The Agent Skills Directory

[Command Execution for Resource Management]: The skill utilizes shell commands and the Python subprocess module to manage Google Cloud resources via the gcloud CLI, such as deleting services during cleanup. This is a standard approach for infrastructure automation and is used to provide the user with a streamlined setup and cleanup process.
[External Package Management]: During installation, the skill fetches several dependencies from the official Python Package Index (PyPI). These include official Google Cloud SDKs and common utilities necessary for the skill's functionality, such as google-cloud-bigquery and python-dotenv.
[Indirect Prompt Injection Considerations]: The skill processes external product data which is then retrieved and presented by the agent. This represents a common pattern for semantic search applications where the agent summarizes indexed content.
Ingestion points: Product catalog data is loaded from user-provided files (CSV, JSON, or JSONL) in scripts/ingest_bigquery.py and scripts/ingest_vertex_search.py.
Boundary markers: The agent's core instructions in scripts/agent.py provide clear guidance on result presentation, though they do not use strict delimiters for the tool's output.
Capability inventory: The system performs database queries and vector searches; the installation scripts have standard permissions to write configuration files and install packages in a virtual environment.
Sanitization: Input data undergoes schema validation to ensure field types are correct before indexing, while natural language content is processed by the language model's inherent safety mechanisms.

retail-product-search