The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data and has capabilities to write to the filesystem, creating a potential attack surface.
Ingestion points: The skill reads design/gdd/game-concept.md, production/review-mode.txt, and .claude/docs/technical-preferences.md in Phase 0 to initialize the workflow. It also reads existing content from design/art/art-bible.md if present.
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands when passing data from game design documents to sub-agents (e.g., the art-director agent).
Capability inventory: The skill is configured with Read, Write, Edit, and Task (for agent delegation) tools. These are used to read project context and author the final Art Bible document across multiple phases.
Sanitization: There is no evidence of sanitization or validation of the text extracted from game design documents before it is used as input for sub-agent tasks.

art-bible