The Agent Skills Directory

Local Configuration Management: The skill reads and writes to specific local paths (e.g., ~/.claude/plugins/config/claude-for-legal/) to maintain a persistent 'playbook' of legal positions. This allows the tool to provide consistent triage results based on organization-specific rules and is core to its intended functionality.
Authorization and Privilege Safeguards: The skill explicitly instructs the agent to verify the recipient's 'privilege circle' before generating output, helping to prevent the accidental waiver of legal privilege. It also restricts certain high-confidence outcomes (GREEN status) until attorney-reviewed positions are established in the configuration.
Indirect Input Processing: As a document review tool, the skill processes untrusted input in the form of NDA text. It mitigates potential manipulation by requiring checks against the structured playbook (CLAUDE.md) and prompting for human confirmation before any new positions are recorded or acted upon.
Resource Locality: The skill operates entirely within the local environment, referencing local configuration files rather than external network resources or remote scripts, which minimizes its external attack surface.

nda-review