The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the agentbox CLI to manage isolated coding environments, including provisioning sandboxes, launching agent sessions, and interacting with running boxes. These operations are the primary intended functionality of the skill.
[EXTERNAL_DOWNLOADS]: The instructions reference the @madarco/agentbox package on npm and integrate with well-known cloud providers such as Vercel, Hetzner, and Daytona for sandbox provisioning.
[PROMPT_INJECTION]: The orchestration recipes describe a pattern where a host-side agent monitors the status of an in-box agent (via agentbox agent wait-for input-needed) and takes actions based on its state. This creates an indirect prompt injection surface.
Ingestion points: Host-side scripts ingest state information and terminal snapshots from the sandboxed environments.
Boundary markers: No explicit prompt delimiters or instructions to ignore embedded content are described in the orchestration examples.
Capability inventory: The orchestration scripts can execute a variety of CLI commands, including simulated user input and file transfers between environments.
Sanitization: No sanitization of in-box agent output or status data is mentioned before it influences host-side decision making.

agentbox-info