The Agent Skills Directory

Input Validation: The skill validates GitHub handles using a specific regular expression pattern before they are used in commands. This practice mitigates potential command injection risks by ensuring only well-formatted identifiers are processed.
Secure Command Execution: When interacting with the GitHub API, the skill writes query strings to temporary files and uses the @file syntax with the gh CLI. This avoids interpolating untrusted data directly into shell commands, a common security best practice.
Indirect Prompt Injection Awareness: The skill identifies that external content (PR titles, comments, and issue descriptions) is untrusted and could contain malicious instructions. It explicitly instructs the agent to treat this data as information rather than instructions and to flag any suspected injection attempts.
Dynamic Execution of Local Tools: The skill utilizes a pre-flight checker (setup_preflight) located in a local directory (.apache-magpie-local). This is used for project configuration verification and state consistency checks at the start of execution.
Privilege and Scope: The skill is designed to be read-only and does not perform any mutations on GitHub. It generates drafts for maintainer review, ensuring human-in-the-loop control for all final actions.

contributor-to-committer