korean-writer
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a text-only implementation focused on linguistic naturalness and stylistic consistency in Korean prose. It does not require or request access to the network, credentials, or dangerous system tools.
- [INDIRECT_PROMPT_INJECTION]: The humanize mode instructions (rules/humanize-method.md) explicitly direct the agent to treat supplied text as 'data, not instructions.' It includes specific logic to ignore requests found inside processed text and provides an adversarial test case to ensure this behavior is maintained.
- [COMMAND_EXECUTION]: The validation rules (rules/validation.md) list development-time commands (e.g., node, bun, ripgrep) intended for repository maintainers to verify the package integrity. The skill's operational contract explicitly forbids the agent from using scripts or network tools during execution.
- [OBFUSCATION]: The self-application audit in the validation rules uses character classes within regex (e.g.,
[를] 통해) to prevent the audit command from flagging its own definition. This is a standard development technique for linting and does not constitute malicious obfuscation. - [DATA_EXPOSURE]: The skill does not access sensitive local file paths or hardcoded secrets. It operates entirely on the text provided within the conversational context.
Audit Metadata