The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external Game Design Documents (GDDs), level designs, and character profiles which could contain malicious instructions designed to influence the main agent or the spawned sub-agents.
Ingestion points: Content is read from various files including design/gdd/[target-name].md, design/levels/[target-name].md, and design/narrative/characters/[target-name].md using the Read tool.
Boundary markers: The skill does not employ explicit delimiters or system instructions to tell the sub-agents (art-director and technical-artist) to ignore instructions embedded within the source material.
Capability inventory: The skill possesses the Write and Edit tools, allowing it to modify the local filesystem, and the Task tool to spawn sub-agents. This creates a surface where injected instructions could lead to unauthorized file changes or deceptive visual prompts.
Sanitization: There is no evidence of sanitization, filtering, or validation of the text extracted from design documents before it is passed into the prompts for sub-agents.

asset-spec