setup-warehouse-bigquery

Warn

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill requires the dangerouslyDisableSandbox: true flag for multiple operations, including telemetry setup, cloud configuration via gcloud and bq, and direct API interactions using curl. This allows the agent to execute shell commands with elevated access to the host environment and external networks.
  • [CREDENTIALS_UNSAFE]: The skill manages sensitive information by storing a clientSecret (acquired from a remote endpoint) and temporary session IDs in plaintext files within the $TMPDIR directory. These files are used to authenticate subsequent telemetry and API calls. Additionally, the skill involves granting IAM permissions and handling service account emails.
  • [EXTERNAL_DOWNLOADS]: The skill initiates network requests to subdomains of confidence.dev to acquire telemetry keys, publish events, and resolve feature flags for verification purposes.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing data from external sources.
  • Ingestion points: Data is ingested from the output of gcloud and bq CLI tools, as well as JSON responses from the confidence-flags MCP server and external APIs (resolver.confidence.dev).
  • Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when processing tool or API outputs.
  • Capability inventory: The skill possesses significant capabilities, including the ability to write to the filesystem, perform network operations, and execute arbitrary shell commands via the Bash tool with the sandbox disabled.
  • Sanitization: There is no evidence of sanitization or validation of the data received from external tools before it is used to drive agent logic or interpolated into further shell commands.
  • [PRIVILEGE_ESCALATION]: The skill guides the user through granting high-privilege IAM roles (e.g., roles/iam.workloadIdentityUser, roles/bigquery.jobUser) to service accounts and requires the agent to operate with the sandbox disabled, effectively escalating the agent's operational permissions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 18, 2026, 07:20 AM
Security Audit — agent-trust-hub — setup-warehouse-bigquery