The Agent Skills Directory

Dynamic Execution / Configuration Loading: The skill executes a local Python module (setup_preflight) and consults local and project-wide override files (.apache-magpie-local/issue-deduplicate.md and .apache-magpie-overrides/issue-deduplicate.md). Loading and executing logic from computed local paths is a security consideration as it relies on the integrity of the files within these directories to define agent behavior.
Indirect Prompt Injection Surface: The skill processes untrusted external content from issue bodies, titles, and comments during the triage process. This creates a potential surface where malicious instructions embedded in an issue could attempt to influence the agent's logic (e.g., attempting to bypass confirmation). The skill addresses this through specific instructions to treat external content as data and by requiring explicit maintainer confirmation for all actions.
Ingestion points: Issue data fetched via gh issue view in Step 1.
Boundary markers: The skill includes explicit rules (Golden Rule 5) to treat external content as data and scan for hidden directives or injection attempts.
Capability inventory: The agent has the capability to post comments, add labels, and close issues using the gh command-line tool.
Sanitization: All state-changing actions are presented as a proposal to a human maintainer, requiring manual review and explicit confirmation before execution.

deduplicate