The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute unity-cli subcommands for tasks like refreshing the asset database, creating materials, and analyzing dependencies. Documentation suggests selecting a binary from the PATH or using project-local build tools.
[DYNAMIC_EXECUTION]: The references/runtime-checklist.md file suggests using cargo run to compile and execute the unity-cli tool from the local workspace if a global binary is not installed, which is a standard development workflow pattern.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data, such as asset paths and material names, which are interpolated into JSON payloads for shell commands. While there is an attack surface for maliciously crafted inputs, the command execution is scoped to the unity-cli tool. Findings indicate:
Ingestion points: User-provided asset paths, material properties, and sprite paths in SKILL.md.
Boundary markers: JSON payloads are passed as quoted strings in the Bash examples, but explicit boundary markers for user-provided strings are absent.
Capability inventory: The skill uses Bash(unity-cli:*) for file and asset database operations.
Sanitization: No explicit sanitization or validation of the user-provided paths is described in the prompt construction.

unity-asset-management