humanizer-en

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill's design involves processing arbitrary user-supplied text or reference files, which introduces a potential surface for indirect prompt injection.\n
  • Ingestion points: Processes text input from the user or reads files from provided paths to perform style analysis (SKILL.md).\n
  • Boundary markers: The instructions do not define explicit delimiters to separate untrusted user content from the agent's system instructions.\n
  • Capability inventory: The skill uses Read, Write, and Edit tools to interact with the file system (SKILL.md).\n
  • Sanitization: Input content is not sanitized or validated before processing.\n
  • Mitigation: The skill includes a 'Safety Tier' (Tier M) that instructs the agent to show results and wait for explicit user approval before overwriting any files.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 08:28 AM
Security Audit — agent-trust-hub — humanizer-en