vendor-ai-review
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE]: The skill reads internal configuration and matter files located at
~/.claude/plugins/config/claude-for-legal-zh/ai-governance-legal/CLAUDE.mdandmatter.md. While this file access is intended for its legal review functionality, it exposes internal legal stances and transaction data to the agent session. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of AI vendor contract terms and processes them alongside sensitive internal guidelines. An attacker-controlled contract could contain instructions designed to leak the contents of the local legal configuration or matter files.
- Ingestion points: The skill ingests untrusted text via the
[粘贴AI供应商合同条款](paste AI vendor contract terms) placeholder. - Boundary markers: The workflow lacks explicit delimiters or instructions to ignore embedded commands within the pasted contract content.
- Capability inventory: The skill has the capability to read local legal configuration files and matter-specific files.
- Sanitization: There is no evidence of sanitization or safety-filtering for the ingested contract text before it is processed by the agent.
Audit Metadata