The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation scripts (scripts/install.sh and scripts/install.ps1) download the AWS CLI v2 installer from Amazon's official servers at https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip. This is a well-known and trusted source for the tool.
[COMMAND_EXECUTION]: The skill utilizes elevated privileges during the installation process. The Linux and macOS installation scripts use sudo to execute the AWS installer. While this is the standard recommended installation method for the AWS CLI, it involves granting the process root permissions.
[DATA_EXFILTRATION]: The skill's diagnostic scripts (scripts/aws-preflight.sh and others) specifically check for and reference sensitive AWS credential and configuration files located at ~/.aws/credentials and ~/.aws/config. Although accessing these files is necessary for the skill's primary function, their contents represent high-value targets for data exposure.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it retrieves and processes metadata from external AWS resources (e.g., S3 bucket names, EC2 tags, or IAM role names) that could be controlled by an attacker. If these resources are named maliciously, they could influence the agent's behavior when processed.
Ingestion points: Metadata returned from aws sts get-caller-identity, aws s3 ls, aws ec2 describe-instances, and other service inspection commands.
Boundary markers: No specific boundary markers or "ignore embedded instructions" warnings are used when processing command output.
Capability inventory: The skill possesses the ability to execute shell commands via the CLI and write diagnostic files to the local disk.
Sanitization: While character escaping is performed for JSON formatting in scripts/aws-diagnostics.sh, there is no evidence of validation or filtering intended to prevent instructional content from AWS resource names from being interpreted by the agent.

aws