The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill clones an external repository (https://github.com/hardrave/NIGHTRUN.git) and executes code using cargo xtask, cargo run, and shell scripts. Executing code from an unverified external source is a high-risk activity.
[COMMAND_EXECUTION]: Instructions guide the agent to perform hardware flashing on block devices (/dev/sdX) using a shell script (install.sh). This is a destructive operation that can lead to permanent data loss on the host system or connected media.
[EXTERNAL_DOWNLOADS]: The skill downloads a complete software environment from a GitHub repository not associated with the skill author or a trusted vendor.
[PROMPT_INJECTION]: The skill processes untrusted GGUF model files, creating a surface for indirect prompt injection if the model content influences the agent's behavior during the build or test phases.
Ingestion points: User-provided GGUF files processed by the nrconvert tool.
Boundary markers: The documentation mentions CRC-32 verification but lacks evidence of sanitization for embedded instructions.
Capability inventory: The environment allows shell command execution, file system modifications, and hardware device writes.
Sanitization: No explicit sanitization or filtering of external model content is described.

nightrun