The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary game project files (C#, C++, GDScript, etc.) which can contain malicious instructions intended to influence the reviewer's output.
Ingestion points: Phase 1 through Phase 7 involve scanning and reading various source files (*.cs, *.h, *.cpp, *.gd, etc.).
Boundary markers: The skill lacks explicit instructions to treat code comments or string literals as data rather than instructions.
Capability inventory: The skill has file-read capabilities and writes detailed architecture reports and telemetry files.
Sanitization: There is no evidence of sanitization or filtering of the content ingested from the project files.
[PERSISTENCE]: The skill contains logic to record execution metadata in a hidden directory within the user's home folder.
Evidence: The "Self-evolution telemetry" section appends data to ~/.claude/projects/skill-telemetry.md if the directory exists.
Impact: This provides a mechanism for the skill to maintain logs and state across multiple independent sessions by writing to a platform-specific configuration path.

game-code-review