The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing external, potentially untrusted data from Google Cloud logs and resource configurations, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill reads external data using commands such as gcloud logging read (referenced in references/logging-monitoring.md) and gcloud asset search-all-resources (referenced in references/architecture-framework.md).
Boundary markers: While the skill encourages the use of structured formats like JSON, it does not specify explicit boundary markers or instructions to ignore embedded commands within the ingested data.
Capability inventory: The agent using this skill is empowered to perform sensitive operations, including IAM policy mutations (add-iam-policy-binding), resource deletions, and service deployments (gcloud run deploy), as outlined across the core rules and reference files.
Sanitization: The instructions do not provide specific sanitization or filtering logic for processing the content of log entries or metadata fields that could contain malicious payloads.
[SAFE]: The skill incorporates robust security controls, such as requiring human approval for all destructive and privilege-modifying actions (Core Rule 5 in SKILL.md) and strongly discouraging the use of static service account keys in favor of Workload Identity Federation (documented in references/iam.md).

gcp