The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides installation instructions for macOS, Linux, and Windows that download and execute shell scripts directly from the vendor's domain.
Evidence: curl -fsSL https://unity.com/install.sh | bash and irm https://unity.com/install.ps1 | iex in SKILL.md.
[DYNAMIC_EXECUTION]: The skill facilitates the execution of arbitrary C# code and script files within a running local Unity Editor instance through specialized command-line tools.
Evidence: Usage of unity command eval and unity command eval_file documented in SKILL.md and references/integration-advanced.md.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined attack surface for processing untrusted project data and tool outputs.
Ingestion points: Ingests data from Unity project files (.unity, .prefab, .asset), Editor logs (Editor.log), and version control system outputs.
Boundary markers: Implements terminal hardening to strip C0 controls and non-SGR escape sequences from output.
Capability inventory: Performs subprocess execution (Bash), file manipulation (project/template creation), and local Editor command invocation.
Sanitization: Uses percent-encoding for service-specific annotations (GitHub Actions) and structured JSON for machine-readable output to prevent injection into consuming shells.
[EXTERNAL_DOWNLOADS]: The skill retrieves the CLI binary, Editor modules, and the Pipeline package from vendor-controlled infrastructure.
Evidence: Downloads from public-cdn.cloud.unity3d.com and the Unity UPM registry are documented across multiple reference files.

unity-cli