implementing-canary-tokens-for-network-intrusion

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt instructs generating and embedding real-looking secrets (AWS access keys, webhook URLs, Thinkst API tokens, etc.) into files, commands, and alerts—requiring the model to handle and potentially output those secret values verbatim, which creates an exfiltration risk.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs modifying system-level files and infrastructure (e.g., /etc/hosts, internal DNS zone files, Active Directory GPOs, /etc/app/db.conf, writing to /var/log, deploying fake AWS creds, and requires "administrative access"), which pushes the agent to perform privileged, state-changing actions on the host or network.