The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the sprite exec command to run arbitrary shell commands and binaries on remote Sprite VMs, giving the agent full administrative control over the remote environment.
[REMOTE_CODE_EXECUTION]: The core functionality of the skill is to facilitate the execution of code on external infrastructure (Fly.io VMs) through a local CLI interface.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for transferring files and data between the local environment and remote VMs using commands like cat and base64, which can be used to move sensitive information or deploy external payloads.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface (Category 8). Evidence: 1. Ingestion points: The agent reads untrusted data from the remote VM via sprite exec output, including file contents from /home/sprite/ and log files from /.sprite/logs/. 2. Boundary markers: No delimiters or specific instructions are provided to help the agent distinguish between remote data and its own system instructions. 3. Capability inventory: The agent has the ability to execute arbitrary commands, destroy VMs via sprite destroy, and modify network authentication/policies via sprite url update and the REST API. 4. Sanitization: No sanitization or validation of the remote output is performed before processing.

sprites-dev