The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Phase 5 of the Unity Test Helpers reference instructs the agent to install the 'Unity CLI' if it is missing. The skill does not provide a specific source, package name (e.g., from npm or PyPI), or verification mechanism for this tool, which could lead to the installation of unverified or malicious software from the internet.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform operations such as unity status, unity mcp, and git diff --check. While these are functional for Unity development, the execution of shell commands based on internal logic provides a surface for command injection if inputs are manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read repository-local instruction files to guide its behavior, creating a surface for indirect prompt injection.
Ingestion points: Files named AGENTS.md, CLAUDE.md, and existing test files within Assets/Tests/** are read to determine project patterns and rules.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when reading these files.
Capability inventory: The skill has the ability to write files (Write, Edit) and execute shell commands (Bash), which could be abused if malicious instructions are ingested from the repo files.
Sanitization: There is no mention of sanitizing or validating the content of the ingested markdown or project files before following the instructions contained within them.

test-helpers