vendor-agreement-review
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill is designed to process and analyze untrusted text from third-party vendor agreements (MSAs, SaaS agreements). Ingesting external documents creates a potential surface where embedded instructions within the contract could theoretically influence agent behavior. The skill mitigates this through a structured comparison workflow against a local 'playbook' and by requiring human review for all actions.
- Ingestion points: Processes vendor agreements and counterparty metadata in Step 1 and Step 3.
- Boundary markers: Uses structured comparison tables and memo templates, though explicit data delimiters for the contract text are not detailed.
- Capability inventory: Writing to matter-specific folders (
~/.claude/plugins/config/claude-for-legal/commercial-legal/matters/) and invoking MCP tools for CLM and DocuSign. - Sanitization: No explicit sanitization or filtering of the contract text is described before interpolation.
- External Content Retrieval: The skill provides instructions for the agent to fetch and read Data Processing Agreements (DPAs) from external URLs found within the vendor agreements. While retrieval of these documents is necessary for a complete review, fetching content from arbitrary URLs provided by counterparties is an external network operation that warrants careful handling. The skill recommends specialized privacy tools for this purpose as a best practice. (Evidence:
SKILL.md - 'DPA-by-reference handling').
- Matter Workspace Isolation: The skill manages data by reading from and writing to specific local directories under
~/.claude/plugins/config/claude-for-legal/tailored to individual legal matters. This practice supports workspace isolation and organizational consistency, involving expected file system access for its stated purpose. (Evidence:SKILL.md - 'Matter context').
- Professional Safety Guardrails: The skill implements robust security and ethical checks, such as a 'Destination check' to protect attorney-client privilege and mandatory role-based disclosures. It requires explicit confirmation from non-lawyer users before performing actions with legal consequences, such as generating signature envelopes. (Evidence:
SKILL.md - 'Destination check' and 'Escalation routing').
Audit Metadata