Reviews → GitHub issues

• The user wants app store feedback in their issue tracker, once or on a schedule.
• Not for answering reviews (use reply-to-every-review with a connected store account) or for a one-off sentiment summary (use app-audit-lite).

• app (required): store URL, Apple id, Play package, bundle id or name.
• repo (required): owner/name of the GitHub repository. Default: the current repo (gh repo view --json nameWithOwner -q .nameWithOwner).
• since (optional): ISO date; default 7 days ago. Overlapping windows are safe (dedupe by review id).

Limits per run unless the user says otherwise: country us (add others on request), at most 5 new issues per run, at most 3 review pages.

Costs in AppGoat credits; GitHub calls are free. Without the MCP, use the REST equivalents in the REST calls below.

1. get_app {app} - 1 (+1 if refreshed). Note store, app_id, title, current version.
2. Existing topics (free): gh issue list --repo <repo> --label appgoat-reviews --state all --limit 200 --json number,title,state,body. Read the <!-- appgoat:topic=<slug> app=<store:id> --> markers so you reuse existing slugs for the same problem instead of inventing new ones.
3. get_reviews {app, country: "us", rating: "1-3", since, limit: 100} - 1 per page (+1 if the US reviews are refreshed). Follow next_cursor for up to 3 pages. Reviews carry AI tags (sentiment, intent, severity 0–3, topic) once tagged; untagged ones have them null. Use a tagged review's topic as the starting point for the slug and severity ≥ 2 as "high". Do not filter by intent/topic here: the filters drop untagged reviews, so read the whole 1–3★ window and classify the untagged ones yourself.
4. Group the reviews into topics (no calls). A topic is one concrete problem ("login loop after update", "widget doesn't refresh"), not a mood ("bad app"). Give each a lowercase slug (login-loop), reuse an existing slug when it is the same problem. Skip praise, spam, pure pricing rants with no actionable ask, and reviews under 20 characters.
5. For each topic, from most reviews to fewest:
   • bash scripts/upsert_issue.sh find --repo <repo> --app <store:app_id> --topic <slug> prints the issue (or none) and the review ids it already links. Drop those reviews.
   • Apply the decision rules, write the body to a temp file, then bash scripts/upsert_issue.sh upsert --repo <repo> --app <store:app_id> --topic <slug> --title "<title>" --body-file <file> [--reopen].
   • With the GitHub MCP server instead of gh: search_issues with repo:<repo> label:appgoat-reviews "appgoat:topic=<slug>" in:body, then add_issue_comment, update_issue (state open) or create_issue with the marker as the first line of the body and the label appgoat-reviews.
6. Report what was created, commented and skipped.

Budget: 2–10 credits per run (1 app lookup + 1–3 review pages + refreshes).

• New issue when the topic has no issue yet and either ≥ 2 reviews in the window, or 1 review mentioning a crash, data loss, a payment/subscription charge problem, a login lockout or a security/privacy problem. Otherwise list it under "watching" in the run report, no issue.
• Comment on the open issue when it exists and there are new (not yet linked) reviews.
• Closed issue: reopen (--reopen) when a new review is for a version released after the issue was closed or names the same symptom within 30 days of closing; otherwise create a new issue (the script links the old one).
• Never paste reviewer names or any personal data into issues; quote at most 200 characters per review; do not edit or close issues; do not assign people.
• Title: <Problem> (<n> reviews, <store>), ≤ 80 characters, e.g. Login loop after 3.2 update (4 reviews, App Store).
• Labels: always appgoat-reviews; add bug for crashes/errors or enhancement for feature requests only if those labels already exist in the repo.
• Severity line: high = crash, data loss, payment, lockout; medium = a broken feature with a workaround; low = UX friction or a feature request.
• Stop after 5 new issues; mention the remaining topics in the report.

Review links: Play https://play.google.com/store/apps/details?id=<package>&reviewId=<review_id>; Apple has no per-review URL, so use https://apps.apple.com/<country>/app/id<app_id>?see-all=reviews and give the review id. Always write review:<review_id> on the review's line so later runs dedupe.

New issue body (the script prepends the marker):

**Severity:** high · **Store:** App Store (us) · **Versions:** 3.2.0, 3.2.1 · **First seen:** 2026-10-18

<2-3 sentences: what users experience, when it started, any pattern (device, OS, version).>

### Reviews
- ★1 · v3.2.1 · 2026-10-20 · "Since the update it logs me out every time I open it…" · [store](https://apps.apple.com/us/app/id284882215?see-all=reviews) · review:11223344
- ★2 · v3.2.0 · 2026-10-19 · "…" · [store](…) · review:11223399

### Suggested next step
<one concrete debugging or product step>

_From AppGoat review data. Updated automatically; new reports are added as comments._

Comment on an existing issue:

**<n> new reviews** (<since> → <today>), versions <v>:
- ★1 · v3.2.1 · 2026-10-21 · "…" · [store](…) · review:11224000

Run report (to the user):

Reviews → issues for <app> since <date>: <n> reviews read, <t> topics.
- Created #123 Login loop after 3.2 update (4 reviews)
- Commented #98 Widget not refreshing (+2)
- Watching: dark mode request (1), price complaint (1)
Credits used: <sum>.

Without the AppGoat MCP server, call the API directly. Base URL https://api.appgoat.ai/v1, header x-api-key: $APPGOAT_API_KEY (create a key at https://dash.appgoat.ai). Responses are { "data": …, "meta": { "credits_used", "credits_remaining", "next_cursor", … } }; errors carry error.hint.

APP=com.spotify.music
curl -s -G -H "x-api-key: $APPGOAT_API_KEY" "https://api.appgoat.ai/v1/apps/$APP/reviews" \
  -d country=us -d rating=1-3 -d since=2026-10-15 -d limit=100 \
  | jq -r '.data[] | "\(.rating)★ v\(.version) \(.created_at) review:\(.review_id) \(.body[0:200])"'

Review fields: review_id, country, rating, title, body, version, thumbs_up, reply, replied_at, created_at, and the AI tags sentiment (positive|neutral|negative), intent (bug|crash|feature_request|pricing|ux|praise|spam|other), severity (0–3), topic, confidence (null until a review is tagged). author is present but must not be copied into issues.