The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the gcloud CLI for authentication and resource enumeration. Specific commands include gcloud auth activate-service-account --key-file=sa.json and gcloud projects get-iam-policy, which involve sensitive credential management and system discovery.
[DATA_EXFILTRATION]: The instructions describe how to extract sensitive tokens from the Google Cloud Metadata Server (metadata.google.internal) and how to target Cloud Storage (GCS) buckets and Secret Manager for data access. It specifically targets the Metadata-Flavor: Google header to bypass security checks for Server-Side Request Forgery (SSRF).
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external GCP API responses to determine subsequent attack steps.
Ingestion points: Data enters the context via responses from gcloud commands (IAM policies, project lists) and the GCP metadata server.
Boundary markers: None identified; instructions do not specify delimiters for parsing external API output.
Capability inventory: The skill uses gcloud, gcloud-iam, and ScoutSuite to interact with cloud infrastructure and modify permissions.
Sanitization: There is no mention of sanitizing or validating the integrity of the data retrieved from the cloud environment before it influences the agent's logic.

cloud-gcp