The Agent Skills Directory

Indirect Prompt Injection Surface: The skill ingests documents from external eDiscovery platforms to build the deposition outline. As this data comes from external sources, it represents a potential surface for indirect prompt injection. However, the skill mitigates this by instructing the agent to strictly use verbatim quotes, pinpoint citations, and specific verification markers, which helps ensure the output remains grounded in the source material.
Local File Access: The skill accesses configuration and matter log files located in the ~/.claude/plugins/config/claude-for-legal/ directory. This access is necessary for maintaining consistency with the case theory and enforcing the 'Conflicts gate,' which prevents the agent from processing matters that have not undergone a formal intake process.
Privilege Protection Guardrails: The 'Destination check' feature is a proactive security measure that prompts the user to verify if the output destination (e.g., a Slack channel or email list) is within the 'privilege circle.' This helps prevent the accidental disclosure of sensitive legal work product to counterparties or unauthorized third parties.
Data Integrity and Verification: The skill includes robust 'Marker discipline' requiring markers like [VERIFY] and [UNCERTAIN] for factual or legal propositions. It also requires explicit source attribution for all citations (e.g., [Westlaw], [CourtListener]), which serves as a safeguard against AI hallucinations and ensures the reliability of the legal advice generated.

deposition-prep