The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends downloading and installing the homekit-cli package from the NPM registry and the HomeClaw application from the Mac App Store.
[REMOTE_CODE_EXECUTION]: The instructions utilize npx -y homekit-cli to execute code directly from the NPM registry and npx skills add l3wi/homekit-cli to fetch and integrate the skill from GitHub.
[COMMAND_EXECUTION]: The skill makes extensive use of the homekit CLI to interact with the local operating system and a specific Unix socket (~/Library/Group Containers/group.com.shahine.homeclaw/homeclaw.sock) for communication with the HomeClaw provider app. It also includes a readiness shell script (scripts/homekit-readiness.sh) that executes multiple status commands.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface where it ingests untrusted data from the HomeKit environment, creating a potential vector for malicious metadata to influence the agent's logic.
Ingestion points: Device names, room names, zone names, and HomeKit event logs accessed via commands like homekit accessories list and homekit events list (referenced in SKILL.md and references/inspection-workflows.md).
Boundary markers: The instructions emphasize using the --format json flag for precise data parsing and mandate resolving fuzzy names to exact IDs to prevent ambiguity.
Capability inventory: The CLI can control safety-critical hardware (locks, garage doors, gates, security systems) and perform structural mutations (deleting or modifying scenes and automations).
Sanitization: The skill relies on operator-level safety checks, including mandatory user confirmation for physical actions and explicit safety flags (--allow-actuation, --allow-mutation) to authorize changes.

homekit