The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py automates cloud security operations by executing AWS CLI commands through the subprocess.run function. These operations include administrative tasks such as disabling compromised IAM access keys (aws iam update-access-key), attaching deny-all policies to identities (aws iam attach-user-policy), and isolating compromised EC2 instances (aws ec2 modify-instance-attribute). The implementation correctly utilizes argument lists within the subprocess calls to mitigate the risk of shell injection attacks.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential vulnerability surface for indirect prompt injection because it ingests untrusted external data from cloud audit logs via aws cloudtrail lookup-events. An adversary with an existing presence in the cloud environment could potentially inject malicious instructions into log fields (such as User Agent strings or resource names) that the agent later processes or summarizes. 1. Ingestion points: The aws_query_cloudtrail function in scripts/agent.py fetches event data from AWS. 2. Boundary markers: The script does not use specific delimiters or instructions to ignore embedded content within the log data. 3. Capability inventory: The agent has capabilities to write to the local file system (json.dump) and perform administrative actions in the AWS environment. 4. Sanitization: There is no evidence of specific filtering or sanitization of the content retrieved from CloudTrail beyond standard JSON parsing.
[SAFE]: The requests library is listed as a dependency in the references/api-reference.md documentation, but it is not imported or used in the scripts/agent.py code. The skill relies exclusively on the pre-installed AWS CLI for all remote service interactions.

conducting-cloud-incident-response