policy-diff
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (regulation text or summaries provided by the user). While this presents a theoretical surface for prompt injection, the skill effectively mitigates risk by:
- Ingestion points: Regulatory text provided via the
argument-hintor pasted by the user. - Boundary markers: The skill mandates strict extraction based on the internal policy index (
CLAUDE.md) as a source of truth. - Capability inventory: Accesses localized configuration files (
~/.claude/plugins/config/claude-for-legal-zh/) and utilizes research MCP/search tools for verification. - Sanitization: Explicitly instructs the agent to warn users that citations are AI-generated and must be verified against primary legal sources.
- [DATA_EXPOSURE_EXFILTRATION]: The skill reads from localized plugin configuration paths (
~/.claude/plugins/config/claude-for-legal-zh/regulatory-legal/CLAUDE.md). These paths are specific to the tool's legitimate function and do not involve sensitive system credentials or user home directories outside the plugin scope.
Audit Metadata