The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses a central execution script, assets/score.sh, to extract auditing logic from its markdown documentation (e.g., references/reliability.md) and execute it as a bash script. This design ensures that the auditing rules are transparent and documented for the user while remaining functional. The execution is strictly limited to logic contained within the skill's own local reference files.
[INDIRECT_PROMPT_INJECTION]: The skill audits cluster configuration by ingesting large amounts of metadata from the target environment (e.g., pod names, deployment labels, and IAM roles). This untrusted data is then processed to generate the final narrative report, representing an indirect injection surface. This risk is inherent to auditing tools and is mitigated by standard agent guardrails and the tool's focus on deterministic scoring.
[EXTERNAL_DOWNLOADS]: Remediation instructions in the reference files contain URLs to official manifests and Helm charts from trusted organizations (e.g., AWS, Kubernetes, and Aqua Security). The instructions explicitly prohibit the agent from executing these downloads or applying them directly, leaving them for the user to review and execute if desired.
[COMMAND_EXECUTION]: The skill heavily utilizes the aws and kubectl CLIs. It employs defensive programming techniques to ensure these commands are executed safely, such as verifying that the Kubernetes context matches the AWS cluster through CA certificate comparisons and performing pre-execution permission checks. It also strictly validates cluster names and region codes to prevent command or path injection.

eks-well-architected-review