The Agent Skills Directory

[External Instruction Fetching]: The skill retrieves a significant number of markdown files from the github/gh-aw repository to define its behavior for specific tasks. These files provide the core logic for designing, debugging, and updating workflows. As these are hosted on a well-known service, they are integrated as part of the skill's extended instruction set.- [Indirect Prompt Injection Surface]: The skill is configured to load and prioritize instructions from .github/aw/instructions.md located within the user's active repository. This repository overlay mechanism is designed to allow local customization but creates a path for untrusted data to influence agent behavior.
Ingestion points: The file .github/aw/instructions.md is read from the local repository environment.
Boundary markers: No specific delimiters or safety warnings are implemented to separate repository instructions from the agent's core logic; the skill explicitly states that these instructions override defaults.
Capability inventory: Once loaded, the agent is directed to "follow it directly," potentially utilizing any tools or capabilities available to the agent session.
Sanitization: The skill does not perform sanitization or validation on the content of the repository-provided instructions.

agentic-workflows