firebase-ai-logic

Warn

Audited by Gen Agent Trust Hub on Apr 5, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [METADATA_POISONING]: The skill falsely identifies as an 'Official skill' despite being provided by an external author. It references non-existent API endpoints in the Firebase SDK (firebase/ai) and hallucinated model versions such as 'Gemini 2.5' and 'Nano Banana', which may mislead the agent or users into attempting to use unsupported features.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill processes user-supplied text prompts and multi-modal files (images, audio, video, PDF) through functions like analyzeImage and sendMessage in references/usage_patterns_web.md.
  • Boundary markers: Absent. There are no instructions to the model to ignore instructions contained within the external files or data parts being processed.
  • Capability inventory: The skill utilizes network capabilities to send data to generative AI backends (model.generateContent).
  • Sanitization: Absent. External content is interpolated directly into the API calls without validation or escaping.
  • [COMMAND_EXECUTION]: The skill instructs the user to execute shell commands using the Firebase CLI (npx -y firebase-tools@latest) for project management and initialization. While these target a well-known service, the init # Choose AI logic command references a non-existent feature in the official CLI tool.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 5, 2026, 06:41 PM
Security Audit — agent-trust-hub — firebase-ai-logic