revenuecat-ad-tracking

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references several external SDKs and libraries from RevenueCat and Google. Specifically, it instructs the user to install packages from NPM (e.g., react-native-purchases-admob) and fetches Swift packages from RevenueCat's official GitHub repository (https://github.com/RevenueCat/purchases-ios-admob). It also mentions importing a .unitypackage for Unity integrations.
  • [INDIRECT_PROMPT_INJECTION]: The skill contains a pattern where it processes untrusted data from the user. Section 0 of SKILL.md instructs the agent to ingest a prompt pasted from the RevenueCat dashboard, stating it should be treated as the 'source of truth' for configuration details like the platform, mediator, and integration path. This represents a potential vulnerability surface where a maliciously crafted prompt could influence the agent's code generation.
  • Ingestion points: User-provided text (the dashboard prompt) and external ad events.
  • Boundary markers: The skill suggests specific sections (Context, Task, Verify) to look for, but lacks strict validation of the pasted content.
  • Capability inventory: The skill involves writing and modifying project source code, build scripts (build.gradle, Podfile), and dependency manifests (package.json, pubspec.yaml).
  • Sanitization: No explicit sanitization or safety check is performed on the pasted prompt before the agent adopts its claims as truth.
  • [COMMAND_EXECUTION]: The skill provides various shell commands for package management (npm install) and Unity package importing (-importPackage). These are standard development tasks for the intended purpose of SDK integration.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 12:37 PM