The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process information from external sources, including web searches and legal research tools (such as Westlaw or CourtListener). This data is interpolated into the agent's context to generate legal answers. The ingestion of untrusted external content represents a surface for indirect prompt injection, where malicious instructions could be embedded in the retrieved content. The skill incorporates mitigation strategies by requiring explicit source attribution (e.g., tagging sources as [web search — verify]) and requiring the agent to stop for human instruction if coverage is thin.
Scoped File Interaction: The skill performs read and write operations on the local file system within the ~/.claude/plugins/config/ directory to manage jurisdictional footprints and matter-specific context. These operations are intended for maintaining workflow state and matter workspaces. While the access is limited to the skill's specific configuration paths, the interaction with local files is a consideration for the overall security posture.

wage-hour-qa