The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to target and harvest Google Cloud Service Account (SA) keys and tokens. It specifically references the local file service-account.json and the GCP metadata server endpoint metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/token for token extraction.
[PRIVILEGE_ESCALATION]: The instructions detail how to identify and exploit IAM privilege escalation paths within GCP. It lists specific permissions such as iam.serviceAccounts.actAs, iam.serviceAccounts.getAccessToken, and setIamPolicy as methods to reach "owner" status or run as a higher-privileged service account.
[COMMAND_EXECUTION]: The skill utilizes command-line tools like gcloud and ScoutSuite to perform enumeration, authentication, and policy manipulation. Example commands include gcloud auth activate-service-account and gcloud projects get-iam-policy.
[DATA_EXFILTRATION]: Instructions include techniques for identifying and accessing sensitive project information and data from services like Google Cloud Storage (GCS), Cloud SQL, and Secret Manager after obtaining credentials.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes untrusted data from GCP metadata services and tool outputs while maintaining significant capabilities.
Ingestion points: Metadata server responses, tool outputs (ScoutSuite, gcloud) in SKILL.md.
Boundary markers: Absent.
Capability inventory: Command execution via gcloud and ScoutSuite, network access to GCP APIs and metadata endpoints, file reading of service account keys.
Sanitization: Absent.

cloud-gcp