The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes natural-language files which may contain untrusted data that could influence the behavior of the LLM during the humanization process.
Ingestion points: Reads content from user-specified files via the <filepath> argument in SKILL.md.
Boundary markers: Explicitly treats fenced code blocks and technical segments as read-only; however, it lacks specific delimiters or instructions to the LLM to ignore malicious instructions embedded within the prose sections it is tasked with rewriting.
Capability inventory: File system read/write (overwrites original files with humanized versions), subprocess execution of local Python scripts, and network calls to Anthropic's API or local CLI fallbacks.
Sanitization: Includes a post-processing validation step to ensure technical blocks remain identical and to check for residual AI-isms, which provides a layer of verification against accidental or malicious modifications.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python module using python3 -m scripts <absolute_filepath>. While this is standard for skill functionality, it involves local code execution and shell interaction that relies on the agent's ability to safely handle absolute file paths provided as arguments.

unslop-file