Copycat Reviews

Turn selected iOS and Android links into a compact, evidence-backed competitor study. The bundled Python-only script is the data layer; agent-written analysis is the judgment layer.

Run the pipeline

1. Resolve <skill-dir> as the directory containing this SKILL.md. Default to <workspace>/.copycat/apps. Ask for market codes when scope is absent; if an unattended orchestration cannot ask, use bounded worldwide and report its limits.

2. Run a mixed batch. --markets accepts comma-separated two-letter codes or worldwide:

   python3 <skill-dir>/scripts/app_reviews.py run <app-url> [<app-url> ...] --markets <codes-or-worldwide>
   

   This adds only new platform/stable-ID pairs, samples requested markets, deduplicates one review across observed markets while retaining observed_markets, and rebuilds compact evidence. Folders use <slug>--<platform>--<stable-id> under .copycat/apps.

3. Use sync to refresh configured apps and brief to rebuild summaries without network access.

4. Read <review-root>/portfolio-evidence.md first, then relevant <app-folder>/evidence.md. Treat reviews.jsonl as storage, not context.

5. Verify uncertain insights with a small raw slice: