demand-draft
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Data Processing and Indirect Injection Surface: The skill processes content from local files (
intake.mdandCLAUDE.md) to generate legal drafts. This ingestion of data represents a potential surface where external instructions could be introduced into the agent's context. However, the skill includes significant mitigating factors such as mandatory manual gates, explicit user review of all drafts, and factual verification requirements before final output. - Controlled File System Access: The skill is configured to interact with specific directories within the user's home folder (
~/.claude/plugins/config/claude-for-legal/). This localized file access is necessary for the skill's purpose of managing legal matter records and templates, and it does not appear to access sensitive system files or credentials. - Skill Orchestration: The workflow involves handing off tasks to other components, such as a
docxgeneration skill and amatter-intakeskill. This is a routine method for modularizing complex legal workflows and does not involve unusual or high-privilege operations.
Audit Metadata