The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data provided by the user either as pasted text or through specified files. While the instructions provide a detailed rewrite process focused on accuracy and style, there is an inherent attack surface for indirect prompt injection.
Ingestion points: Processes user-pasted text and reads files specified by the user as described in SKILL.md.
Boundary markers: Absent. The instructions do not mandate the use of specific delimiters or 'ignore' commands when reading the input text to prevent the AI from following instructions embedded within the content to be humanized.
Capability inventory: Includes file-writing capabilities as the skill is instructed to write the final humanized text back to a file when in 'File mode'.
Sanitization: Absent. No explicit sanitization or filtering of the input content is performed beyond the stylistic rewriting process.
[EXTERNAL_DOWNLOADS]: The skill contains references to external sources, including Wikipedia and GitHub repositories (github.com/blader/humanizer and github.com/op7418/Humanizer-zh), for attribution and to document the origin of its linguistic patterns. These are static informational links and do not involve automated runtime downloads or execution of remote code.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md define a 'File mode' where the agent is expected to write the final output to a file system. This is a functional requirement for the skill's stated purpose of editing documents and is not associated with unauthorized command execution.

humanizer