The Agent Skills Directory

Indirect Prompt Injection Surface: The skill facilitates the loading of an optional overlay file (.github/aw/instructions.md) from the local repository. These instructions are given precedence over default prompts, which could allow untrusted content within a repository to influence the agent's behavior if the repository is not thoroughly vetted.
Ingestion points: The agent is explicitly instructed to check for and load .github/aw/instructions.md from the current working directory.
Capability inventory: The skill enables complex tasks such as workflow design, creation, debugging, and optimization, which typically involve repository file modifications and potential execution of CLI tools.
Boundary markers: The instructions do not specify any delimiters or warnings to the agent to treat the local overlay content as potentially untrusted data.
Sanitization: There are no provisions for sanitizing or validating the content of the overlay file before it is incorporated into the agent's instruction set.
Remote Resource Retrieval: The skill fetches various Markdown-based prompt and configuration files from the github/gh-aw repository. These resources are essential for the skill's dispatcher functionality and are retrieved from a well-known service provider's infrastructure.

agentic-workflows