The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill scans project test files and design documents to tailor its output. While processing untrusted local content is a known surface for indirect prompt injection, the risk is negligible here as the tool only generates boilerplate code, does not execute it, and includes a mandatory human-in-the-loop confirmation step before any file operations occur.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads local project metadata and architecture documentation to understand the project's requirements. This is a functional requirement for code generation and does not involve accessing sensitive user credentials, private keys, or performing unauthorized network transmissions.
[DYNAMIC_EXECUTION]: The skill generates script files (GDScript, C#, C++, and Rust) based on predefined templates. It does not compile or execute the generated code at runtime. Additionally, the skill explicitly prevents overwriting existing files, protecting against accidental or malicious data loss.

test-helpers