The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute unity-cli during the audit phase to verify engine versions and pipeline configurations. This is a standard project management operation and does not involve passing unvalidated user input to the shell.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined attack surface where it ingests untrusted data from the repository, specifically repository instructions (AGENTS.md, CLAUDE.md) and existing project code. While necessary for context-aware implementation, these ingestion points could potentially contain malicious instructions if the repository itself is compromised.
Ingestion points: AGENTS.md, CLAUDE.md, and path-scoped rules identified in Phase 1 of the workflow.
Boundary markers: The skill does not explicitly define delimiters for these external files, though it instructs the agent to follow established project motion and accessibility settings.
Capability inventory: The agent has capabilities to write scripts, modify existing codebases, and execute project-specific CLI tools like unity-cli.
Sanitization: No specific sanitization or escaping of the repository instruction content is mentioned before it enters the agent's context.
[SAFE]: The skill implements several security and performance best practices, including object pooling for damage numbers and audio voices to prevent runtime stutters, and a "Liquid UI" architecture that ensures gameplay logic remains decoupled from presentation (juice) effects.
[SAFE]: Robust accessibility guardrails are included, requiring the agent to respect project-wide reduced-motion preferences by skipping disruptive effects like camera shake, screen flashes, and time-scale modifications while maintaining static success/failure cues.

game-feel-polish