written-consent
Pass
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: The skill implements strong safety logic, including a 'Hard Stop' mechanism that prevents the generation of 'ready-to-sign' documents for high-stakes corporate actions under time pressure without external legal counsel.
- [SAFE]: The skill strictly monitors user roles, providing detailed 'Instructions for Counsel' and warnings to non-legal users before outputting finalized drafts.
- [SAFE]: File access is restricted to the plugin's own configuration and matter directories (~/.claude/plugins/config/claude-for-legal/), avoiding sensitive system-level paths or credentials.
- [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it ingests untrusted user descriptions and historical resolution precedents to generate new drafts.
- Ingestion points: User-provided action descriptions and precedent files retrieved from the resolution repository defined in CLAUDE.md.
- Boundary markers: Absent; the skill interpolates retrieved data directly into the resolution structure.
- Capability inventory: Performs file reading and writing within the local plugin directory; no network or shell execution capabilities identified.
- Sanitization: Not explicitly implemented for the retrieved resolution text or user input.
Audit Metadata