The Agent Skills Directory

[SAFE]: The skill is primarily composed of documentation and structured runbooks (SOPs) for troubleshooting AWS Glue. It relies on standard, read-only AWS CLI commands (such as aws glue get-job, aws logs filter-log-events, and aws ec2 describe-security-groups) to gather diagnostic information.
[DATA_EXPOSURE]: Every runbook includes a dedicated section for "Data Sensitivity" and "Prohibited Actions." These sections correctly identify that CloudWatch logs and metadata may contain sensitive info and provide explicit instructions to the agent to avoid exposing S3 paths, database credentials, or KMS key policies externally.
[PROMPT_INJECTION]: The skill implements extensive defensive measures against improper agent behavior. It includes a specific glue-hallucination-patterns.yaml file and glue-guardrails.md to prevent the AI from making dangerous assumptions, such as confusing executor and driver memory errors or assuming job bookmarks are automatic.
[COMMAND_EXECUTION]: While the skill lists CLI commands, they are presented as read-only diagnostic steps for an administrator role. The remediation advice (e.g., modifying security group rules) is standard for AWS troubleshooting and is constrained by clear "Safety Ratings" and "Escalation Conditions."

glue-diagnostics