The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and rewrite user-supplied Chinese text to reduce AI-generated signals, which creates a surface for indirect prompt injection.
Ingestion points: The skill processes arbitrary Chinese text provided by the user, referred to as text.txt or paper.txt in the CLI examples.
Boundary markers: There are no explicit instructions for the agent to use delimiters (like triple quotes or XML tags) to separate untrusted user content from its own instructions.
Capability inventory: The skill possesses the capability to rewrite, summarize, and change the style of the provided text, which could be exploited to bypass intended outputs if the input contains malicious instructions.
Sanitization: No sanitization or filtering logic is specified to identify or neutralize embedded instructions within the input text.
[COMMAND_EXECUTION]: The skill provides examples for a "CLI Flow" that involves executing local Python scripts.
Evidence: The instructions suggest running commands such as python3 scripts/detect_cn.py text.txt -v and python3 scripts/style_cn.py text.txt --style xiaohongshu -o out.txt.
Analysis: While these commands are presented as optional for users with a local clone of the toolkit, they involve executing external code (scripts/*.py) that is not part of the skill itself. Users should audit these scripts before execution.
[METADATA_POISONING]: There are inconsistencies in the attribution of the skill's source and author.
Evidence: The frontmatter lists the source as antigravity, the repository URL points to a user named sickn33, the body text attributes the work to voidborne-d/humanize-chinese, and the platform identifies the author as ranbot-ai.
Analysis: While these discrepancies may reflect forks or project renames, inconsistent metadata can be used to mask the true origin of a skill or mislead users about its maintenance status.

humanize-chinese