The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The wrapper script scripts/hab.sh is designed to download platform-specific binaries for the 'hab' CLI from a GitHub repository (github.com/balloob/home-assistant-build-cli).
[REMOTE_CODE_EXECUTION]: The skill executes the downloaded binary file directly using the exec command within the bash script.
[PRIVILEGE_ESCALATION]: The script performs a permission modification using chmod +x on the binary downloaded at runtime to the user's cache directory (~/.cache/hab-cli).
[COMMAND_EXECUTION]: The skill utilizes shell commands such as curl, wget, and chmod to facilitate the installation and execution of the external CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes state and configuration data from an external Home Assistant instance, creating a potential attack surface.
Ingestion points: Data enters the agent's context through commands like hab entity list, hab entity get, and various dashboard or automation schema inspection commands.
Boundary markers: The skill relies on JSON envelopes for structured output but lacks explicit delimiters or instructions to ignore instructions embedded within Home Assistant configuration fields.
Capability inventory: The hab tool possesses significant capabilities, including the ability to restart the system, delete backups, and modify network settings.
Sanitization: No explicit sanitization or validation of the data retrieved from Home Assistant is documented before processing.

hass-builder