The Agent Skills Directory

[SAFE]: Resource Scoping: The skill limits its file system operations to a specific project workspace (~/.claude/plugins/config/claude-for-legal/). It includes explicit checks to prevent accidental cross-context data leakage between different legal matters.
[SAFE]: Integrity Guardrails: The skill mandates the use of source-tracking tags for all legal citations and prohibits the agent from using untrusted web results or model knowledge to fill missing regulatory data without user consent. This prevents the silent incorporation of potentially poisoned or inaccurate data.
[SAFE]: Functional Ingestion: While the skill ingests external content such as Product Requirement Documents (PRDs), this is conducted as part of a structured interview process. The instructions emphasize verifying information against authoritative legal sources, reducing the risk of indirect instruction obedience.

pia-generation