use-case-triage
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection: The skill accepts user-supplied descriptions of data processing activities to determine legal and privacy obligations. This pattern introduces a surface where crafted inputs could attempt to influence the agent's classification logic.
- Ingestion points: User-provided feature descriptions or data activity summaries entered as the skill's primary argument.
- Boundary markers: The skill does not define specific delimiters (such as XML tags or unique markers) to isolate the user-provided description from the agent's internal reasoning instructions.
- Capability inventory: The skill is designed to read authoritative configuration files and write assessment results to matter-specific directories within the
~/.claude/plugins/config/claude-for-legal/privacy-legal/file path. - Sanitization: There are no explicit instructions for the agent to sanitize or filter the input descriptions for potential instruction-like text before processing them against privacy rules.
- File System Interaction: The skill manages sensitive legal context by reading from and writing to a dedicated local directory structure. While this is necessary for matter management, it relies on the integrity of the underlying configuration files to provide accurate legal triage advice.
Audit Metadata