The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to delete or modify AWS CloudTrail logs (e.g., aws cloudtrail delete-trail), which are critical for audit visibility. These commands allow for the suppression of security alerts and the removal of evidence of malicious activity within the cloud environment.
[EXTERNAL_DOWNLOADS]: Fetches various security assessment and exploitation tools from public repositories, including RhinoSecurityLabs, NetSPI, BishopFox, NCC Group, and Duo Labs.
[PRIVILEGE_ESCALATION]: Contains detailed techniques for escalating IAM permissions. This includes creating new access keys for target users, attaching administrative policies to the current identity, and adding users to administrative groups.
[DATA_EXFILTRATION]: Outlines methods for harvesting sensitive credentials and data, such as extracting temporary IAM credentials from the EC2 Instance Metadata Service (IMDSv1/v2) and Fargate container environment variables. It also includes instructions for snapshotting and mounting EBS volumes to bypass OS-level file permissions.
[DYNAMIC_EXECUTION]: Includes instructions for modifying Lambda function code at runtime to inject malicious logic (aws lambda update-function-code) and executing arbitrary shell commands on EC2 instances via the AWS Systems Manager (SSM) send-command API.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection by ingesting untrusted data from external sources.
Ingestion points: Reads contents from S3 buckets (aws s3 sync), retrieves Lambda function source code, and mounts EBS volume snapshots containing arbitrary files.
Boundary markers: None identified; the skill does not use delimiters or instructions to ignore embedded commands in processed data.
Capability inventory: Includes powerful capabilities such as delete-trail, send-command (RCE), update-function-code (code injection), and IAM policy modification.
Sanitization: There is no evidence of sanitization or validation of the data retrieved from external AWS resources before it is presented to the agent or used in subsequent logic.
[PERSISTENCE]: Documents methods for maintaining access, including backdooring Lambda functions and container images within Amazon Elastic Container Registry (ECR).

aws-penetration-testing