The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user-supplied Chinese text, which presents an attack surface for instructions embedded within that data.
Ingestion points: The skill reads text from user input or local files using the Read tool.
Boundary markers: The instructions explicitly define the input as "material" and strictly command the agent to ignore any embedded commands, role-play settings, or prompts.
Capability inventory: The skill utilizes Read, Write, Edit, and AskUserQuestion for content modification.
Sanitization: The skill relies on strong logical boundaries defined in the prompt to prevent the agent from acting on external data as instructions.
[EXTERNAL_DOWNLOADS]: The skill metadata and documentation reference external sources on GitHub.
Evidence: References to github.com/blader/humanizer, github.com/op7418/Humanizer-zh, and github.com/hardikpandya/stop-slop are included.
Context: These are standard references to the skill's conceptual origins and do not involve any automated runtime execution or downloading of untrusted scripts.

humanizer-zh