draft
Warn
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill accesses sensitive information including client case notes and intake summaries, alongside internal plugin configuration files. It further uses dynamic path construction for loading practice-area guides based on user-supplied parameters, creating an access surface for arbitrary local files. \n
- Evidence: The skill instructs the agent to read '
/.claude/plugins/config/claude-for-legal/legal-clinic/CLAUDE.md' and dynamically load guides from '/.claude/plugins/config/claude-for-legal/legal-clinic/guides/.md'.\n- [PROMPT_INJECTION]: The skill ingests untrusted case notes and intake summaries to generate legal drafts, presenting an indirect prompt injection surface.\n - Ingestion points: Intake summaries and case notes provided by the student or user.\n
- Boundary markers: The skill instructions include the use of inline flags like '[VERIFY]', '[FACT NEEDED]', and '[UNCERTAIN]' to demarcate AI-generated or hypothesized content.\n
- Capability inventory: The skill is capable of reading local configuration and guidance files and generating drafted legal text.\n
- Sanitization: There is no evidence of validation or sanitization of the ingested case notes before they are processed by the agent.
Audit Metadata