The Agent Skills Directory

Local Module Execution (Pre-flight Check): The skill executes a Python module (setup_preflight) from a local directory (.apache-magpie-local). This represents a dynamic execution pattern intended to verify the project's environment and configuration. While executing local code is a consideration, it is implemented as a pre-flight validation step within the skill's managed lifecycle.
Indirect Prompt Injection Surface: The skill ingests untrusted data from external sources, which presents a potential injection surface. This is handled with the following safeguards:
Ingestion points: The skill reads GitHub issue and PR titles, bodies, and comments via gh issue view (Step 2 of the Runtime loop).
Boundary markers: The skill requires explicit human maintainer confirmation before posting any comments (Step 8 and 9) and includes instructions to ignore instructions found in external content.
Capability inventory: The skill can execute local Python modules and the GitHub CLI (gh).
Sanitization: The skill instructions explicitly direct the agent to treat external content as data rather than instructions and to flag potential injection attempts to the user.
Command Line Tool Usage: The skill uses the GitHub CLI (gh) to interact with remote repositories. This usage is confined to viewing thread information and posting drafted comments after maintainer approval, which is consistent with the skill's intended purpose of PR management.

mentor