The Agent Skills Directory

Command Execution: The skill utilizes local shell scripts and standard tools like curl and jq for its operations. This approach is well-implemented, using structured parsing to ensure that inputs and API responses are handled securely without being treated as executable code.
External API Interaction: The skill communicates with bigquery.googleapis.com, which is a well-known service from Google. This interaction is necessary for the skill's primary functionality and follows standard HTTPS protocols.
Credential Management: Authentication is managed via bearer tokens injected by the runtime environment. The documentation follows security best practices by advising against manual token setup or debugging, ensuring credentials are managed within the expected security scope.
Indirect Prompt Injection Surface: As the skill processes data from BigQuery, it represents a potential surface for indirect prompt injection. However, the risk is mitigated by the use of structured data formatting.
Ingestion points: Data returned from the BigQuery API (found in scripts/bq_query.sh).
Boundary markers: The skill outputs data in structured TSV or JSONL formats.
Capability inventory: Capabilities are limited to shell-based API orchestration within the skill's scripts.
Sanitization: The skill employs jq for all data extraction and JSON construction, maintaining a strict separation between the data content and the logic of the tool.

bigquery-api