Reference for all 13 Adspirer skills — the agent persona, the tool-call contract, six platform skills, four cross-platform workflows, and docs access.
Adspirer ships 13 skills — instruction files that teach an AI assistant how to run advertising
campaigns properly. They follow the open Agent Skills format, so the same
files work in Claude, ChatGPT, Codex, Cursor, Gemini CLI, and OpenClaw.You don’t invoke them by name. Describe what you want — “how are my Google Ads doing?”, “launch a
TikTok campaign” — and your assistant loads the right skill.
Skills load on demand. Your assistant sees only each skill’s name and description until a task
matches, then reads the full instructions.Two skills underpin everything:
adspirer-agent sets the safety rules — campaigns are created paused, nothing spends without
your explicit approval — and routes to the right skill.
adspirer-mcp carries the tool-call contract: how to reach each platform’s tools, which
account id each platform expects, and the budget units (Meta budgets are in cents, Google’s in
dollars).
Everything else builds on those two. A workflow skill like “launch a campaign” delegates to the
platform skill for that platform’s field rules, rather than duplicating them.
Every skill inherits the same safety contract from adspirer-agent:
Nothing spends without your approval
Campaigns are created paused on every platform. The agent never resumes a campaign, raises a
budget, or changes a bid strategy without you agreeing to that specific action first.
Read before write. The agent looks at what’s actually in the account before proposing a change.
Verify after write. After creating or updating anything, it reads the object back and reports
what exists — not what it asked for.
Never invent a number. If a tool fails, the agent says so. It does not estimate metrics.
Destructive means destructive. Deleting a creative or removing keywords is named as such, and
confirmed, before it happens.
In Claude, a scorecard, a set of ad-copy variants, or a campaign plan can be published as an
artifact — a live page that updates as work continues, private to you by default.In ChatGPT, something you’ll revisit or send to a client can be a hosted Site. The agent
defaults to invitation-only, because these pages carry your spend and conversion data.
Ask for a weekly review or a pacing check and the agent offers to schedule it, using whatever your
assistant calls it:
Assistant
What it’s called
Runs when your computer is off?
ChatGPT
Scheduled tasks
Yes
Codex
Scheduled tasks (automations)
Yes
Claude Cowork
Scheduled tasks, via /schedule
Yes — they run remotely
Claude Code
Routines — a local task or a remote routine
Local: no. Remote: yes
A Claude Code local scheduled task only runs while the desktop app is open and your computer is
awake. A laptop asleep at 9am skips the run. For budget pacing or spend alerts, the agent sets up a
remote routine instead.
Every scheduled run calls Adspirer tools again, so it draws on your monthly tool-call quota exactly
like a live conversation. The agent tells you what a cadence will cost before setting it up — weekly
is the sensible default for a performance review.If your assistant has no scheduler, Adspirer’s own monitoring tools cover the same ground
server-side: threshold alerts on ROAS, CPA and spend, and scheduled reports delivered by email.
