The Agent Skills Directory

Remote Instruction Fetching: The skill is designed to dynamically fetch and follow markdown instruction files from a remote GitHub repository. This allows the agent to update its logic and strategies based on the latest guidance from the source repository. These downloads are targeted at official repositories hosted by a well-known service.
Indirect Prompt Injection Surface: The skill implements a mechanism to load and follow instructions from a target repository's local .github/aw/instructions.md file. This allows repository-specific overrides of upstream default behaviors. This architecture creates a surface where content within a repository being processed could influence the agent's subsequent actions.
Secure Credential Handling: The instructions include a specific security practice to ensure that authentication headers or tokens are never sent to public raw URLs during file fetching. This reduces the risk of credential exposure when the agent switches between authenticated API calls and public file downloads.
Instructional Execution Chain: The skill functions by chaining together various instructional documents, where the output of one selection determines the next set of loaded instructions. This modular design requires consistent validation of the instructions being loaded at each step of the workflow.

agentic-workflows