chronology

Pass

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads from and writes to sensitive local configuration and matter directories located at ~/.claude/plugins/config/claude-for-legal/litigation-legal/. This is the primary mechanism for managing case theory, document logs, and the resulting work product.
  • [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The workflow incorporates data retrieval from well-known external legal technology services and document storage providers (including Everlaw, Relativity, DISCO, Aurora, Google Drive, and SharePoint) through MCP connectors when available.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents from various sources to extract and analyze facts, creating a surface for indirect injection via malicious content in legal productions.
  • Ingestion points: Reads from user-provided file paths, email exports, and external platforms as defined in 'Step 1: Identify document sources'.
  • Boundary markers: Employs a 'No silent supplement' rule and mandatory provenance tags (e.g., [web search — verify], [user provided]) to distinguish between document-sourced facts and model-generated assertions.
  • Capability inventory: Performs local file reads to retrieve case context (Step 2) and writes chronology files to designated local paths (Step 8).
  • Sanitization: Implements an unbypassable 'Privilege gate' (Step 0) to define data handling posture and uses explicit [SME VERIFY] flags for borderline significance or privilege classifications.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 3, 2026, 03:58 PM
Security Audit — agent-trust-hub — chronology