The Agent Skills Directory

Structured Local File Access: The skill interacts with local files within a specific configuration directory (~/.claude/plugins/config/claude-for-legal/). This is used to manage policy indices and gap trackers, which is central to its purpose as a legal assistant tool.
Write Protection Guardrails: A primary safety feature of this skill is its strict prohibition against overwriting source documents. It is instructed to write outputs to new, uniquely named files, which protects the integrity of existing policy libraries.
Rule Verification Workflow: The skill performs external research (via web search or research tools) to confirm the current status of regulations. This ensures that redrafts are based on up-to-date legal requirements and alerts the user if a rule's status is unverified.
Contextual Ingestion: The skill processes external text provided by the user (policy text and rule text). While this represents a standard processing surface, the skill implements 'no silent supplement' rules and mandatory provenance tagging to maintain clarity on the source of all information.
Human-Centric Design: The skill includes explicit instructions to avoid 'applying' changes or closing tracking items automatically, maintaining a clear boundary between AI-generated proposals and human policy ownership.

policy-redraft