The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from the UniFi API, such as device names and hostnames, which could contain malicious instructions.
Ingestion points: Data enters the agent context through API endpoints such as /stat/sta (clients), /stat/device (hardware), and /rest/networkconf (VLANs/SSIDs) in SKILL.md.
Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized when presenting network data to the agent.
Capability inventory: The skill utilizes Bash, python3, curl, and unifi CLI tools for system and network operations across SKILL.md.
Sanitization: There is no evidence of sanitization or filtering of external data retrieved from the UniFi controller.
[COMMAND_EXECUTION]: Instructions include the creation and execution of a local bash wrapper script (~/.local/bin/unifi) and the use of the op run command to inject secrets into the execution environment.
[EXTERNAL_DOWNLOADS]: The skill downloads OpenAPI specifications from Ubiquiti's official developer portal (developer.ui.com) and utilizes the Homebrew package manager to install the restish utility.
[DATA_EXFILTRATION]: The skill accesses highly sensitive network information, including client throughput, IP/MAC addresses, firewall rules, and WiFi passphrases. While it uses 1Password to handle the API key securely, the agent's ability to read and potentially transmit this configuration data represents a data exposure surface.
[COMMAND_EXECUTION]: The skill documentation recommends disabling SSL/TLS certificate verification using ssl._create_unverified_context() in Python and the -k flag in curl to accommodate self-signed certificates common in local UniFi deployments.

unifi-api