genkit

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill guides the installation of the genkit-cli and official plugins from the NPM registry and GitHub repositories associated with Firebase Genkit.\n- [REMOTE_CODE_EXECUTION]: Includes a setup command that fetches and executes a script from cli.genkit.dev. This is the official installation method for the Genkit CLI tool.\n- [COMMAND_EXECUTION]: Provides instructions for running local commands such as npx tsx for development and firebase deploy for production deployment.\n- [PROMPT_INJECTION]: Defines AI flows that interpolate untrusted input directly into prompts, creating an indirect prompt injection surface.\n
  • Ingestion points: Inputs for flows in SKILL.md (e.g., text in summarizeFlow, question in agentFlow).\n
  • Boundary markers: Generally absent in standard template literals; Dotprompt examples utilize {{variable}} delimiters.\n
  • Capability inventory: Includes model plugin integration for network LLM calls and dev-local-vectorstore for local file system interaction.\n
  • Sanitization: Uses Zod for structural validation of inputs, but does not provide specific filtering for malicious LLM instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 06:51 AM