The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from AWS APIs, such as IAM usernames and policy descriptions, which are included in generated audit reports. This creates a surface for indirect prompt injection if an agent later processes these reports and interprets the metadata as instructions.
Ingestion points: scripts/agent.py retrieves entity information from the AWS IAM and Access Analyzer APIs.
Boundary markers: The audit output is stored in JSON/CSV formats without explicit delimiters to isolate untrusted metadata from instructions.
Capability inventory: The skill performs local file writes to save report data and prints findings to the console.
Sanitization: No sanitization is applied to the strings retrieved from the AWS environment before they are recorded.
[COMMAND_EXECUTION]: The skill provides documentation for executing AWS CLI commands for security auditing and includes a Python script that automates these audits using the boto3 library.
[EXTERNAL_DOWNLOADS]: The skill utilizes the boto3 Python package, which is the official and trusted SDK provided by Amazon Web Services for interacting with cloud infrastructure.

securing-aws-iam-permissions