The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from cloud environments (such as instance names, tags, and bucket names) and possesses the capability to execute administrative changes, creating a surface for indirect prompt injection.
Ingestion points: The skill fetches cloud resource details using boto3 (e.g., ec2.describe_instances, s3.list_buckets) in resources/implementation-playbook.md.
Boundary markers: While the skill advises validation in staging, it lacks explicit prompt delimiters to isolate external metadata from instructions.
Capability inventory: The skill provides logic for high-privilege operations including ec2.stop_instances, ec2.modify_instance_attribute, and lambda.put_function_concurrency.
Sanitization: The generated HTML dashboards in resources/implementation-playbook.md do not demonstrate explicit escaping for resource names, which could lead to script injection in the monitoring interface.
[DYNAMIC_EXECUTION]: The skill uses a template-based approach to generate executable scripts and configuration files at runtime.
Evidence: Functions such as create_rightsizing_automation, create_lifecycle_policies, and create_network_optimization_script in resources/implementation-playbook.md return strings containing Python code and shell scripts intended for execution.
[COMMAND_EXECUTION]: The skill instructions and associated playbook involve the execution of commands and scripts to manage cloud infrastructure.
Evidence: The implementation playbook relies heavily on the boto3 library to programmatically interact with AWS services and modify resource states.

database-cloud-optimization-cost-optimize