The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches workflow instructions and strategies from the github/gh-aw repository on GitHub. This is used as a fallback when local files are unavailable or authentication fails. Neutral language is used to guide the agent in fetching these resources without transmitting authentication tokens to public endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill implements a mechanism to load and follow instructions from a target repository's .github/aw/instructions.md file. These overlay instructions are designed to override default behaviors, creating a surface where malicious instructions in a repository could influence the agent's behavior.
Ingestion points: The agent is instructed to load .github/aw/instructions.md from the repository it is currently processing.
Boundary markers: No delimiters or explicit ignore instructions are provided to separate the overlay instructions from other data or to prevent the agent from obeying embedded commands.
Capability inventory: The skill allows for complex tasks including workflow design, creation, and debugging, which may involve the use of CLI tools and file system modifications as described in the associated documentation and test fixtures.
Sanitization: There is no evidence of sanitization or validation for the content loaded from the overlay files before it is adopted as authoritative instructions by the agent.

agentic-workflows