The Agent Skills Directory

[DYNAMIC_EXECUTION]: The calculate method within skill-template/index.ts uses the Function() constructor to evaluate a string as JavaScript. This enables arbitrary code execution within the agent's runtime environment if the input string is influenced by an attacker-controlled prompt or untrusted user input.
[INDIRECT_PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection as it ingests untrusted user data for character definitions and story themes without performing any validation or sanitization before use in high-privilege operations. * Ingestion points: Character appearance, backstory, and story theme inputs accepted by tools in skills/cinematic-script-writer/index.ts. * Boundary markers: The skill fails to use delimiters or 'ignore embedded instructions' warnings when constructing complex prompts in prompt-builder.ts. * Capability inventory: The skill has permissions to write files to Google Drive (saveToStorage) and make external network requests to the Google Drive API. * Sanitization: User-provided text is directly interpolated into generated prompts for downstream AI (like Sora or Midjourney) and project files without escaping or filtering, allowing an attacker to inject malicious instructions that could influence subsequent AI generations or corrupt persistent storage content.

cinematic-script-writer