The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: Several scripts in the references/ directory interpolate shell variables directly into command-line arguments and JSON strings, creating a surface for injection if used with untrusted input.
Ingestion points: The scripts in references/cloudflare-dns-configuration.md and references/dns-failover-script.md accept command-line arguments and environment variables (e.g., DOMAIN, type, name, content, NEW_PRIMARY) that are passed directly to curl and aws commands.
Boundary markers: None are present to distinguish between trusted instructions and untrusted data in the variable inputs.
Capability inventory: The skill utilizes aws CLI for Route53 management, curl for Cloudflare API interaction and health checks, and jq for JSON processing.
Sanitization: Input variables are quoted in the shell context but are not escaped or validated before being interpolated into JSON payloads or shell commands, which could lead to malformed payloads if the data contains quotes or special characters.
[COMMAND_EXECUTION]: The skill uses bash scripts to perform DNS updates and monitoring. This is consistent with the skill's stated purpose of DNS management and does not involve unauthorized or hidden command execution.

dns-management