Terminal
Terminal/Skill
Checks a connected Apple Ads account for serving problems, duplicate keywords, missing negatives and install gaps against Adapty (it makes no changes).
By Adapty
Works with
npx skills add adaptyteam/apple-ads-cli --skill apple-ads-auditRead the account, explain what is observable, and stop before any mutation. This skill has no write
path. Route proposed fixes to apple-ads only after the user chooses an action.
When the full plugin is installed, use the operating skill's synced CLI references for command syntax:
../apple-ads/references/asa-management.md for entity and catalog reads.../apple-ads/references/asa-metrics.md for performance reads and metric definitions.When this skill was installed alone and those files are absent, use
adapty asa <topic> <command> --help for the installed version. The playbooks name the allowed
read commands. Never reconstruct a missing flag or endpoint from memory.
Run adapty asa whoami first in every session.
adapty-cli-setup and stop.apple-ads-strategy.Open references/INDEX.md and read the one playbook it names.
references/playbooks/account-health.md.references/playbooks/structure-audit.md.apple-ads, then its weekly-review.md playbook.Do not run both audit playbooks by default. Finish the requested audit, then offer the other as a separate follow-up if the evidence points there.
Allowed work:
whoami, list, get, metrics, and metrics overview reads.Forbidden work:
create, update, add, sync, pause, enable, or automation mutation.When the user asks to fix a finding, close the audit and hand the chosen finding to apple-ads as a
new task. The operator must re-resolve ids and obtain its own confirmation.
revenue, roas,
arpu, arppu, arpas, or roi.unknown or
not_applicable; do not score them as failures.Every audit response contains:
summary — a short plain-language answer.scope — company, app, entity scope, and identifiers used.date_window — exact dates and grouping, plus a cohort window only for a cohort root.findings — ordered by severity.evidence — ids that map every claim to returned data.confidence — overall confidence and the limiting missing data.unknowns — questions the available surface cannot settle.recommended_actions — read-only follow-ups or operator handoffs.Each finding contains:
| Field | Meaning |
|---|---|
status | healthy, attention, critical, unknown, or not_applicable |
observation | What the returned data directly shows |
explanation | The simplest supported interpretation, clearly labeled |
evidence | Evidence ids, metric names, windows, and affected entity ids |
confidence | high, medium, or low, with a reason |
recommendation | The next read or an apple-ads handoff |
limitations | What prevents a stronger conclusion |
Use critical only for a confirmed serving failure or a user-defined limit that was actually
breached. Use attention for a supported concern without that proof. Never invent a numeric health
score or universal threshold.
This is the entire attribution scope of this skill:
total_installs from Apple with adapty_installs from Adapty for the same scope and exact
date window.absolute_gap = apple_installs - adapty_installs.relative_gap = absolute_gap / apple_installs only when apple_installs > 0.unknown. If Apple installs are zero, say the
percentage is undefined.Name correlation as correlation: use "coincided with" when change history is unavailable, never "was caused by".
unknown, not unprofitable.429 that reaches the agent means stop. Do not sleep and retry in a loop.apple-ads-strategy.apple-ads after the audit is complete.