The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an automated integration workflow that is susceptible to indirect prompt injection. In references/basic-integration-1.0-begin.md, the agent is instructed to read multiple files from the user's project to generate a tracking plan. This data ingestion occurs without boundary markers or instructions to ignore potentially malicious embedded prompts. The agent subsequently uses this information to perform code modifications and execute build scripts, creating a path for untrusted data to influence agent behavior.
[COMMAND_EXECUTION]: The documentation and integration scripts reference standard system commands for React Native development. This includes package installation (npm install), native dependency management (pod install), and environment configuration. The README.md within the example project includes the command sudo gem install cocoapods, which is a common but privileged instruction for setting up the iOS development toolchain on macOS.
[DATA_EXFILTRATION]: The skill's primary function is to capture application usage data and transmit it to PostHog. This involves sending identified user properties and error stack traces to the vendor's official domain (us.i.posthog.com). The implementation correctly uses environment variables via react-native-config to manage sensitive project tokens, which is a recommended security practice.
[SAFE]: An automated security alert regarding proguard-rules.pro was evaluated and determined to be a false positive. The file contains only comments and a legitimate link to official Android developer documentation. The binary data observed in android/app/debug.keystore is consistent with standard Android debug certificates used in development environments.

integration-react-native