The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes untrusted data from arguments or external files to drive its execution pipeline.
Ingestion points: Data enters the context via $ARGUMENTS or by reading the contents of local files specified by the user in Step 2 of the behavior section.
Boundary markers: The skill does not implement delimiters (e.g., XML tags or triple quotes) or 'ignore embedded instructions' warnings for the input data, allowing malicious instructions within the text to potentially influence the 'humanization' phases.
Capability inventory: The skill utilizes the Read tool for file system access and executes a complex multi-phase instruction set (Phase 0 through results delivery) defined in the humanize-korean procedure.
Sanitization: No input validation or instruction filtering is applied to the processed content.
[DATA_EXFILTRATION]: The skill allows the agent to read local file contents based on paths provided directly in $ARGUMENTS. While the instructions specify .txt and .md extensions, this capability could be used to expose sensitive project documentation or configuration files to the model context if they match those extensions.

humanize