policy-diff

Pass

Audited by Gen Agent Trust Hub on May 15, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions define a legitimate and structured workflow for legal policy analysis. It performs authorized file reads on targeted internal configuration files (~/.claude/plugins/config/claude-for-legal/regulatory-legal/CLAUDE.md) and workspace transaction files (matter.md) necessary for its primary function.
  • [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection as it processes untrusted regulatory text provided by users or retrieved from the web.
  • Ingestion points: User-pasted regulation text and external web search results (via tools).
  • Boundary markers: None explicitly defined to delimit the ingested text from instructions.
  • Capability inventory: Local file read access for policy documents.
  • Sanitization: The skill mitigates risks by requiring the agent to verify the legal status of regulations (Step 0) and implementing a 'no silent fill' policy that prevents the agent from guessing or hallucinating missing regulatory details.
Audit Metadata
Risk Level
SAFE
Analyzed
May 15, 2026, 07:06 AM
Security Audit — agent-trust-hub — policy-diff