The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from the UniFi Access controller, including event logs, visitor metadata, and user descriptions, which could potentially contain malicious instructions.
Ingestion points: Data enters the context through access_list_events, access_recent_events, access_list_users, and access_list_visitors (as described in SKILL.md and access-tools.md).
Boundary markers: The skill requires a mandatory 'Coverage and Limitations' section and forces the agent to separate 'Observations' from 'Hypotheses', which effectively creates a structured reasoning boundary that reduces the likelihood of the agent following instructions embedded in log data.
Capability inventory: The skill can execute physical actions (locking/unlocking doors), manage security credentials, and reboot hardware devices through tools like access_unlock_door, access_create_credential, and access_reboot_device.
Sanitization: The server automatically redacts sensitive fields such as pin_code and token values (replacing them with ***REDACTED***) before the agent receives the data, preventing the exposure of credentials.

unifi-access