launchpad-ux-custom-frontend

Fail

Audited by Gen Agent Trust Hub on Oct 30, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill's architecture relies on a configuration file (sdk-config.json) to store sensitive information including OAuth Client IDs and Secrets. While the file contains placeholders, the instructions guide users to populate these with real credentials.
  • [CREDENTIALS_UNSAFE]: The webpack.config.js file is configured to use CopyWebpackPlugin to copy sdk-config.json directly into the public distribution directory (dist/). This ensures that the file, containing any populated secrets, is served to the client-side browser and is accessible to any user.
  • [CREDENTIALS_UNSAFE]: The authentication logic in examples/workmanagement/src/context/PegaAuthProvider.tsx includes an initializeAuthentication function that reads mashupPassword from the configuration and uses window.atob() to decode it, confirming the intent to store and process passwords in the client-side configuration.
  • [COMMAND_EXECUTION]: The provided package.json and build instructions involve executing shell commands for building and serving the application, including webpack serve, webpack, and npx shx.
  • [EXTERNAL_DOWNLOADS]: The project configuration in package.json initiates downloads of numerous external Node.js packages from the NPM registry during installation. This includes vendor-specific packages like @pega/auth and @pega/constellationjs, as well as standard libraries like Material UI and React.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by ingesting and rendering data from the Pega Launchpad Server.
  • Ingestion points: Data entering the application context from the Pega server via the Constellation engine (PCore) and rendered through the <PegaContainer /> component in examples/workmanagement/src/components/Dashboard/index.tsx.
  • Boundary markers: Absent; there are no instructions or delimiters provided to the agent to treat data returned from the Pega server as untrusted or to ignore instructions embedded within that data.
  • Capability inventory: The skill has the capability to write data back to the server via PCore.getMashupApi().createCase and perform authenticated network operations (found in examples/workmanagement/src/context/PegaReadyContext.tsx).
  • Sanitization: The implementation relies on React's default output escaping, but does not implement specific sanitization for data flowing into the PCore lifecycle hooks.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 30, 2026, 08:38 AM