humanize-korean
Warn
Audited by Gen Agent Trust Hub on Jul 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill attempts to execute several local scripts and shell commands. Specifically, the SKILL.md file references 'scripts/prepare_monolith_input.py' which is not included in the provided file bundle. Additionally, multiple utility agents are instructed to use bash for executing tests, verifying file differences, and managing the project workspace. The metrics scripts also perform dynamic path manipulation to handle local imports.
- [COMMAND_EXECUTION]: The skill contains utility agents, such as the 'post-editese-metric-engineer', that are designed to generate new Python code (e.g., metrics_v2.py) and immediately execute it. This dynamic code generation and execution pattern, while intended for maintenance, allows for runtime modification of the skill's logic.
- [PROMPT_INJECTION]: The skill processes untrusted user input to humanize text, which presents an attack surface for indirect prompt injection. Malicious instructions embedded in the input text could attempt to override the agent's constraints or the skill's specific objective of style-based rewriting.
Audit Metadata