The Agent Skills Directory

Legal Workflow Safeguards: The skill includes robust procedural checks, such as a 'Privilege gate' that requires users to define a posture (cleared, mixed, or aborted) before processing documents. This is a positive security and ethical feature designed to prevent accidental data exposure or waiver of privilege.\n- Project-Specific Resource Management: The skill interacts with files in the ~/.claude/plugins/config/claude-for-legal/ directory. These paths are consistent with the skill's role as a specialized tool for legal matter management and do not indicate unauthorized file system access.\n- Integration with Professional Services: The skill identifies and utilizes eDiscovery connectors (e.g., Everlaw, Relativity, DISCO) to pull document metadata. These represent integrations with well-known, industry-standard legal technology services and are used according to the skill's primary purpose.\n- Data Processing Ingestion Surface: As its core function, the skill processes untrusted external data (PDFs, emails, .docx files). While this creates a surface for indirect prompt injection, the skill mitigates this through transparency and human-in-the-loop verification.\n
Ingestion points: Processes documents from user-provided paths, shared drives, and eDiscovery connectors (SKILL.md).\n
Boundary markers: The skill does not explicitly define character-based delimiters for the extracted text but uses structured output formats to isolate facts.\n
Capability inventory: Reading and writing local matter files and accessing external metadata via authenticated MCP connectors.\n
Sanitization: Employs [SME VERIFY] tags for all extracted significance calls and privilege flags, ensuring that an expert reviews the model's output before it is used in legal work product.

chronology