The Agent Skills Directory

Instruction Sourcing from External Repository: The skill directs the agent to load instructional content from an external source ('github/gh-aw') to handle specific tasks. While this supports centralized prompt management, it establishes a dependency on the integrity of the remote source.
Repository Overlay Surface: The skill allows for repository-local files, such as '.github/aw/instructions.md', to override default behavior. This pattern presents an indirect prompt injection surface where a file within a processed repository could attempt to influence the agent's logic.
Ingestion points: Data is ingested from multiple markdown files located in '.github/aw/' and specified external repositories.
Boundary markers: The skill does not implement explicit boundary markers or 'ignore' instructions when loading these external prompts.
Capability inventory: The skill acts as a router to other prompt files which may have capabilities including file system access and tool execution.
Sanitization: Instructions from external or local files are loaded and followed directly without a validation or sanitization step.

agentic-workflows