The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to read and process untrusted source code from various game projects. This risk is inherent to its function as a code reviewer.
Ingestion points: Processes engine-specific source files (*.cs, *.h, *.cpp, *.gd, etc.) and project metadata files (package.json).
Boundary markers: No explicit delimiters are used to separate reviewed code from the agent's internal instructions.
Capability inventory: Includes file system read operations and local file system write (telemetry logging).
Sanitization: None observed; the skill interprets code content directly to perform its architectural analysis.
[SAFE]: The skill records execution metadata to a local skill-telemetry.md file within the ~/.claude/projects/ directory. This is a standard local persistence mechanism for maintaining execution history and does not involve network exfiltration or sensitive data exposure.

game-code-review