The Agent Skills Directory

Local Data Processing: The skill reads from and writes to specific local paths within ~/.claude/plugins/config/claude-for-legal/. This scoping ensures that the agent's operations are confined to its intended legal project environment, minimizing the risk of unauthorized file system access.
Human-in-the-loop Review: A security consideration for legal tools is the potential for generating incorrect or privileged content. This skill addresses this by only creating drafts (in markdown or Gmail) and including a 'Send Gate' notice that mandates attorney review. This prevents the agent from autonomously communicating with outside parties.
Indirect Data Ingestion: The skill processes information from matter.md and history.md to populate email drafts. While these files are local, they represent an entry point for data that influences agent output. The skill mitigates this by using a structured template and explicit instructions to match the 'house style' defined in CLAUDE.md, which helps maintain consistency and reduces the likelihood of following unintended instructions embedded in matter histories.
Secure Integration: The use of a Gmail MCP for creating drafts is a standard method for interacting with external services. It relies on the user's existing authentication and restricts the agent's capability to drafting only, rather than sending emails automatically.

oc-status