The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to generate shell scripts, such as replay.sh and target-global-preflight.sh, to facilitate rule migrations and deployments, which are subsequently executed via the system shell.
[COMMAND_EXECUTION]: The primary function of the skill is carried out by executing commands using the xgg CLI tool to interact with cloud and local IoT gateways.
[EXTERNAL_DOWNLOADS]: The skill utilizes npm for updating its CLI tool and allows the configuration of custom package registries through the XGG_UPDATE_REGISTRY environment variable.
[CREDENTIALS_UNSAFE]: The skill manages sensitive Xiaomi Home authentication data (e.g., ~/.config/mijia-api/auth.json) and local gateway passcodes. Direct handling of these credentials by the agent is required for operation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, such as device specifications and automation logs, to drive its rule-authoring logic.
Ingestion points: The agent reads device specifications via xgg device spec and retrieves runtime logs using xgg rule logs.
Boundary markers: While the skill suggests using validate and lint commands, it lacks strict delimiters to prevent data from being interpreted as instructions by the agent.
Capability inventory: The agent has the ability to write and enable automation rules, and perform direct device control via the cloud API.
Sanitization: The documentation emphasizes schema validation but does not detail specific methods for sanitizing data before prompt interpolation.

xgg-rule-authoring