The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, including code diffs, PR descriptions, commit messages, and repository guidelines. This creates a surface for indirect prompt injection. The skill mitigates this by using a multi-engine integration process that requires findings to be 'grounded' against the actual source code in the main context, effectively filtering out instructions embedded in data that the AI engines might otherwise follow.
[DYNAMIC_EXECUTION]: The skill provides instructions for the agent to generate and execute Python scripts using the 'pty.spawn' module. This approach is used to handle specific terminal (TTY) requirements for the 'agy' CLI tool during headless execution. As the code is generated from a known template for infrastructure purposes, it is considered a low-severity dynamic execution finding.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading the 'agy' (Antigravity) CLI tool from Google's official installation domain and associated plugins from GitHub. These sources are categorized as trusted and the downloads are for standard development tools necessary for the skill's primary function.

judge