The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes information from Game Design Documents (GDD) and existing Architecture Decision Record (ADR) files to inform its drafting process. This ingestion of potentially untrusted project data creates a surface for indirect prompt injection. \n
Ingestion points: Content is read from design/gdd/ and docs/architecture/ during the context-gathering phase. \n
Boundary markers: The skill does not employ specific delimiters or instructions to ignore embedded commands within the processed files. \n
Capability inventory: The skill can perform file system modifications (Write, Edit) and delegate tasks to specialized subagents. \n
Sanitization: The primary mitigation is the use of AskUserQuestion at critical junctures, requiring the user to verify all assumptions, drafted content, and registry updates before they are committed. \n
[SAFE]: The skill manages environment-specific knowledge by reading local engine reference documentation. This approach ensures technical accuracy for specific engine versions without necessitating external network access or remote data fetching. \n
[SAFE]: Task delegation to 'Engine Specialists' and the 'Technical Director' is handled via the internal Task tool using predefined agent roles, maintaining execution within the local agent environment.

architecture-decision