The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a local Python script (scripts/lint_flow.py) for static analysis and validation of the generated YAML automation flows. The agent is instructed to run this script using python3 to ensure script quality before delivery.
[EXTERNAL_DOWNLOADS]: The skill guides the agent to use curl and jq to fetch UI structure data and screenshots from a user-specified host IP on port 18182. This is a core part of the workflow to understand the target application's interface.
[REMOTE_CODE_EXECUTION]: The skill is designed to generate scripts that are transmitted to and executed by a remote VMOS Edge engine on port 47218. The communication protocols and payload structures are documented for the agent to facilitate this remote execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests UI hierarchy data (accessibility dumps) from the target device to generate automation selectors. While this data is external and untrusted, it is necessary for the skill's operation.
Ingestion points: Accessibility dump output (screen.xml) retrieved via the device API.
Boundary markers: None specified for the UI data ingestion.
Capability inventory: Shell command generation, network requests, and local script execution.
Sanitization: The provided linter validates engine-specific syntax but does not filter for potential injection patterns in UI text.

vmos-edge-flowsmith