The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes content from various external markdown files and passes the data to subagents, creating a surface for indirect prompt injection. * Ingestion points: Data is ingested from the design/gdd/, design/levels/, design/narrative/, and production/ directories. * Boundary markers: The skill does not provide instructions or delimiters to the subagents to ignore or isolate potentially malicious instructions within the ingested documents. * Capability inventory: The orchestrator and its delegated subagents have the ability to execute tasks, perform file writes, and run bash commands. * Sanitization: There is no indication that the ingested data is sanitized, validated, or escaped before being provided to the agents as context.

team-level