The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of ssh ha 'bash -lc ...' to execute shell commands directly on the Home Assistant OS host, bypassing standard agent abstractions.
[PRIVILEGE_ESCALATION]: The skill uses sudo docker to interact with containers on the host system, which requires elevated privileges and allows the skill to escape standard container isolation.
[CREDENTIALS_UNSAFE]: The skill involves manual handling of sensitive credentials, including:
Accessing and using the $SUPERVISOR_TOKEN for API authentication.
Managing rclone.conf files containing plaintext OAuth tokens (access_token, refresh_token).
Generating temporary local files (opts.json, payload.json) that may contain plaintext secrets.
[EXTERNAL_DOWNLOADS]: The skill fetches a third-party add-on repository from GitHub (https://github.com/jcwillox/hassio-rclone-backup) to install the backup utility.
[DYNAMIC_EXECUTION]: A Python script is generated at runtime via a heredoc (`python3
<<'EOF'`) to manipulate JSON configuration data before sending it to the Supervisor API.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes Home Assistant backup files located in the /backup directory.
Boundary markers: No specific boundary markers or sanitization logic is provided to separate backup data from the logic that handles file movement and deletion.
Capability inventory: The skill possesses the capability to perform network uploads (rclone copy) and execute file deletions (rclone delete) based on file naming patterns.
Sanitization: The skill does not perform sanitization of backup file metadata or content before processing them in the pipeline.

haos-cloud-backup