The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The Python scripts omada_api_client.py and deploy_omada_ssid.py contain hardcoded credentials (PASSWORD = "SecurePassword123" and PASSWORD = "SecretPassword123"). Furthermore, both scripts explicitly disable SSL certificate verification (self.session.verify = False and urllib3.disable_warnings), which exposes session tokens and credentials to man-in-the-middle attacks.
[COMMAND_EXECUTION]: The skill provides numerous instructions for executing privileged shell commands using sudo. This includes modifying system firewall rules (sudo ufw allow), managing system services (sudo systemctl restart), and deleting files in protected system directories (sudo rm -f /opt/tplink/EAPController/data/db/mongod.lock).
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill retrieves and displays untrusted data from the Omada Controller API, including device names, model identifiers, client connection statistics, and rogue AP alerts.
Boundary markers: The skill lacks boundary markers or instructions to treat data retrieved from the network controller as untrusted.
Capability inventory: The agent is authorized to perform privileged command execution (sudo), file system modifications, and network operations.
Sanitization: There is no evidence of sanitization or escaping of the metadata retrieved from the network controller before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The gpt.md file includes a Docker Compose configuration that pulls the third-party image mbentley/omada-controller:5.14 from a public registry.

omada-controller