google-cloud
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
gcloudandbqcommand-line interfaces to interact with cloud infrastructure. - Evidence: The file
SKILL.mdcontains numerous examples ofgcloudandbqcommands for resource discovery, data sampling, and log reading.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes data from external, untrusted sources. - Ingestion points: The agent reads BigQuery table contents (
bq head,bq query) and Cloud Logging entries (gcloud logging read) inSKILL.md, which are populated by external events and data. - Boundary markers: While the skill advises the agent to only quote relevant parts of logs, there are no explicit delimiters or instructions to treat the ingested data as untrusted content to prevent it from influencing the agent's logic.
- Capability inventory: The agent has the ability to execute shell commands (
gcloud,bq) and perform filesystem writes to a local context file (~/.gcloud/google-cloud-skill/context.json). - Sanitization: There is no evidence of sanitization or filtering of the external data before it is processed or presented to the agent context.
Audit Metadata