The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes hardcoded example values for sensitive cluster configuration parameters.
Evidence: A static gossip encryption key (A4F4+oAxyOvAThY4c3Kt0sMN0+SIpxEwbjxZTr/BRh0=) and a master ACL token (a34adf98-9b1d-4be5-8f6a-5c1f2a3b4c5d) are provided in the HCL configuration blocks in SKILL.md.
[COMMAND_EXECUTION]: The skill provides numerous shell commands for system and cluster administration.
Evidence: Includes instructions for service management via systemctl, modification of network resolver settings in /etc/resolv.conf, and interaction with binaries such as consul, dig, and helm.
[EXTERNAL_DOWNLOADS]: The skill fetches infrastructure resources from established providers.
Evidence: Configures and installs packages from HashiCorp's official Helm repository (https://helm.releases.hashicorp.com).
[INDIRECT_PROMPT_INJECTION]: The skill processes external data sources that could be manipulated to inject malicious instructions into the agent's context.
Ingestion points: Data entering through the Consul catalog (service names/metadata), health check outputs, and the KV store (SKILL.md, references/consul-service-discovery.md).
Boundary markers: Not explicitly implemented in the prompt generation logic.
Capability inventory: The skill can execute scripts defined in health checks, render files via consul-template, and trigger infrastructure updates using consul-terraform-sync.
Sanitization: No explicit validation or filtering of ingested service or KV data is described.
[DYNAMIC_EXECUTION]: The skill utilizes automation tools that generate and apply configuration dynamically based on cluster state.
Evidence: consul-template renders files based on KV values, and consul-terraform-sync executes Terraform modules in response to catalog changes.

consul