The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text provided by the user and resides in an environment with file-modification capabilities. This creates a potential surface for indirect prompt injection where malicious instructions hidden in the text could be processed by the agent.
Ingestion points: The skill ingests text directly from the user chat or via the agent's file-reading tools when requested to 'humanize' specific content.
Boundary markers: The instructions do not define strict delimiters (e.g., XML tags or specific block markers) to separate the data being analyzed from the processing instructions.
Capability inventory: The skill environment, as specified in the YAML configuration, allows for the use of Read, Write, Edit, Grep, and Glob tools.
Sanitization: No explicit instructions are provided to the agent to sanitize the input or to disregard commands embedded within the text during the rewriting process.
[EXTERNAL_DOWNLOADS]: The documentation within README.md and WARP.md provides instructions for the user to download the skill repository from a third-party GitHub repository.
Details: The instructions suggest using git clone https://github.com/blader/humanizer.git to install the skill. This is a standard installation procedure for third-party extensions and is not an automated runtime execution by the agent itself.

writing-humanizer