gcp-monitoring
Pass
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches the Ops Agent installation repository script from Google's official distribution domain (dl.google.com).
- [REMOTE_CODE_EXECUTION]: Executes the official Google Cloud Ops Agent setup script via bash during the installation process.
- [COMMAND_EXECUTION]: Utilizes
sudofor installing the Ops Agent package and managing its systemd service status (restart/status). - [DATA_EXFILTRATION]: Provides commands to read cloud logs and configure sinks that export data to external destinations like BigQuery, GCS, or Pub/Sub. This is standard log management functionality within GCP.
- [PROMPT_INJECTION]: The skill ingests data from Google Cloud Logging which may contain untrusted strings from external applications, serving as a surface for indirect prompt injection.
- Ingestion points:
gcloud logging readandgcloud logging tailcommands in SKILL.md. - Boundary markers: None identified for delimiting log content from agent instructions.
- Capability inventory: Subprocess execution via
gcloudandshelltools as defined in the skill's frontmatter. - Sanitization: No content filtering or validation is specified for the ingested log data.
Audit Metadata