The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external data from a Home Assistant instance, which presents a surface for potential indirect prompt injection attacks.
Ingestion points: Entity states, attributes, and logbook entries are fetched via curl in scripts/ha.sh and used to inform agent actions.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potential commands embedded within entity names or state descriptions.
Capability inventory: The skill has the capability to perform network requests, read/write local files, and control physical hardware via HA service calls.
Sanitization: Data is parsed using jq to ensure valid JSON, but the content of the data is not sanitized for potential natural language instructions.
[COMMAND_EXECUTION]: The scripts/ha.sh CLI wrapper executes shell commands using curl and jq. It interpolates variables such as entity IDs and service names directly into these commands and JSON payloads.

home-assistant