The Agent Skills Directory

GCP Metadata Processing Surface: The skill performs automated discovery of agent runtimes, telemetry settings, and linked datasets by querying the Google Cloud project environment. Ingesting this external metadata into the agent's context is an indirect prompt injection surface where manipulated resource labels or configurations could theoretically influence the agent's reasoning.
Ingestion points: scripts/gather_agent_info.py (via Cloud Asset Inventory and AI Platform APIs).
Boundary markers: Absent in scripts processing API outputs.
Capability inventory: The agent can execute gcloud, bq, terraform, and python (specified in allowed-tools), perform network requests to Google Cloud APIs, and write configuration files.
Sanitization: Relies on the structured format of Google Cloud service responses.
System Utility Execution: The skill uses Python's subprocess module to invoke system utilities like gcloud, bq, and terraform for environment auditing and configuration deployment.
Evidence: scripts/gather_agent_info.py executes gcloud asset search-all-resources and gcloud beta monitoring metrics-scopes list to identify project parameters.
Context: The skill implements a mandatory 'Safety & Confirmation' protocol in SKILL.md, requiring explicit user approval for any 'Tier B' actions involving resource creation or billing.

agent-platform-alert-configuration