Agent Skills for Advertising
Ready-to-run skills for Claude, ChatGPT, and any AI agent — it does the ad work, you approve every dollar.
The agent skill library
48 of 48 skills
Set up a clean A/B test (one variable, proper split, a decision rule and duration for significance), read it honestly, and call the winner only when the data supports it.
Clarifies copy editing vs copy writing up front: this is a true copywriting job, not a copy-edit pass. Produces platform-correct ad copy variations (headlines, primary text, CTAs) for Meta, Google RSA, and LinkedIn, grounded in a real offer and audience, with angle rationale — including Chinese writing to copy and paste or Japanese writing copy and paste for non-English markets.
Systematically mine a brand's Meta Ads Library — creative angles, hooks, formats, longevity (long-running = winning) — to search competitor patterns and inform your own creative.
Inventory what the client actually has, audit tracking before launching, and reach first campaigns with a reporting cadence agreed.
Launch a real product ads Amazon campaign: Sponsored Products eligibility, keyword/ASIN targeting, bids, and negatives built and verified before any spend goes live on your amazon ads campaign.
Follow the clues behind a sudden campaign change, explain what is happening in plain language, and route the next step to the right person.
Stand up an app-growth stack: UAC + Meta app campaigns, install + in-app event tracking, cohort/LTV read, and a scaling plan — end to end, gated.
Build a cold-to-hot audience structure sized to the account's real event volume, with overlap and exclusions handled deliberately.
Produce a client-ready monthly report: performance vs goals, what changed and why, wins and learnings, next month's plan — clear, honest, and defensible.
Choose a practical Codex skill pack for paid media and run it with the right context, connectors, and approval boundaries.
Read retention by acquisition cohort against what each cohort cost, and say plainly when it is too early to judge.
Research a competitor's active advertising: ad-library angles, landing pages (Firecrawl if connected), traffic mix (Similarweb) — synthesize what they're doing and where the gaps are. Facts, not verdicts.
Calculate CPC, compare traffic costs fairly, and explain why a cheaper click is not automatically a better business result.
Turn positioning and account history into a brief with distinct angles, real hooks, and the platform limits that stop work bouncing at review.
Watch frequency and CTR/CPA decay per ad; flag creatives that have fatigued and should be refreshed, with the evidence. Read-only, alert-only.
Read the placements a campaign really serves, then compose the master into each required ratio with safe areas and file limits checked.
Calculate CTR, compare engagement on a consistent basis, and explain what the percentage does and does not prove.
Extract real brand tokens from Figma and build ad templates at the sizes campaigns actually serve, with rules for what may change.
Audit Google Ads with evidence, rank the risks that matter, and prepare a safe remediation plan before account changes.
Check the conversion event and asset coverage first, then build an App campaign paused with bidding matched to real event volume.
Launch google sem ads campaigns the disciplined way: tightly-themed ad groups, keyword match-type strategy, RSAs, a seed negative list, conversion tracking verified, launched paused and diffed against the plan before spend. Uses Ahrefs for volumes if connected.
Prove the feed is healthy, structure Shopping around margin rather than a flat target, and build it paused for review.
Group keywords by what the searcher intends, match types to that intent, and build the negative list from real search-term history.
Mirror the live ad's promise on a dedicated page, wire tracking, and verify it works before any spend points at it.
Route leads into the CRM without losing their origin, then judge campaigns on qualified leads instead of form fills.
Size the firmographic audience before building, set honest cost expectations, and launch a lead form campaign paused.
Trace B2B leads from ad to deal stage, then rank campaigns on pipeline created rather than forms completed.
Sync a media plan into a tracker (Airtable if connected): campaigns, budgets, flight dates, status — one source of truth the team can see.
Verify the WhatsApp account link and number before building, then launch a messaging campaign paused for approval.
Turn a brief and assets into a complete Meta prospecting to retargeting to conversion structure — built paused, pixel/CAPI verified, checked against the approved plan, launched only on explicit go. Optional Canva resize leg.
Mine the search-terms report for wasted spend and recommend the correct negative keyword match type — exact, phrase, or broad — for clean Google Ads negative keyword hygiene, the single highest-ROI Search action. Changes proposed, applied on approval.
Build a Performance Max campaign paused, check it against the live account, and hand the launch decision back to the user.
Separate the three Quality Score components, rank the problems by spend at stake, and fix the causes in the order that pays.
Calculate ROAS, compare scenarios, and explain what the result can—and cannot—say about campaign profitability.
Keep catalog, campaigns and profitability in one loop, so spend follows margin rather than platform-reported return.
From one concept, produce a batch of on-brand static ads via Canva — headline/visual/CTA variations sized to platform placements, with a creative-testing structure in mind.
When taking over an existing account — a new client handoff, an internal transfer, or an account you have just acquired — produce a full evidence-backed audit: structure, tracking health, wasted spend, quick wins, and risk, across Google Ads, Meta, and TikTok, before you spend another dollar.
Account for the gap between platform, analytics and backend numbers, and say which source answers which question.
Judge whether each campaign's bid strategy suits its real conversion volume, and separate a stale target from a structural mismatch.
Find capped winners and weak spenders across a campaign set, and propose moves sized to avoid resetting learning.
Recommend a channel split justified by each platform's mechanics for the stated objective, with honest reasoning and named uncertainty.
Every morning, compare spend vs plan across every connected account and raise honest, evidence-backed alerts only when something needs a human. Read-only; hands off to the reallocator for changes.
From a brief (product, audience, budget, goal, market), produce a defensible media plan: audience definitions, budget split across channels, flighting calendar, and a channel-mix rationale. Cross-platform. Uses Similarweb/Semrush for competitive context if connected. Plan doc, not a live launch.
Report a quarter against the goals that were actually set, separate what worked from what merely happened, and propose next quarter honestly.
Tier retargeting audiences against real list sizes, sequence the message per tier, and suppress converters deliberately.
Assemble a TikTok campaign against the platform's real setup rules, build it paused, and show every deviation from the plan before launch.
Produce UGC-style video ads via Pika + HeyGen — script, avatar/voice, cut to Meta video ads specs and TikTok/Reels dimensions — that feel native, not polished-ad.
Map the business outcome to an eligible YouTube format, check the conversion signal, and build the campaign paused.
What Claude skills are, and how these work
A skill is a folder of instructions an AI agent loads when it recognises the job you are describing — the open Agent Skills specification that Claude, Claude Code, ChatGPT, Cursor, Codex, Gemini, and Windsurf all read. Unlike a prompt you paste, a skill travels with its own procedure, safety rules, and stop conditions, so the agent behaves the same way every time it runs. Every skill in this library is free, MIT-licensed, and downloadable as a folder or a ZIP.
Claude skills and Claude Code plugins
In claude.ai and Claude Desktop, upload a skill ZIP under Settings → Skills. In Claude Code, install the
Adspirer marketplace with /plugin marketplace add amekala/ads-mcp, then install
/plugin install adspirer-advertising-agent@adspirer-marketplace,
run /reload-plugins, then sign in.
Each skill page has the exact command for your client.
ChatGPT skills and the Adspirer app
Adspirer is an official ChatGPT App Store app. Open Apps, search Adspirer, choose
Connect, and sign in — no developer mode and no URL to paste. Then download any skill ZIP here
and upload it in ChatGPT, or point the agent at the raw
.md file.
Claude MCP and connectors for ad accounts
A skill describes the work; the MCP connection does it. Adspirer is the MCP server that reaches Google
Ads, Meta, Amazon, LinkedIn, TikTok, and ChatGPT Ads — 400+ tools across six platforms, plus GA4 and
Klaviyo. Add it as a custom connector at
https://mcp.adspirer.com/mcp.
Building an AI marketing agent
Stack several skills and you have an AI marketing agent that plans, launches, monitors, and reports on real campaigns. New campaigns are always created paused, destructive actions are gated behind an explicit confirmation, and every dollar waits for your approval. See what Adspirer can and cannot do.
GET adspirer.com/skills/agent.txt · catalog.json · /skills/<slug>.md