The Agent Skills Directory

[PRIVILEGE_ESCALATION]: Provides specific IAM exploitation methods (e.g., using 'iam:CreateAccessKey' or 'iam:AttachUserPolicy') and code samples to elevate user permissions to the 'AdministratorAccess' level.
[PERSISTENCE]: Includes instructions for establishing and maintaining access to compromised environments, such as backdooring Lambda functions and disabling or deleting AWS CloudTrail logs to hide malicious activity.
[DATA_EXFILTRATION]: Details techniques for harvesting temporary security credentials by exploiting Server-Side Request Forgery (SSRF) against the Instance Metadata Service (IMDSv1 and IMDSv2).
[COMMAND_EXECUTION]: Instructs the agent on how to use AWS Systems Manager (SSM) to execute arbitrary shell commands on managed EC2 instances through the 'AWS-RunShellScript' document.
[REMOTE_CODE_EXECUTION]: Includes Python code snippets designed to be injected into Lambda functions to modify IAM policies and grant elevated rights to the attacker.
[EXTERNAL_DOWNLOADS]: Recommends fetching various third-party exploitation frameworks and tools from external GitHub repositories, including Pacu, enumerate-iam, and weirdAAL.
[DYNAMIC_EXECUTION]: Provides guidance on generating and deploying malicious ZIP files and script payloads for runtime execution within the target's cloud infrastructure.
[INDIRECT_PROMPT_INJECTION]: Identifies an attack surface where the skill processes unvalidated data from AWS CLI outputs, S3 bucket enumeration tools, and external discovery URLs without using boundary markers or input sanitization.
Ingestion points: Processes JSON outputs from AWS CLI commands (e.g., 'iam list-users', 'sts get-caller-identity') and data from external discovery tools like bucket_finder.
Boundary markers: No delimiters or instructions are present to separate untrusted tool outputs from the agent's logic.
Capability inventory: Extensive capabilities including shell command execution, file system writes (malicious.zip), and network operations for SSRF.
Sanitization: No validation or escaping logic is described for handling external content before processing.

AWS Penetration Testing