cloud-iam-privesc

Warn

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructs the agent on methods to bypass cloud security boundaries and gain administrative privileges. Evidence includes instructions for using specific IAM primitives such as iam:CreatePolicyVersion, iam:AttachUserPolicy, and iam:PassRole for AWS, as well as setIamPolicy for GCP.
  • [COMMAND_EXECUTION]: The skill facilitates the use of several command-line interfaces to modify cloud infrastructure. It explicitly guides the agent to use tools like awscli, gcloud, and az to enumerate permissions and change IAM policies.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for processing untrusted data to perform high-privilege operations.
  • Ingestion points: The skill processes potentially untrusted cloud identities such as leaked keys, service-account tokens, and assumed roles (SKILL.md).
  • Boundary markers: None identified to separate system instructions from identity metadata.
  • Capability inventory: The skill utilizes awscli, pacu, and gcloud to perform cloud resource modification and IAM policy updates (SKILL.md).
  • Sanitization: No sanitization or validation of the ingested credentials or identity metadata is identified.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 20, 2026, 11:02 PM
Security Audit — agent-trust-hub — cloud-iam-privesc