The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): Step 1 of the skill recommends installing the Genkit CLI using curl -sL cli.genkit.dev | bash. This 'piped remote execution' pattern is dangerous because it executes unverified scripts from the internet directly on the host machine without integrity checks.
EXTERNAL_DOWNLOADS (MEDIUM): The skill requires installing multiple Node.js packages (genkit, @genkit-ai/google-genai, genkit-cli) from external registries, increasing the software supply chain attack surface.
COMMAND_EXECUTION (LOW): The skill utilizes command-line tools like npm, npx, and genkit start to initialize projects and run development servers, which involve executing shell commands with user-defined paths.
PROMPT_INJECTION (LOW): The skill provides templates for building RAG pipelines and tool-calling agents, which are inherently vulnerable to indirect prompt injection from untrusted data sources.
Ingestion points: Data enters the agent via inputSchema in flows and through external tool outputs (e.g., the fetchWeather example).
Boundary markers: No explicit delimiters or boundary markers are used in the prompt templates to separate instructions from untrusted data.
Capability inventory: The agents created can perform network operations (fetch) and execute arbitrary code via the flow runner.
Sanitization: The provided examples do not demonstrate any sanitization or validation of external content beyond schema definition.

genkit