description on a dataset so future sessions inherit it.
Deprecated:list-dataflows — replaced by list-datasets.
Skill frontmatter
The Agent Skills spec allows only six top-level frontmatter keys — name, description, license, compatibility, metadata, allowed-tools — and a SKILL.md carrying anything else can be rejected outright. Claude Code's own frontmatter reference accepts considerably more (when_to_use, model, effort, context, agent, argument-hint, disable-model-invocation, allowed-tools, disallowed-tools, and others), but those are Claude-only extensions that other runtimes ignore. We target the portable six, so everything of ours lives under metadata:
---
name: sales-analytics
description: Use this skill when the user wants to analyze sales pipeline, review win rates, … # for the agent: what it does and when to reach for it
metadata:
short_description: Pipeline health, win rates, sales velocity and rep performance, straight from your CRM. # for humans: one line, shown in the UI
category: sales
sources:
- Salesforce
---
description is the agent's routing signal, so it is long and stuffed with trigger phrases — unreadable in a skill catalog. short_description is the human-facing counterpart: one sentence, no "use this skill when", safe to render in a card or list. It is optional; consumers should fall back to a truncated description when it is absent.
python .github/scripts/generate_skill_index.py --check enforces the length limits:
Field
Limit
Source
name
64 characters
Spec. Renaming means renaming the folder too — the two must match.
Ours. It renders in the UI, where anything longer than a sentence stops being scannable.
Above 500 characters a description only earns a warning — it still passes. There is no consensus on the right length: obra/superpowers ships descriptions of 79–234 characters, while coreyhaines31/marketingskills runs 432–1014. The warning is a nudge to check the tail is still earning its context, not a rule — agents weigh the opening sentence most heavily.
Both descriptions land in skills-index.json, which is what the API and the MCP server read.
Contributing and development
The machine-readable catalog is skills-index.json. Do not edit it by hand — it is generated from SKILL.md frontmatter. Change the skill files, then run python .github/scripts/generate_skill_index.py, or let the GitHub Action regenerate it on merge to main.