The Agent Skills Directory

Indirect Prompt Injection Surface: The skill aggregates and processes external content from GitHub, including pull request titles, bodies, and issue comments. While such content can contain adversarial instructions, the skill includes explicit directives to treat all external text as data rather than instructions, and it employs boundary markers to maintain this distinction.- Local Command and Dynamic Execution: The skill instructs the agent to execute shell commands using the GitHub CLI (gh) and a local Python module (setup_preflight) for pre-flight environment checks. The Python execution involves modifying the PYTHONPATH to include a local directory (.apache-magpie-local), which is a mechanism for project-specific validation.- Input Sanitization and Injection Guards: The skill demonstrates proactive security measures by validating GitHub handles against a strict regular expression before use in shell commands or API paths. It also uses a temporary file pattern for GraphQL queries to prevent shell metacharacter injection when interacting with the GitHub CLI.- Data Exposure and Network Operations: The skill performs network operations to fetch contributor activity metrics. These operations are scoped to authoritative GitHub APIs and Apache project metadata services, which are established and trusted infrastructure for this skill's operational context.

nomination