The Agent Skills Directory

Structured File Access: The skill reads and writes data within the ~/.claude/plugins/config/claude-for-legal/regulatory-legal/ directory. This is used to manage the policy library index and matter-specific workspaces, which is consistent with the skill's documented purpose for regulatory compliance tracking.
External Data Handling: It processes regulatory text provided by users or external sources. To mitigate the risk of processing inaccurate or malicious content, the skill requires explicit source attribution (e.g., [web search — verify], [user provided]) and prohibits 'silent supplements' or filling gaps using model knowledge without user approval.
Integrity Banners: The skill automatically generates warning banners if a rule's status cannot be verified or if the user requests a restricted scope. These banners are designed to be carried through to downstream artifacts to prevent the misrepresentation of compliance data.
Matter Separation: It includes logic to enforce matter-level data isolation, specifically instructing the agent not to read files from other matter workspaces unless cross-matter context is explicitly enabled.

policy-diff