The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py performs highly sensitive AWS operations using the boto3 library.
It intentionally opens security group ingress to the public internet (0.0.0.0/0) on port 22 within the experiment_open_security_group function.
It disables CloudTrail logging via the ct.stop_logging call in experiment_stop_cloudtrail, which can be used to blind security monitoring and auditing.
[PRIVILEGE_ESCALATION]: The experiment_create_admin_user function in scripts/agent.py creates a new IAM user and attaches the AdministratorAccess managed policy, granting full administrative control over the target AWS account.
[PERSISTENCE]: The creation of a new administrative IAM user is a classic technique for maintaining persistent access to a cloud environment, which the skill facilitates through its experiment logic.
[INDIRECT_PROMPT_INJECTION]: The skill features a vulnerability surface where it ingests and processes untrusted data from external security services.
Ingestion points: scripts/agent.py reads data from the outputs of gd.list_findings, config_client.get_compliance_details_by_config_rule, and cw.describe_alarms.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are present when the agent processes these API outputs.
Capability inventory: The script possesses broad administrative capabilities across IAM, EC2, CloudTrail, Config, GuardDuty, and CloudWatch.
Sanitization: There is no evidence of sanitization, validation, or escaping of the data received from these external AWS services before it is used to influence the agent's decision logic.

implementing-security-chaos-engineering