App Store reviews
Any app's App Store reviews, newest first, from Apple's public feed. No key and read-only; the agent never replies to a review. In a helpdesk queue, the bugs they report become cards beside its tickets.
The App Store pack gives the agent one tool, appstore_reviews, that reads the reviews of any app on the App Store, yours or anyone else's. Ask in plain words: "show the latest App Store reviews for apps.apple.com/us/app/acme-notes/id1234567890", "what did the 1- and 2-star reviews say this month", "how is the new version being received".
Switch it on with configure-skills → App Store. There's no key and no Apple account involved: it reads Apple's public review feed. It never replies to a review.
What comes back
| Each review | Rating, title, text, the reviewer's nickname, the app version, the date and the storefront |
| Ratings | How many reviews got each star rating in the window |
| Coverage | Which storefronts had reviews, how many had none, and any that failed |
through | The moment of the read, so a later one can start exactly there |
Pass a link (apps.apple.com/…/id1234567890) or the numeric id, and optionally a window (7d, 3m, any, or a date), star filters and a limit. A limit of 0 returns the ratings and coverage with no review text, which is how sync counts reviews without reading them.
Storefronts
Reviews are kept per country, and there's no global feed, so a read is per storefront. You don't have to list countries:
- The first read of an app sweeps all 183 storefronts, US first. That takes about a minute and a half.
- After that the agent reads the storefronts that have produced reviews, plus 15 others in rotation, so a market that starts reviewing is found within a few reads.
- A full sweep runs again once a month.
What it learns lives in .state/appstore/<id>.json in your home. Name countries to narrow a read (storefronts: [us, gb]), or pass ['all'] for a full sweep.
When the feed comes back short
Apple's feed keeps roughly the 500 most recent reviews per storefront, and it answers many requests with an empty page: on some days most of them. Its cache then keeps that empty answer for minutes. The tool works around what it can and says what it couldn't:
- A blank page is asked again with a random value that gets past the cache: three more times, a few seconds apart, in a storefront that has had reviews before, and once elsewhere.
- A read where nothing came back gives no count at all, never "zero reviews".
- A storefront that has had reviews before and still comes back blank is named in the warnings, and the read isn't marked complete. So is one that fails (Apple sometimes refuses a few during a full sweep). The next read starts from the last complete one, so reviews a blank page hid aren't skipped.
- A long window can come back short: a blank page past a storefront's first can't be told from the end of its reviews.
In a helpdesk queue
Add the app to a queue profile and the helpdesk reads its reviews beside the tickets:
---
statuses: [Open, Awaiting agent reply]
appstore: 1234567890
---helpdesk reviews acme reads the reviews since the last complete read and turns the bugs they report into cards:
- It reads every review for a bug, whatever its stars: people who love an app still leave five stars when something breaks.
- Reviews that describe the same break become one group.
- For each group it gathers what's already known: open and closed cards, tickets that mention it, older reviews of the same break, and planned tasks.
- It offers one card that carries them all, or a comment on the card already open for it. The profile's
## Raisingrules still apply, such as which versions a card may come from.
A card quotes each review with its rating, version, country and date, never the reviewer's nickname, and lists its tickets as links. One with no ticket is titled (App Store reviews). The pulse sends its review groups through the same steps, and sync counts the day's reviews by star and offers helpdesk reviews acme when new ones arrived. Nothing replies to a review, from the helpdesk or anywhere else.
Helpdesk
Your support queue in SupportCandy or Pylon, worked the way your team works it. The agent reads, sorts and drafts; a person sends every reply. A queue profile per project holds its own pulse, bug lifecycle, routines and context.
Slack
Read-only Slack through your own account. A pulse across your channels, the replies you owe, threads where someone outside the workspace is waiting, and channels you joined that the agent doesn't know about yet. It never posts and never reads DMs.