skills/lev-os/agents/humanizer/Gen Agent Trust Hub

humanizer

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and rewrite untrusted text provided by the user or from local files. While the instructions emphasize maintaining factual accuracy and avoiding inventions, the interpolation of external data into the agent's context for rewriting creates a surface for indirect prompt injection.
  • Ingestion points: Untrusted data enters the context via user pastes or by reading file paths provided to the /humanizer command (SKILL.md).
  • Boundary markers: The skill lacks explicit XML-style or unique delimiters to wrap the input text, though it instructs the agent to 'read the sample first' and 'find AI patterns'.
  • Capability inventory: The skill has read access to local files and the ability to output rewritten text to the user or write it back to a file in 'File mode'.
  • Sanitization: There is no evidence of programmatic sanitization or escaping of the input text before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 07:21 AM
Security Audit — agent-trust-hub — humanizer