humanize
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains an 'OVERRIDE RULE' that explicitly instructs the agent to ignore 'OMC autonomous modes' and specific patterns like 'The boulder never stops' or 'autopilot'. This is a form of instruction bypass intended to ensure the skill's logic takes precedence over other system or skill constraints.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted manuscript files which could contain hidden malicious instructions intended to influence the behavior of the sub-agents (G5, G6, F5).
- Ingestion points: Manuscripts are read from the user-provided
{target_file}and passed as input during multiple pipeline stages. - Boundary markers: The skill does not define specific delimiters or instructions to treat the manuscript text as passive data when passing it to sub-agents.
- Capability inventory: The skill can spawn multiple sub-agents (diverga:g5/g6/f5), call MCP tools for text analysis, and perform file-write operations.
- Sanitization: There is no evidence of sanitization, filtering, or escaping of the manuscript content before it enters the agent's context.
Audit Metadata