The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from external patent abstracts and claims which could potentially contain malicious instructions.
Ingestion points: Patent data retrieved via search_patents and get_patent methods from the BigQueryPatentSearch class.
Boundary markers: Absent; there are no specific instructions or delimiters to isolate retrieved patent text from the agent's control logic.
Capability inventory: Bash, Read, Write tools allow for command execution and file operations.
Sanitization: No sanitization of the retrieved text is described.
[Data Exposure & Exfiltration] (SAFE): No hardcoded credentials or sensitive file access patterns were detected. The skill uses standard Google Application Default Credentials (ADC).
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The skill relies on the standard google-cloud-bigquery library and references official Google Cloud documentation for setup.
[Command Execution] (SAFE): While the skill utilizes the Bash tool, it is intended for executing the internal Python search scripts as part of its primary functionality.

bigquery-patent-search