irac-practice

Pass

Audited by Gen Agent Trust Hub on Jul 25, 2026

Risk Level: SAFE
Full Analysis
  • Local File Access and State Management: The skill reads and writes to specific paths within the user directory (e.g., ~/.claude/plugins/config/claude-for-legal/). This is used to maintain a session tracker and load learning profiles, which allows the agent to identify long-term patterns in student writing. This behavior is standard for agents requiring persistent state between interactions.
  • Input Processing Surface: The skill ingests user-provided text, such as legal essays and hypothetical prompts. This represents a potential surface for indirect prompt injection; however, the skill mitigates this by enforcing a rigid output structure and providing specific instructions that prioritize structured analysis over arbitrary text processing.
  • Operational Safety Guardrails: The instructions include a 'Real-matter check' that triggers when the user provides details suggesting a real legal situation. It effectively redirects the user to professional legal resources, minimizing the risk of the AI being misused for legal advice. Additionally, it explicitly warns users that AI-generated citations may be inaccurate and must be verified.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 25, 2026, 12:36 AM
Security Audit — agent-trust-hub — irac-practice