ip-clause-review

Pass

Audited by Gen Agent Trust Hub on Jul 3, 2026

Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted agreement content from files, links, or direct user input without implementing boundary markers.
  • Ingestion points: Agreement content provided via file paths, Google Drive links, or pasted text as described in SKILL.md.
  • Boundary markers: Absent. The instructions do not define delimiters for the untrusted content or instruct the agent to ignore commands embedded within the agreement text.
  • Capability inventory: The skill possesses read/write capabilities for the local file system, specifically within the ~/.claude/ configuration hierarchy.
  • Sanitization: Absent. No validation or filtering is specified for the input agreement content.
  • [DATA_EXFILTRATION]: The skill performs file operations within the ~/.claude directory, which is a hidden configuration location for the agent platform. It reads a practice profile from ~/.claude/plugins/config/claude-for-legal/ip-legal/CLAUDE.md and writes matter-specific data to subfolders in that same tree.
  • [DATA_EXFILTRATION]: The skill constructs file paths using a user-supplied (e.g., writing to ~/.claude/plugins/config/claude-for-legal/ip-legal/matters//). Without explicit sanitization of this input, there is a risk of path traversal if a user provides a malicious slug intended to overwrite sensitive files elsewhere in the file system.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 3, 2026, 03:59 PM
Security Audit — agent-trust-hub — ip-clause-review