The Agent Skills Directory

[PROMPT_INJECTION]: Potential for indirect prompt injection from untrusted data sources.
Ingestion points: Data is retrieved from the UniFi Access controller via tools like access_list_events, access_recent_events, access_list_users, and access_list_visitors (found in references/access-tools.md).
Boundary markers: Absent; there are no instructions provided to the agent to treat external controller data as potentially untrusted or to ignore instructions embedded within retrieved fields.
Capability inventory: The skill can perform high-impact physical actions such as door locking/unlocking, security credential creation/revocation, and visitor pass management.
Sanitization: Absent; the skill does not specify validation or filtering of the content returned by the controller before it is processed by the agent.
[COMMAND_EXECUTION]: Use of dynamic execution for tool selection.
The access_execute tool enables the agent to invoke any tool by its name as a string. While part of the meta-tool discovery and lazy-loading architecture of the server, this provides a mechanism for dynamic operation selection based on input strings.

unifi-access