The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by instructing the AI agent to ingest data from an external executable's help output and use that data to construct subsequent commands.
Ingestion points: The agent is instructed to run unity <relevant-command> --help and parse the resulting stdout to determine available syntax (referenced in SKILL.md).
Boundary markers: Present. The skill provides clear instructions to treat help output as authoritative but specifically mandates preflight checks and user authorization before any state-changing command is executed.
Capability inventory: The agent is capable of executing arbitrary Unity CLI commands, including project creation, license activation, and build execution.
Sanitization: Absent. The help output is not sanitized before being processed by the model.
[EXTERNAL_DOWNLOADS]: Fetches the official Unity CLI installer from a well-known service domain.
Source: https://public-cdn.cloud.unity3d.com/hub/prod/cli/install.sh and install.ps1 (referenced in references/installation-and-upgrade.md).
[REMOTE_CODE_EXECUTION]: Provides instructions for the agent to install the Unity CLI by piping remote content directly into shell interpreters.
Pattern: curl -fsSL [URL] | bash and irm [URL] | iex for the official Unity CLI installer.

unity-cli