The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes high-privilege IAM management capabilities, including CreateRole, PutRolePolicy, and AttachRolePolicy. This creates an attack surface where an agent could be tricked by untrusted data into modifying cloud permissions or creating unauthorized administrative identities.
Ingestion points: IAM entity names and policy documents are passed as parameters to the agent during operation.
Capability inventory: The IAMTool provides extensive mutation and deletion capabilities for sensitive cloud resources (SKILL.md).
Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from executing malicious commands embedded within policies or resource descriptions.
Sanitization: The skill does not define validation or sanitization mechanisms for the JSON policy documents passed to the underlying AWS services.
[PRIVILEGE_ESCALATION]: The toolset allows for direct modification of the cloud account's security posture and permission boundaries.
Evidence: Commands such as PutRolePolicy, AttachRolePolicy, and UpdateAssumeRolePolicy enable the modification of trust relationships and access rights (SKILL.md).
Evidence: Identity lifecycle commands like CreateUser and CreateRole provide the ability to establish new persistent access points in the AWS environment (SKILL.md).

awsflow-iam