The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to analyze external, potentially untrusted data such as application source code, cloud infrastructure definitions, and billing reports. This ingestion surface, combined with the agent's ability to execute powerful cloud management commands, constitutes a surface for indirect prompt injection (Category 8).
Ingestion points: Analyzes user-supplied code snippets, Dockerfiles, Kubernetes manifests, and cloud configuration files (referenced in SKILL.md and cloud-and-infra.md).
Boundary markers: The skill does not provide explicit delimiters or instructions to the agent to isolate the analyzed data from its core system instructions.
Capability inventory: The skill utilizes shell commands including aws s3api, aws logs, aws budgets, and kubectl (documented in SKILL.md, cloud-and-infra.md, and services-and-finops.md).
Sanitization: There is no evidence of input validation or sanitization mechanisms to prevent the execution of instructions embedded within the analyzed data.
[COMMAND_EXECUTION]: The skill provides numerous pre-written shell command templates for AWS and Kubernetes management (e.g., aws s3api put-bucket-intelligent-tiering-configuration, aws logs put-retention-policy, kubectl top pods, aws budgets create-budget). While these commands are correctly aligned with the stated purpose of cost optimization, they represent significant system capabilities that require careful oversight during execution.

cost-reducer