The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing content from multiple user-modifiable markdown files.
Ingestion points: Reads extensively from design/gdd/*.md, design/art/art-bible.md, design/narrative/*.md, and design/levels/*.md (SKILL.md, Phases 0b, 1, and 2).
Boundary markers: No delimiters or instructions to ignore embedded commands are present when interpolating file content into prompts for the main session or spawned agents.
Capability inventory: The skill utilizes Write, Edit, and Task tools. Injected instructions could potentially manipulate file outputs or influence the behavior of the spawned art-director and technical-artist agents.
Sanitization: No evidence of content sanitization or validation was found prior to processing external file data.

asset-spec