The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could contain malicious instructions intended to influence the agent's behavior.
Ingestion points: The DecodeAuthorizationMessage command accepts an EncodedMessage string, and AssumeRoleWithSAML/AssumeRoleWithWebIdentity accept authentication tokens from external providers.
Boundary markers: None identified in the skill instructions to separate data from instructions.
Capability inventory: The skill can perform critical AWS actions including AssumeRole, GetFederationToken, and GetSessionToken, which generate temporary security credentials.
Sanitization: There are no explicit instructions for the agent to sanitize or validate the content of the authorization messages before processing.
[CREDENTIALS_UNSAFE]: The skill is designed to handle and generate sensitive AWS security credentials.
The STSTool provides commands like GetSessionToken and AssumeRole that return temporary access keys, secret keys, and session tokens.
The GetAccessKeyInfo command allows looking up account information for specific access keys.
The skill enables the agent to manipulate its own permission set and identity within the AWS environment by design.

awsflow-sts